Robotics and Mechatronics: Richard (Chunhui) Yang Yukio Takeda Chunwei Zhang Gu Fang Editors

Mechanisms and Machine Science 72
Richard (Chunhui) Yang

Yukio Takeda
Chunwei Zhang
Gu Fang Editors
Robotics and
Mechatronics
Proceedings of the Fifth IFToMM
International Symposium on Robotics &
Mechatronics (ISRM 2017)
Mechanisms and Machine Science
Volume 72
Series Editor
Marco Ceccarelli, Department of Industrial Engineering, University of Rome Tor
Vergata, Roma, Italy
Editorial Board
Alfonso Hernandez, Mechanical Engineering, University of the Basque Country,
Bilbao, Vizcaya, Spain
Tian Huang, Department of Mechatronical Engineering, Tianjin University,
Tianjin, China
Yukio Takeda, Mechanical Engineering, Tokyo Institute of Technology, Tokyo,
Japan
Burkhard Corves, Institute of Mechanism Theory, Machine Dynamics and
Robotics, RWTH Aachen University, Aachen, Nordrhein-Westfalen, Germany
Sunil Agrawal, Department of Mechanical Engineering, Columbia University,
New York, NY, USA
This book series establishes a well-defined forum for monographs, edited Books,
and proceedings on mechanical engineering with particular emphasis on MMS
(Mechanism and Machine Science). The final goal is the publication of research that
shows the development of mechanical engineering and particularly MMS in all
technical aspects, even in very recent assessments. Published works share an
approach by which technical details and formulation are discussed, and discuss
modern formalisms with the aim to circulate research and technical achievements
for use in professional, research, academic, and teaching activities.
This technical approach is an essential characteristic of the series. By discussing
technical details and formulations in terms of modern formalisms, the possibility is
created not only to show technical developments but also to explain achievements
for technical teaching and research activity today and for the future.
The book series is intended to collect technical views on developments of the
broad field of MMS in a unique frame that can be seen in its totality as an
Encyclopaedia of MMS but with the additional purpose of archiving and teaching
MMS achievements. Therefore, the book series will be of use not only for
researchers and teachers in Mechanical Engineering but also for professionals and
students for their formation and future work.
The series is promoted under the auspices of International Federation for the
Promotion of Mechanism and Machine Science (IFToMM).
Prospective authors and editors can contact Mr. Pierpaolo Riva (publishing
editor, Springer) at: [email protected].
Indexed by SCOPUS and Google Scholar.
More information about this series at http://www.springer.com/series/8779

Richard (Chunhui) Yang
•
Yukio Takeda Chunwei Zhang

• •
Gu Fang
Editors
Robotics and Mechatronics

Proceedings of the Fifth IFToMM
International Symposium on Robotics &
Mechatronics (ISRM 2017)
123
Editors
Richard (Chunhui) Yang Yukio Takeda
Western Sydney University Department of Mechanical Engineering
Penrith, NSW, Australia Tokyo Institute of Technology
Tokyo, Japan
Chunwei Zhang
Qingdao University of Technology Gu Fang
Qingdao, China Western Sydney University
Penrith, NSW, Australia
ISSN 2211-0984 ISSN 2211-0992 (electronic)

Mechanisms and Machine Science
ISBN 978-3-030-17676-1 ISBN 978-3-030-17677-8 (eBook)
https://doi.org/10.1007/978-3-030-17677-8
© Springer Nature Switzerland AG 2019
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
1 Architecture Choice of a Robotic Hand for Deep-Sea Exploration

Based on the Expert Gestures Movements Analysis . . . . . . . . . . . . 1
C. Mizera, M. A. Laribi, D. Degez, J. P. Gazeau, P. Vulliez
and S. Zeghloul
2 A Vision-Based Strategy for a Cost-Effective Flexible
Robotic Assembly System Without Using RCC Devices
and Compliant Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
C. Y. Weng and I. M. Chen
3 A Comparative Survey of Climbing Robots and Arboreal
Animals in Scaling Complex Environments . . . . . . . . . . . . . . . . . . 31
C. Webster, A. Jusufi and D. Liu
4 A Service Baxter Robot in an Office Environment . . . . . . . . . . . . . 47
Gu Fang and Brendan Cook
5 The Design of a New Rotary Hexapod with a Single Active
Degree of Freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
A. S. Fomin, V. A. Glazunov and J. K. Paik
6 Modelling and Model-Based Control of a Four-Bar Mechanism
with a Flexible Coupler Link . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Nguyen Van Khang, Nguyen Sy Nam and Nguyen Phong Dien
7 Structural Synthesis of 2-DOF, 7-Link, 2-Basic Loops Simple
and Multiple Joint Mechanisms for Robotics . . . . . . . . . . . . . . . . . 83
V. I. Pozhbelko and E. N. Kuts
8 Numerical Synthesis of Stephenson Six-Bar Mechanism
Using a CAD Geometric Approach . . . . . . . . . . . . . . . . . . . . . . . . . 95
Yanhua Zhang and Chunhui Yang
v
vi Contents
9 Foot Reaction Analysis of Whole Body Dynamic via Screw

Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Ming-June Tsai and Wan-Shan Yin
10 Comparison of Motion/Force Transmissibility in a 3-SPR
Parallel Manipulator and a 6-SPS Equivalent Mechanism . . . . . . . 119
M. Russo, M. Ceccarelli and Y. Takeda
11 Kinematic and Dynamic Dimensional Synthesis of Extended
Delta Parallel Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
J. Brinker, B. Corves and Y. Takeda
12 An Influence Based Error Identification for Kinematics
Calibration of Serial Robotic Manipulators . . . . . . . . . . . . . . . . . . 145
D. A. Patel, T. F. Lu and L. Chen
13 Autonomous Welding Seam Detecting and Tracking
Using Vision and Sound Sensors in Robotic Gas Metal
Arc Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Chao Chen, Gu Fang, Yanling Xu, Na Lv and Dinham Mitchell
14 Applications of a 3-Revolute Orientation Sensing Mechanism
(3-ROSM) in Controlling a Camera . . . . . . . . . . . . . . . . . . . . . . . . 177
Shasa A. Antao, Vishnu S. Nair and Rajeevlochana G. Chittawadigi
15 Contrast Enhancement Using Region Based Image Fusion
in Wavelet Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
S. Sharma, J. J. Zou and G. Fang
16 Real-Time Monitoring of Fatigue Cracks in Machine Parts
Using Microwave and Laser Imaging Techniques . . . . . . . . . . . . . . 199
P. Giri, S. Kharkovsky, B. Samali and R. Salama
17 Modeling and Characterization of the Double Helical
Compliant Joint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Rasheed Kittinanthapanya, Yusuke Sugahara, Daisuke Matsuura
and Yukio Takeda
18 Optimization of the Geometry of a Cable-Driven Storage
and Retrieval System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
C. Reichert and T. Bruckmann
19 Cascade ADRC Based Fault-Tolerant Control for a Hovering
PVTOL Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Xinli Xu and Chunwei Zhang
20 Power Analysis of Epicyclic Gear Transmission
for Wind Farm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
N. Hu, S. Liu, D. Zhao and C. Chen
Contents vii
21 Influences of Angular Position of Unbalanced Force on Crack

Breathing Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
H. M. Mobarak, Helen Wu and Chunhui Yang
22 Effects of Elliptical Crack Shape Ratio on Transverse Trajectory
of a Cracked Shaft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Joseph Spagnol, Helen Wu and Chunhui Yang
23 Seismic Performance Comparison of Base-Isolated
and Conventional RC Frames Compliant with ASCE 7–10
Using FAHP Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Longjun Xu, Yabin Chen and Qingyang Liu
24 A Pitman Style Fiber Bragg Grating Displacement Sensor
Based on Wedge Cavity Structure . . . . . . . . . . . . . . . . . . . . . . . . . 313
Sun Li, Su Zhongxin, Zhang Chunwei and Song Yansheng
Chapter 1
Architecture Choice of a Robotic Hand
for Deep-Sea Exploration Based
on the Expert Gestures Movements
Analysis
C. Mizera, M. A. Laribi, D. Degez, J. P. Gazeau, P. Vulliez and S. Zeghloul
Abstract In this paper, a method to choose the hand architecture of an end effector
used for deep-sea exploration will be presented. This method is based on the study
of the movements of an archeologist and on the analysis of its recorded gestures.
These observations allow an objective description of the archaeologist tasks, which
are for the most part specific to this activity. From this preliminary study and from
several criteria based on the fingertips workspace and on the quality of grasps, the
hand architecture has been defined. To sum up the study of the archeologist gesture,
a new taxonomy specific to this activity was established.
Keywords Design · Robotic hand · Motion capture · Submarine · Archaeology
1.1 Introduction
Human hand specificity is the ability to perform a wide range of grasps with objects
of different dimensions, masses, shapes and fragilities. Adaptability and human hand
dexterity are the main reasons why robotic grippers try to imitate our hands. Human
C. Mizera · M. A. Laribi (B) · D. Degez · J. P. Gazeau · P. Vulliez · S. Zeghloul

GMSC Department, PPRIME Institute, University of Poitiers, RoBioSS, CNRS, UPR 3346,
Poitiers, France
e-mail: [email protected]
C. Mizera
D. Degez
J. P. Gazeau
P. Vulliez
S. Zeghloul
© Springer Nature Switzerland AG 2019 1

R. Yang et al. (eds.), Robotics and Mechatronics, Mechanisms
and Machine Science 72, https://doi.org/10.1007/978-3-030-17677-8_1
2 C. Mizera et al.
grasp has been studied before to have a better understanding of the interaction
between objects and hand [1, 2]. This would allow the design of more effective
robotic hand that would reproduce human grasp with more precision. The study [3]
showed that some grasps are predominant for some activities while not used at all
for others. Consequently, an end effector adapted for a specific task or set of tasks
will not suit another.
In this paper, we will investigate the geometric design of a robotic hand that would
be used for submarine archaeology. For that purpose, we will observe and analyses the
gestures of an archaeological expert. Indeed, submarine exploration require specific
gestures to extract carefully fragile objects.
Few grippers of different shapes and dexterity level have been designed before for
this specific activity. Previous subsea end effectors were quite perfunctory, with only
two degrees of freedom like a simple grip, like for example the ORION manipulators
or the AMADEUS hand [4], and consequently unsuitable for the manipulation of
objects of different shape and fragility. More recent end effectors were made for that
purpose, like the HEU Hand [5], the Ocean One hand [6] or the Bologna University
hand [7]. These hands are more adaptable to different shapes of objects. Yet, they
only have three fingers, with only 2 DOF each for the Bologna University and the
Ocean One hand; and with 3 DOF each for the HEU hand. Consequently, they cannot
reproduce accurately the delicate and various gestures of the archaeologist. Moreover,
due to their restricted dimension, Ocean One and HEU hands can only grasps small
objects. Bologna University hand is able to grasp both small and large objects, but it
is quite bulky. Only the Bologna University hand have sensors giving an information
on the contact forces, which is necessary not to damage fragile objects. Last, these
hands are not made to go beyond a 100 m depth.
The paper is organized as follows: Sect. 1.2 presents the analysis of the task made
from the recording and the study of the archaeologist gesture. Section 1.3 presents
the choice of the hand architecture of the hand, based on the conclusions of Sect. 1.2.
Criteria will be defined to evaluate the hand workspace and the quality of the grasp
of spherical and prismatic objects. Conclusions and perspectives are presented at the
end of the paper.
1.2 Analysis of the Archaeologist Gesture
1.2.1 Recording of the Motions
In order to have a better understanding of the submarine archaeology, and to design a

robotic hand that would suit this activity, we proceeded to the analysis of the expert’s
gestures. To do this, we simulated an archaeological site by hiding selected objects
in the sand (Fig. 1.1). The objects match the different categories of artefacts we may
find on a wreck, and all of them require specific gesture and precaution. Based on
the expertise of the archaeologist diver, we selected very small and light objects like
1 Architecture Choice of a Robotic Hand for Deep-Sea Exploration … 3
Fig. 1.1 Experimental site
Fig. 1.2 Position of the markers on the hand
ceramic fragments or bullets, containers, broken ceramic object, bulky and heavy
object. We also disposed very fragile objects like what may be found undersea,
damaged by time and water.
Motion capture system was used to record the expert gesture. We used ten cameras,
MXT40, that record the marker coordinates at a frequency of 100 Hz. Passive markers
are placed on the hand and the arm of the archeologist, as depicted on Fig. 1.2. A
model composed by 23 markers is considered.
The placement of the markers allows the recording of every motions of flex-
ion/extension and abduction/adduction of the fingers.
4 C. Mizera et al.
Each finger can produce a movement of flexion at the level of the distal interpha-
langeal (DIP) joint, the proximal interphalangeal (PIP) joint and the metacarpopha-
langeal (MCP) joint. These joints are represented on Fig. 1.2.
1.2.2 Obtained Results
The conducted motion capture study and the observation of the expert gestures lead to
identify two main purposes of the hand in the submarine archaeology context. Firstly,
the hand can be used as a gripper, to grasp tools or underwater objects. Secondly, the
hand can be used directly as a tool, to search through the sand or to clear objects.
The workspace of the hand was identified, by computing all the points swept by
the extremity of the fingers for each task. The flexion angle as well as the joint speed
was computed for each joint on the hand.
1.2.2.1 Hand Used as a Tool
In these examples, the hand itself is used to perform the action. These observed
gestures are quite specifics to submarine archeology.
Observed gestures
Sweeping: As a layer of sediments often hides the objects, it is necessary for the
archeologist to clear the area in order to bring out objects and facilitate access. The
sweeping consists in slowly passing the hand over an object to remove the sand
covering it. This action is made without contact, thanks to the water movement. It’s
important to notice that the expert does not use its thumb during this gesture.
Scratching: When the sweeping is not enough to bring out entirely the desired object,
the expert have to scratch delicately the sand around it for the artifact to be completely
clear. The action of scratching is made with 1, 2, 3 or 4 fingers depending on the
amount of sediment that need to be removed, and on the estimated fragility of the
object. The thumb is not used for this gesture.
Shoveling: For shoveling action, the hand is put in flat configuration and slides under
the object to lift it without damaging. For that purpose, fingers have to push the sand
stuck under the objects.
Observations
For each of these actions we notified that:
• The thumb is barely used. It follows passively the other fingers most of the time.
• The flexion movements of the four fingers (index finger, middle finger, ring finger
and pinky finger) are almost identical. Nonetheless, the expert do not use its pinky
finger every time.
Fig. 1.3 Evolution of the joint configuration of the hand during the scratching gesture
Fig. 1.4 Range of abduction

of the fingers
• The MCP flexion as well as the PIP flexion are coupled. The coupling is approxi-
mately the same for the four fingers; but the coupling value is different and depend
on each action.
Scratching movement as well as sweeping movement do not request abduc-
tion/adduction movements of the four fingers (index, middle, ring and pinky). Exam-
ple of angles evolution during scratching movement are depicted on Figs. 1.3 and
1.4.
By looking at the joint velocity variations, we noticed that the movements are
slow. For the scratching and the swiping, the movements of flexion are composed
of regular cycles: mean cycle of 750 ms for the scratching gesture and mean cycle
6 C. Mizera et al.
Table 1.1 Maximal Joint Maximum rotational speed recorded (°/s)

rotational speed recorded
MCP 204
PIP 192
of 320 ms for the sweeping gesture. The maximum rotational speed recorded are
presented in Table 1.1.
1.2.2.2 Hand Used to Grasp Objects
The hand grasps two categories of objects: the archaeologic artefacts and a wide
variety of tools. The gesture realized to grasp tools are well documented in the
literature, like in Feix taxonomy [2]. Among them, we can find large tools used in
power grasp like hammers or trowel; delicate tools like scalpels or brush in precision
grasp; clamp used to seize tiny parts.
These tools are grasped by the handle, which can have various sizes and shapes.
We will have the possibility to adapt the shape of the handle of the tools to facilitate
the grasping.
Archaeological artefact needs precautions to be lifted. In the next, an atlas of
different artefact categories is presented.
Categories of objects
Very small objects: The small objects and the fragments are grasped with fingertips,
between two or three fingers. It is a precision grasp.
Fragile objects: The very fragile objects, which can disintegrate in the slightest con-
tact, are not directly grasped in the hand. It is necessary to slide a box under the object
to bring it to the surface still surrounded with sediments. This operation can be made
with one or two hands, knowing that the contents of the box can be heavy because we
keep the sand surrounding the object. During the grasp, fingers are placed stretched
out under the box and the thumb in opposition locks the grip.
Dislocated objects: The objects likely to come apart, like brush of brooms, are seized
compressed between two hands. It is important to cover the largest area possible
with the hand, and the fingers are widely spaced in order to secure the grip to the
maximum and avoid losing pieces.
Bulky objects: The very heavy or voluminous objects are taken with two hands. Their
geometry is very variable. According to their shape and to their fragility, they can
either be grabbed from below, with flat hands acting as support, or by sides with
hands in opposition. Some of these grips will be achievable with only one larger
hand.
Others: The other artefacts, which are the majority, are grabbed inside the palm. For
that, there are two possibilities. The hand slides under the object (see the shoveling
action) and the fingers deploy around the surface and conform to its shape. The thumb
may then be deployed in opposition to secure the grasp. The other scenario is that
Fig. 1.5 Evolution of the joint configuration during the grasp of an object
the object is seized by the fingertips and then placed inside the palm thanks to a wrist
rotation. The objects can be roughly divided in three categories, according to their
shapes: spheres, prism and cylinders.
Observations
For each of these gestures we notified that:
• The movement of the wrist is very important to place the hand in relation with the
object, and to dispose the object in the palm.
• The flexion of the four fingers (index, middle, ring and pinky finger) are synchro-
nized.
• The MCP flexion and the IPP flexion are coupled, but the coupling value depends
on the action.
For example on Fig. 1.5, we can observe the angular variation of the different
joints before the actual grasp of the object, on the left of the red line, and when the
grasp is effective on the right. We can also observe on this figure that the thumb joint
makes important movements to secure the grasp. In most of the grasping motion
observed, the thumb is very mobile.
There is few motion of abduction/adduction for the other four fingers during the
grasp of the same object type. Nonetheless when switching from an object type to
another, the angle of abduction varies significantly.
We observed that the coupling ratio between the PIP flexion and the DIP flexion
varies according to the objects being grasped. This ratio does not vary much during
8 C. Mizera et al.
Fig. 1.6 Coupling ratio between the PIP and the DIP flexion
a same gesture. On Fig. 1.6, we can notice the coupling ratio during the grasp of
different objects: a marble, a small ball, a plate, a prismatic box, a bottle and a large
bowl.
1.2.3 Data Analysis
The total fingertips workspace observed, for several gestures is represented on Fig. 1.7
relative to the coordinate system represented next to it. The points correspond to the
different positions of the fingertips during several gestures: sweeping, scratching,
grasping a small object with the fingertips and grasping a larger object in the palm.
All these points form the set WA that represent the archeologist workspace. It could
be contained in half a sphere of 100 mm radius represented on Fig. 1.7 in black.
The workspace center is situated 100 mm under the palm, centered relatively to the
y-axis with an offset toward the fingertips relatively to the x-axis.
The Table 1.2 gives the total angular range for all recorded motion. The angles of
abduction-adduction are measured between the grey line and the line perpendicular
to the red line as shown on the figure on Table 1.2.
The analysis of gesture leads to the following conclusions that will be important
for the hand design:
• The hand needs to have at least four fingers: three fingers on a side of the hand,
and one opposable thumb. Indeed, to lift an object in a flat hand, it is necessary to
maximize the contact surface between the hand and the object.
• Different actions are observed through the motion captures, from grasping fragile
objects to scratching the sand: the three fingers (index, middle and ring fingers)
act always simultaneously.
• It is necessary to have an opposable thumb to perform an efficient grasp. Through
the observation of the workspace of the thumb’s extremity during the different
grasping phases, its movement is preponderant. The observation of the hand’s
workspace during several movements shows that the thumb should have a wide
Fig. 1.7 Total workspace of the archaeologist fingertips
Table 1.2 Total angular range

Thumb Index Middle Ring Pinky
(°) finger finger finger finger
(°) (°) (°) (°)
MCP flexion 31–68 1–38 7–47 12–62 3–60
PIP + DIP flexion 2–58 8–62 9–91 2–66 0–83
Abduction/Adduction −31 −32 40–81 −30 −28
to 3 to 3 to 7 to 4
Reference lines
workspace. It should also be able to be in direct opposition with each of the

three other fingers. This result was also discussed by Kapandji and led to the
Kapandji test [8]. Consequently, the thumb will be able to proceed to a movement
of abduction/adduction in addition to the MCP flexion.
• A small amplitude movement of abduction/adduction for the index, middle, ring
and pinky fingers during a given gesture of the hand used as a tool or for grasping.
• The angle of abduction-adduction varies widely from a grasping configuration
to another. For example, the abduction-adduction angle is much wider for large
objects grasps than for small fragments.
• As the contact surface between the fingers and the object is maximized for the
grasps (except for the fingertips grasp of the smallest objects), it is better to have
at least three phalanges for the finger to conform to the objects shapes.
10 C. Mizera et al.
Based on this analysis, we built a taxonomy of the submarine archaeology activity

is presented in annex. All gestures presented above are illustrated in this taxonomy.
1.3 Choice of the Hand Architecture
The hand architecture definition consists in this preliminary study in defining the
configuration of fingers and thumb on the palm. This definition is based on the
evaluation of the workspace of the archeologist hand and on the stability of the
object.
Regarding the previous observations, several hand geometries seem to suit the
archeologist work with different placements of the thumb, as shown in the Table 1.3.
The only joints represented are the one leading to the abduction/adduction motion,
and the MCP flexion of the thumb.
Due to the compactness problem and the difficult environmental conditions under-
water, the under-actuation is one of the most important constraint to comply.
In order to select the best geometry able to perform the archeologist work an
evaluation procedure based on two criterions is proposed. The first criterion concerns
the total workspace of the hand, and the second one concerns the ability of the hand
to ensure stable grasps.
Table 1.3 Schemes of the possible hand architectures

Anthropomorphic (1) Opposite thumb (2) Thumb on the palm (3)
1.3.1 Total Workspace of the Hand
In this section, the fingertips workspace for each of the 3 investigated hand geometry
has been drawn and evaluated. All hand are considered with the same fingers length,
palm dimension and joints limits.
The dimensions as well as the joint limits of the considered robotic hand, whose
workspace and grasp quality are computed, are summarized up in Table 1.4. The
choice of these dimensions was based on a precedent work of the team [9], with a
1.5 factor.
To evaluate the workspaces, we compare them to the discretized workspace of the
archeologist hand obtained by motion capture, noted WA and represented on Fig. 1.7.
It is composed of approximatively 20000 points.
WA is the set of points Pa reached by the archeologist.
Pa(i) ∈ W A, i = 1, . . . , n, n ∼ 20000 (1.1)
WH is the set of points Ph reachable by the robotic hand. We create this set by
calculating the coordinates of the fingertips of the robotic hand. These coordinate
are obtained by sweeping, every 5 degrees, all the possible combinations of angular
position of the fingers, within the joints limits presented in Table 1.3. We get a set
composed of approximatively 25000 points.
Ph(i) ∈ W H, i = 1, . . . , m, m ∼ 25000 (1.2)
With,

Ph(i) = f q1i , q2i , q3i , θi , q1 pi , q2 pi , q3 pi , θ pi for i = 1, . . . , m, m ∼ 25000
(1.3)
and, q1i , q2i , q3i , θi , q1 pi , q2 pi , q3 pi , θ pi the angular parameters of the hand depicted
on Fig. 1.10.
One notes that the fingers 1, 2 and 3 are coupled and have consequently the same
MCP flexion angle q1 , PIP flexion angle q2 and DIP flexion angle q3 .
The parameter r is defined as a covering ratio and computed as follows.
n
Ci 1 i f Pa(i) ∈ W H
r= 1
With Ci = (1.4)
n 0 i f Pa(i) ∈
/ WH
Table 1.4 Dimensions and joint limits

a L1 = L2 = L3 = L abd θc(◦ ) q 1 (◦ ) q 2 (◦ ) q 1 p (◦ ) q 2 p (◦ ) θ(◦ ) θ p (◦ )
(mm) L1 p L2 p L3 p (mm)
(mm) (mm) (mm)
35 75 52.5 30 20 45 [0, 90] [0, 90] [0, 90] [0, 90] [0, 90] [−40, 40]
12 C. Mizera et al.
Table 1.5 Workspaces of the fingers

Anthropomorphic (1) Opposite thumb (2) On the palm (3)
Workspace 3.47 × 10−3 m3 4.03 × 10−3 m3 6.78 × 10−3 m3
volume
Ratio r 0.931 0.946 0.982
Workspaces
The two sets of points WH and WA are compared. If r = 1, it means that all the
points reachable by the archeologist hand are reachable by the robotic hand.
The workspaces of the hands and the obtained ratios r are shown on Table 1.5.
It is stressed that the anthropomorphic placement of the thumb is not ideal in
term of workspace criterion. This could be explained by superposition of the thumb
workspace and the third finger workspace, which reduce the total workspace and
leads to possible collision.
The third hand geometry presents a larger workspace, and covers the largest area
over the palm. Moreover, this geometry has the largest covering ratio r, which means
that the placement of the thumb allows to sweep a maximum of the points accessible
by the hand of the archaeologist. This ratio will be improved in the future work by
optimizing the geometric parameters of the robotic hand.
1.3.2 Ability to Make a Stable Grasp
The quality of a grasp depends mainly on its stability. It can be evaluated with several
criterions. In a previous work of the team, these criterions were listed and compared
[10].
The three different solutions for the thumb placement are consequently compared
over the quality of the fingertips grasps. For that, we qualify the grasp of objects with
simple geometric shapes that approximate the archaeologic artefacts. The grasps of
spheres with different diameters will be evaluated as well as prisms with different
thicknesses. Indeed, human grasps can be divided in two categories: prismatic and
circular grasp [1] and this shapes allows the evaluation of these grasps.
Figure 1.8 presents the implemented algorithm to compute the criterion evaluating
the grasp quality an object. Several constraints should be validated before computing
the grasp quality. These constraints concern the contact between fingertips and object
surface, the contact orientation and the force closure constraint.
Fig. 1.8 Computing flowchart of the criterion evaluating the grasp quality
The algorithm presented on Fig. 1.8 is implemented to handle two simple geo-
metric shapes: the sphere and the prism. The sphere is located on the center of the
palm. For the sphere, the distances dk are computed between the sphere center and
the fingertips of the four fingers. The radius of the sphere is noted by R as shown on
Fig. 1.8. The contact is possible if and only if:
dk = R for k ∈ [Finger 1, Finger 2, Finger 3, Thumb] (1.5)
For the prism with a thickness t, as illustrated on Fig. 1.8, the grasp is considered
done with three fingers. The three contact points between the fingertips, of the fingers
1 and 3 and the thumb, and the prism surfaces describe a triangle. The altitude of
this triangle is noted by D. The contact is possible if and only if:
D=t (1.6)
If the contact occurs, then it should be oriented toward the inside of the object,
which means that the fingers are located outside the object. This constraint is verified
through the computing of the scalar product between the normal to the surface contact,
14 C. Mizera et al.
Fig. 1.9 Contact between

the fingertips and the object
n, and the last phalanx vector, PQ (see Fig. 1.9). The contact is correctly oriented,
fingers outside the object, if and only if:
n · PQ > 0 (1.7)
For the evaluation of the force closure property, we use the Li Algorithm. The Li
Algorithm computes an angle α that evaluates how far the configuration of the grasp
is from the loss of the property. It is based on the evaluation of the friction cones
disposition at the contact points [11]. Our criterion uses the angle α as described on
Fig. 1.9. The grasp is “Force Closure” if the smallest angle of the friction cone is
non-null.
αmin > 0 (1.8)
The criterion used to determine the best geometry is the evaluation criterion called
“Volume of the ellipsoid in the wrench space” which defines the global contribution
of all the contact forces [12]. This criterion is defined by the grasp matrix G, which
depends on the position of the contact points between the fingertips and the objects
surfaces, computed thanks to the coordinates given below. It is noted Cr.
The matrix G establishes the relation between the fingertip forces f and the net
wrench applied on the object ω, and the relation between velocities at the contact
points v and the twist ẋ. We suppose that v is equal to zero since the object is
motionless.
ω = G · f and v = G T · ẋ (1.9)

Cr = det G · G t (1.10)
The architecture of the hand is presented on Fig. 1.10, with the different parameters
appearing in the expressions of the criteria used in the evaluation process of the hand
workspaces.
The coordinates of the fingertips of the four fingers are given by the following
equations:
Fig. 1.10 Architecture and parameters of the robotic hand
⎛ ⎞ ⎡⎛ ⎞ ⎤ ⎛ ⎞ ⎛ ⎞
x Finger 1 L abd x Finger 2 a
⎜ ⎟ ⎢⎜ ⎟ ⎥ ⎜ ⎟ ⎜ ⎟
⎜y ⎟ ⎢⎜ ⎟ ⎥ ⎜y ⎟ = M +⎜0⎟
⎝ Finger 1 ⎠ = R(θ, z) ∗ ⎣⎝ 0 ⎠ + M ⎦ + ⎝ Finger 2 ⎠ ⎝ ⎠
z 0 z Finger 2 0
⎛ Finger 1 ⎞
(a − L abd ) ∗ cos(θc )
⎜ ⎟
⎜ (a − L ) ∗ sin(θ ) ⎟
⎝ abd c ⎠
0
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
x z thumb −a
⎜ Finger 3 ⎟ ⎜ ⎟ ⎜ ⎟
⎜y ⎟ ⎜y ⎟ ⎜ ⎟
⎝ Finger 3 ⎠ = ⎝ thumb ⎠ = Ni + ⎝ 0 ⎠ f or i = 1, 2 or 3
z Finger 3 z thumb 0
⎡⎛ ⎞ ⎤
L
⎢⎜ abd ⎟ ⎥
R(−θ, z) ∗ ⎢ ⎜ ⎟
⎣⎝ 0 ⎠ + M ⎦ +
⎥
0
⎛ ⎞
(a − L abd ) ∗ cos(θc )
⎜ ⎟
⎜− (a − L ) ∗ sin(θ ) ⎟
⎝ abd c ⎠
0
16 C. Mizera et al.
Fig. 1.11 Evolution of the criterion for different sizes of grasped objects
Where θc is a constant and q1i , q2i , q3i , θi , q1 pi , q2 pi , q3 pi , θ pi are the joints param-
eters whose values vary between the joint limits presented on Table 1.4.
The matrix M is defined as follows:
⎛ ⎞
L1 ∗ cos(q1 ) + L2 ∗ cos(q1 + q2 ) + L3 ∗ cos(q1 + q2 + q3 )
M =⎝ 0 ⎠
L1 ∗ sin(q1 ) + L2 ∗ sin(q1 + q2 ) + L3 ∗ sin(q1 + q2 + q3 )
The matrix Ni is defined as follows for the three different placements of the thumb
depicted on Table 1.3.
⎛ ⎞
cos θ p ∗ L abd + L1 p ∗ cos q1 p + L2 p ∗ cos q1 p + q2 p + L3 p ∗ cos q1 p + q2 p + q3 p
⎜ ⎟
N1 =⎝ sin θ p ∗ L abd + L1 p ∗ sin q1 p + L2 p ∗ sin q1 p + q2 p + L3 p ∗ sin q1 p + q2 p + q3 p ⎠

L1 p ∗ sin q1 p + L2 p ∗ sin q1 p + q2 p + L3 p ∗ sin q1 p + q2 p )
⎛√ √ ⎞
2 − 2 0
⎜ √2 √2 ⎟
N2 =⎜
⎝
2 2 0 ⎟ ∗ N1 ;
⎠
2 2
0 0 1
⎛ ⎞
L1 p ∗ cos q1 p + L2 p ∗ cos q1 p + q2 p + L3 p ∗ cos q1 p + q2 p + q3 p )
⎜ ⎟
N3 =⎝ − sin θ p ∗ L abd + L1 p ∗ sin q1 p + L2 p ∗ sin q1 p + q2 p + L3 p ∗ sin q1 p + q2 p + q2 p ⎠

cos θ p ∗ L abd + cos L1 p ∗ sin q1 p + L2 p ∗ sin q1 p + q2 p + L3 p ∗ sin q1 p + q2 p + q3 p
Figure 1.11 presents the normalized value of the criterion “Volume of the ellipsoid
in the wrench space” Cr.
Cr
Cr = (1.11)
max(Cr )
The criterion Cr is computed for different sizes of grasped objects. When the
value is equal to zero, then it means that it is impossible to get a stable grasp.
One can see that the anthropomorphic placement of the thumb leads to slightly
better prismatic grasp and slightly less good sphere grasp. The two other geometries,
the cases 2 and 3, lead to very similar results. According to the obtained results, the
geometry with the thumb on the palm is chosen as the more suitable solution. This
third hand geometry presents the largest workspace, and covers the largest area over
the palm. The chosen geometry is given on the third column of Table 1.3 and detailed
on Fig. 1.10.
1.4 Conclusion and Future Work
In this paper, we were looking for a hand geometry that would suit the underwater
archaeology. For that purpose, we studied the expert gesture, using motion capture
system in simulated archaeological site. A new task taxonomy peculiar to submarine
archaeologist’s activity has been established.
From the data analysis of the joints evolution, the fingertips workspace and the
joints velocity, we selected the hand geometry that will be able to realize all the
archeology tasks identified. The more suitable geometry has been chosen among the
three candidates geometries that are evaluated.
Two different criteria have been defined and implemented in this purpose: the
workspace of the fingertips and the ability to make a stable grasp.
In future work, the optimal dimensions of the fingers and the palm will be inves-
tigated through an optimization problem. The under actuation will be also studied.
18 C. Mizera et al.
Annex: Taxonomy of the Submarine Archaeology Task
Task / Grasp OpposiƟon Thumb Virtual 1 or 2 Task / Grasp OpposiƟon Thumb Virtual 1 or 2
Type Picture Type Picture
name Type PosiƟon Fingers hands name Type PosiƟon Fingers hands
Two cupped
Sweeping UnderacƟon NA Adducted 2 to 4 1 hands Intermediate Palm Abducted 6 to 10 2
3 fingers 2 fingers 1 finger

Scraping UnderacƟon NA Abducted 2 to 4 1 Hand flat Precision Palm Adducted 3 to 5 1
Extension
Checking out Precision Pad Abducted 2 1 type (F) Power Pad Abducted 2 to 5 1
Palmar pinch
Shoveling Precision Palm Abducted 3 to 5 1 (F) Precision Pald Abducted 2 1
Packaging
light objects Intermediate Palm Adducted 3 to 4 1 Tripod Precision Pad Abducted 3 1
Packaging Medium wrap

heavy objects Power Palm Abducted 6 to 10 2 (F) Power Palm Abducted 3 to 5 1
Wriƫng
Cupped hand Precision Palm Abducted 3 to 5 1 tripod (F) Precision Side Abducted 2 to 3 1
Locked Adducted
cupped hand Intermediate Palm Adducted 3 to 5 1 thumb (F) Power Palm Adducted 2to 5 1
Pad +
Large objects Intermediate Palm Abducted 6 to 10 2 Clamp Power phalanx Abducted 2 to 5 1
Delicate
By the top Precision Pad Adducted 3 to 5 1 clamp Intermediate Pad Adducted 2 to 3 1
Friable
objects Power Palm Adducted 6 to 10 2 (F) : grasp presented in Feix taxonomy
References
1. Cutkosky, M.R.: On grasp choice, grasp models, and the design of hands for manufacturing
tasks. IEEE Trans. Robot. Autom. 5(3), 269–279 (1989)
2. Feix, T., Pawlik, R., Schmiedmayer, H.B., Romero, J., Kragic, D.: A comprehensive grasp
taxonomy. In: Robotics, Science and Systems: Workshop on Understanding the Human Hand
for Advancing Robotic Manipulation, pp. 2–3, June 2009
3. Bullock, I.M., Zheng, J.Z., De La Rosa, S., Guertler, C., Dollar, A.M.: Grasp frequency and
usage in daily household and machine shop tasks. IEEE Trans. Haptics 6(3), 296–308 (2013)
4. Lane, D.M., Davies, J.B.C., Casalino, G., Bartolini, G., Cannata, G., Veruggio, G., et al.:
AMADEUS: advanced manipulation for deep underwater sampling. IEEE Robot. Autom. Mag.
4(4), 34–45 (1997)
5. Meng, Q., Wang, H., Li, P., Wang, L., He, Z.: Dexterous underwater robot hand: HEU Hand II.
In: Proceedings of the 2006 IEEE International Conference on Mechatronics and Automation,
pp. 1477–1482. IEEE, June 2006
6. Khatib, O., Yeh, X., Brantner, G., Soe, B., Kim, B., Ganguly, S., Mullins, P.: Ocean one: a
robotic avatar for oceanic discovery. IEEE Robot. Autom. Mag. 23(4), 20–29 (2016)
7. Bemfica, J.R., Melchiorri, C., Moriello, L., Palli, G., Scarcia, U., Vassura, G.: Mechatronic
design of a three-fingered gripper for underwater applications. IFAC Proc. Vol. 46(5), 307–312
(2013)
8. Kapandji, A.: Clinical test of apposition and counter-apposition of the thumb. Annales de
chirurgie de la main: organe officiel des societes de chirurgie de la main 5(1), 67–73 (1985)
9. Mnyusiwalla, H., Vulliez, P., Gazeau, J.P., Zeghloul, S.: A new dexterous hand based on bio-
inspired finger design for inside-hand manipulation. IEEE Trans. Syst. Man Cybern. Syst.
46(6), 809–817 (2016)
10. Mnyusiwalla, H.: Qualité de prise dans le contexte de la planification de mouvements de
préhension et de manipulation dextre en robotique. Doctoral dissertation, Poitiers (2016)
11. Jin, J.-W., Liu, M.-H., Li, H.: A new algorithm for three-finger force-closure grasp of polygonal
objects. In: International Conference on Robotics and Automation, vol. 2, pp. 1800–1804 (2003)
12. Suárez, R., Cornella, J., Garzón, M.R.: Grasp quality measures. Institut d’Organització i Control
de Sistemes Industrials (2006)
Chapter 2
A Vision-Based Strategy
for a Cost-Effective Flexible Robotic
Assembly System Without Using RCC
Devices and Compliant Control
C. Y. Weng and I. M. Chen
Abstract This paper aims to provide a vision-based self-adaptive strategy to cope

with the uncertainties in a cost-effective robotic assembly system without the adop-
tion of RCC devices and compliant control. Several assumptions are given in the first
stage for the adaption in practical manufacturing. Then various approaches are taken
in our proposed robotic assembly system for dealing with a classical peg-in-hole
insertion process. In the end, the realistic implementations are carried out to verify
the effectiveness of our strategy.
Keywords Computer vision · Robotic assembly · Intelligent system
2.1 Introduction
Based on the fulfillment of sustainability, the demands of industrial robot technol-

ogy from high-growth industries are increasing [1]. As a result, the potential of
the utilization of industrial robots to realize the feature of flexibility in automation
manufacturing has become a main stream view, the value of investing efforts in the
robotic system integration is a forward-looking consideration. In practical, the main
advantage of the deployment of industrial robots is that they can be applied to a
wide range of feasible industrial applications have them be properly programmed
and integrated into manufacturing systems. Most industrial robots can be equipped
with various peripherals, such as different sensors or computer vision systems, pro-
viding the robots some utilizable feedback information on their decision processes,
thereby improving the robots’ abilities to cope with quite a few more arduous tasks
which may need a higher precision guidance or an object recognition process [2].
In most circumstances, the elements in a robotic integration system are an overhead
which cannot be ignored. The better the peripherals employed, the better the per-
formances of the robotic systems are presented. Sometimes, however, we can adopt
C. Y. Weng · I. M. Chen (B)

School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore,
Singapore

22 C. Y. Weng and I. M. Chen
with a simpler and cost-effective measure and peripheral equipment to implement a

robotic task, which can save considerable time and cost for us. In view of this, this
paper proposes a vision-based strategy for a cost-effective flexible robotic assembly
system without using RCC devices (hard 1 DOF grippers produced by 3D printing)
and compliant control (no path searching behavior during assembling).
2.2 Assumptions
The implemented strategy is based on several assumptions made to satisfy the realistic
flexible manufacturing scenarios. The assumptions are given as follows:
1. In most assembly manufacturing workstations, the assembly parts are first trans-
ported to the workstations after they are produced. Plus, after the completion of
assemblies, those assemblies are transported to other workstations. This kind of
behavior is commonly realized through conveyors. Therefore, we can assume
that there is a virtual transportation system in our robotic workspace, that is, the
assembly parts can appear at the beginning and disappear in the end.
2. The assembly parts are always presented in workstations with pre-determined
positions and orientations by the orienting systems to not only manifest the effec-
tiveness of the picking processes of robots but also facilitate the programming
of engineers. Furthermore, different forms of the presentation of assembly parts
are considered in realistic manufacturing scenarios, such as the utilization of
feeders, pallets, and magazines. In view of this, we simply conclude the features
of the presenting methods mentioned previously and assume that there is a week
presenting system to assist the robot, that is, the assembly parts appear in a ran-
dom position in each area, so do their orientation (only the orientation along the
normal direction of the workspace plane are random).
3. Intuitively, from the manual assembly of the scene to consider, workers generally
know what the product is to be assembled as well as the motion sequences to
assemble the product. As a result, we can assume the assembly parts and their
assembly processes are given to the robotic system as prior knowledge.
2.3 Methodologies and Results
Based on the assumptions made previously, we design the vision-based strategies

inspired by the action of charging a cell phone. An ordered motion sequence of
getting a wire terminal and connecting it on a cell phone is given as follows:
1. Find the rough position of the cell phone.
2. Determine how to pick up the cell phone.
3. Pick up the cell phone.
4. Find the rough position of the charging terminal.
2 A Vision-Based Strategy for a Cost-Effective Flexible … 23
5. Determine how to pick up the charging terminal.

6. Pick up the charging terminal.
7. Place (or fix) the cell phone.
8. Insert the charging terminal into the cell phone connector.
9. If we find that the charging terminal cannot be plugged into the cell phone con-
nector, we will generally re-confirm the pose of the charging terminal by eyes,
and then re-insert it again.
In summary of the above steps, we can conclude that the main vision-based strate-
gies are global detection, local detection, and error detection. However, we change
the step of error detection to an offline strategy to reduce the waste of time caused
by pose error in our robotic task, that is, we move the step to the beginning of the
insertion step. The detailed approaches are provided in the following sections.
2.3.1 Global Detection
The goal of global detection is to search the whole workspace plane and find the
rough position of every assembly part presented. In this step, if any of the assembly
parts is missing, the next step will not proceed by the robotic system. This approach
is realized by PCL with the utilization of point cloud data [3]. The working principle
is as follows:
1. Transform the point cloud data p ∈ R3×n to the robot base coordinate q ∈ R3×n ,
which is also the global coordinate:
⎡ ⎤
qx
q = ⎣ q y ⎦ = Tbase
sensor
p, (2.1)
qz
sensor
where n is the number of the points in point cloud data; Tbase is the homogeneous
transformation matrix from the robot base coordinate to the sensor coordinate.
2. Get the point cloud data only on the workspace plane by using the cartesian
threshold filter:
T
qi = qxi
q yi qzi

qxi , qxi ∈ [X L , X U ]
qxi =
N AN , other wise
, (2.2)
q yi , q yi ∈ [Y L , YU ]
q yi =
N AN , other wise

q zi , qzi ∈ [Z L , Z U ]
qzi =
N AN , other wise
where [X L , X U ], [Y L , YU ], [Z L , Z U ] are the lower and upper bounds of x-axis,

y-axis and z-axis in cartesian coordinate, respectively; The point with the element
NAN means that the point does not exist.
3. Remove noise by the radius outlier removal filter:

Visit all the existent qi ∈ q . If the number of the conditions that q j ∈ q satisfies

qi − q j ≤ r, i = j is no less than s1 , then qi is kept; otherwise, qi is removed.
Here both r and s1 are parameters depending on practitioners.
4. Get the cluster of each assembly part by means of the cluster segmentation
method:
Visit all the existent qi ∈ q obtained from the step 3. For those q j ∈ q satisfies

qi − q j ≤ ε, i = j, if the number of q j is of the interval [C L , CU ], then qi
and those q j are allocated into a cluster Ck . Here ε is parameter depending on
practitioners and k is the index of cluster.
5. Calculate the statistical position of each cluster. The result should be the rough
position of each assembly part:
nk
Pk = ql , ql ∈ Ck (2.3)
n k l=1
where Pk is the rough position of cluster Ck ; n k is the number of points in cluster

Ck .
Based on our assumptions made previously, the information of each assembly
part can be determined as we know which assembly part will appear in which fixed
area. Therefore, both the rough position and the information of each assembly part
can be obtained.
In our implementation, four assembly parts are presented on the workspace plane.
As the results are shown in Fig. 2.1, the rough positions of these four assembly parts
in the robot base frame “WAIST” are successfully captured.
Fig. 2.1 Results of global detection

2.3.2 Local Detection
After the completion of global detection, we can obtain the rough position of each
assembly part and move the hand-tip camera mounted on the end-effector to our
target. Then local detection can assist the robot to acquire the exact position and
orientation so that the end-effector of the robot can access the target assembly part
without missing and colliding. Noted that the realization of local detection should
be guaranteed that only one target assembly part is shown in the shooting range of
the hand-tip camera. This approach is realized by OpenCV with the utilization of 2D
RGB image data [4]. The working principle is described as follows:
1. Filter out the rough shape target assembly part based on the binary threshold:
T
pi =
pxi p yi pzi ∈ p

1, pxi ≥ T
pxi =
0, other wise , (2.4)

1, p yi ≥ T
p yi =
0, other wise
T
where p = px p y ∈ N2×1 is the grayscale image converted from the original 2D
RGB image; T is threshold value depending on practitioners.
2. Establish all the contours in image [5].

3. Select the contour with the largest area, C .
4. Perform the principal component analysis (PCA) [6] for the contour to get two
eigenvectors, λ1 (the x-axis of the target assembly part) and λ2 (the y-axis of the
target assembly part).
T
5. Calculate the center c = cx c y ∈ N2×1 and the rotation angle θ of the contour

C:

np
c= 1
np
pm , pm ∈ C

m=1
, (2.5)
−1 T 0
θ = cos λ2
1

where n p is the number of pixels in the contour C .
6. Compensate the pose of the end-effector:

xbase = −k c y − I y + L cx
ybase = −k(cx − Ix ) + L cy
Fig. 2.2 Results of local detection

π + θ, θ ≤ −π
θbase = (2.6)
θ − π, θ ≥ π
T
where k is the ratio of the pixel to the realistic dimension (meters); I = Ix I y ∈
T
N2×1 is the position of the middle pixel in the 2D image; L c = L cx L cy ∈ N2×1
is the position of the hand-tip camera w.r.t. the position of the end-effector joint. The
implementation result is shown in Fig. 2.2.
2.3.3 Error Detection
In this process, the detection is implemented by means of the 2D RGB image data
received by Kinect 2. The goal of error detection in this paper is to prevent the
position error caused by disturbances during the picking process. This approach can
be realized simply by calculating the relative position of the insertion part center to
the gripper claw center (which is also the end-effector center). The relative position
then is compensated to the end-effector pose to accomplish the assembly motion.
Let the line equation of one of the claws in 2D image is Ax + By = C, the center
T
of the claw is w = wx w y ∈ N2×1 and the center of the grasped object is po =
T
pox poy ∈ N2×1 . Then, the position compensation along the x-axis direction of
the robot base frame can be geometrically calculated by:

−ABpox +A2 poy −BC
2 +B 2 − wy
δ = B 2 pox −ABpoy −AC
A ∈ N2×1 (2.7)
A +B
2 2 − w x
The scene and implementation of error detection are shown in Figs. 2.3 and 2.4,
respectively.
Fig. 2.3 Scenario of error detection
Fig. 2.4 Results of error detection
The claw center and insertion part center are illustrated in Fig. 2.5.
2.4 Implementation and Results
The configuration of the integration system is presented in Fig. 2.6, which consists
of a sensor system (one Kinect 2 stereo camera and two hand-tip cameras), a pair
of symmetric grippers and four assembly parts including one-hole part and three
insertion parts. The goal of our implementation is to find the given peg part and
the given hole part from the workspace plane and then accomplish the peg-in-hole
insertion task by means of the proposed vision-based strategy.
Fig. 2.5 Illustration of claw center and the insertion part center
Fig. 2.6 Setup of robotic system
The implementation result of the peg-in-hole insertion task is shown in Fig. 2.7,
which basically involves the following action sequences, including global detection,
local detection, picking action, error detection, placing action and assembly action.
The time spent to accomplish this task with a fixtureless dual-arm robot is about
50 seconds.
Fig. 2.7 Implementation of task
Table 2.1 The adopted Sensor Quantity Price (USD$)

vision system
Microsoft Kinect V2 1 180
2-million-pixel camera 2 56
The adopted vision system is described in Table 2.1.

Besides, the motors for the grippers are two DYNAMIXEL MX-28-T which costs
USD$219.90 each. In addition, the time for producing the remaining 3D printing
components of the two grippers is about 10 hours.
2.5 Conclusion
This paper presents a vision-based strategy to realize a cost-effective robotic assembly

system to deal with the uncertainties in robotic assembly tasks. Under the assumptions
we made, our robotic system successfully and reliably accomplishes the peg-in-hole
insertion task of 0.1 clearance ratios, where the radius of the insertion part is 18
(mm). The successful implementation proves that the peripheral equipment in this
integration system is cost-effective enough to satisfy this assembly task without using
passive compliance features and other kinds of compliant control. In the future,
we will carry on more explorations on the hardware configurations and software
strategies which can be utilized in cost-effective robotic manufacturing systems.
Acknowledgements The author would like to thank for all the supports from the Robotics Research
Centre of Nanyang Technological University. This project is partially supported under National
Robotics Program RECT (SERC1622500003), Singapore.
References
1. The International Federation of Robotics: Statistics, Market Analysis, Forecasts, Case Stud-
ies and Profitability of Robot Investments [Internet]. World Robotics; c2017 [cited 2017 Jun
28]. Available from https://ifr.org/downloads/press/02_2016/Editorial_WR_Industrial_Robots_
2016.pdf
2. Turek, F.D.: Machine Vision Fundamentals, How to Make Robots See [Internet]. NASA Tech
Briefs; c2005-2016 [cited 2017 Jun 28]. Available from http://www.techbriefs.com/component/
content/article/ntb/features/feature-articles/10531
3. Rusu, R.B., Cousins, S.: 3D is here: point cloud library (PCL). In: IEEE International Conference
on Robotics and Automation (2011)
4. Bradski, G.: The OpenCV library. Dr. Dobb’s J. Softw. Tools Prof. Program. 25(11), 120–123
(2000)
5. Suzuki, S., Abe, K.: Topological structural analysis of digitized binary images by border fol-
lowing. CVGIP 30(1) (1985)
6. Pearson, K.: On lines and planes of closest fit to systems of points in space. Philos. Mag. 2(11)
(1901)
Chapter 3
A Comparative Survey of Climbing
Robots and Arboreal Animals in Scaling
Complex Environments
C. Webster, A. Jusufi and D. Liu
Abstract The inchworm-style climbing robot present in current literature will only
take us so far towards the robotic maintenance of transmission towers. To continue to
push toward developing robots that can perform work in reticular structures we must
consider a broader spectrum of animals for inspiration. The abilities of Primates in
climbing have long been a benchmark in climbing standards, but due to the mechani-
cal and control complexity associated with their development, they are seldom seen in
robotics. Birds (specifically Psittaciformes) offer an alternate solution, utilising less
degrees-of-freedom whilst maintaining stability and maneuverability. These ancient
arboreal specialists may hold they keys to unlocking the next stage in the devel-
opment of climbing robotics. This work presents lessons learned from a review on
primates as well as some preliminary observations on the climbing capabilities of
Psittaciformes.
Keywords Climbing · Robot · Reticular structure · Primate · Psittaciforme
3.1 Introduction
Recent advances in climbing robots have shown that versatile climbing robots are
increasingly becoming more viable for applications in monitoring, surveillance,
inspection, cleaning, painting, search and rescue [1–18]. However, none of the robots
so far developed come close to matching the performance of the specialised arboreal
(tree dwelling) animals which populate our planet. The ability to climb has evolved
in nearly every major animal class including mammals, birds, reptiles, amphibians,
insects, and even some species of fish like the Sicyopterus stimpsoni which climb
waterfalls in order to ascend to their breeding grounds [19]. Despite nearly two
decades of research into climbing robots, we still do not have a platform capable of
C. Webster (B) · A. Jusufi · D. Liu

Faculty of Engineering and Information Technology, Centre for Autonomous Systems,
University of Technology Sydney, Ultimo, NSW 2007, Australia

32 C. Webster et al.
both scaling and performing significant work on complex structures. The observed
capability gap relative to how animals climb is a true testament to the significance
of the challenges that nature has overcome when it comes to navigating the most
complex structures.
One of the most challenging environments to conquer are the compliant, retic-
ulated structure of vegetation such as trees and bushes. These environments are
characterized by their non-rigid, discontinuous supports. Being non-rigid, they yield
to the winds and the mass of animals attempting to traverse them, making them a
volatile medium to negotiate. Structures of comparable difficulty to scale can also
be found in civil infrastructure. Perhaps one of the most common examples of these
structures is the transmission tower, or electricity pylon (Fig. 3.1). In New South
Wales, Australia, approximately 15,000 of these style towers carry the high voltage
conductors that enable electricity to be transported about the state. Moving over to
a more populous country like Japan, some 50,000 towers populate the Tokyo region
alone. These space-frame type structures are also often found in bridges, the roofs
of large open buildings and stadiums, construction scaffolding, human-made satel-
lites, and space structures. They are found anywhere where a high strength-to-weight
ratio is desirable. The primary differences between our human-made structures and
naturally occurring ones is the uniformity and stiffness of the structure and its mem-
bers. Civil infrastructure is typically made up of rigid members, though tensegrity
structures where rigid members are held in place by compliant tensile cables are also
not completely uncommon. Trees also vary from relatively rigid to non-rigid as we
move away from the trunk and toward the extremities. The opportunities for the use
of robotics in the maintenance of civil infrastructure was described by Liu et al. [20].
As with all civil infrastructure, transmission towers require periodic maintenance to
ensure their longevity. This is a costly and dangerous job for humans, and hence a
prime candidate for automation should the technology be up to the challenge.
Unfortunately, the technology is not yet up to the task for large-scale deployment
of platforms that might be capable of work in this environment. Physically, the
knowledge required to develop the technology is embodied in the adaptations of the
arboreal animals who navigate these structures. Whilst we’ve come a long way in
understanding the fundamental operating principles of many animals, we’re still yet
to truly understand them in a way that enables us to create truly versatile machines that
can be deployed to perform work. In this paper, we review previous work on robotic
transmission tower maintenance and robots operating on reticular structures from a
more general view point (Sect. 3.2). In Sect. 3.3 we present the results and discussion
on a review of literature on arboreal primates (with a focus on Gibbons) and discern a
few of the characteristics of their morphology that are suitable for adoption in robotic
development. We also look at climbing birds and the lessons they have to offer. As the
material on climbing birds is limited, we also present the preliminary observations
on climbing parrots, identifying what typical manoeuvers these arboreal tripeds are
capable of, and how this could be of benefit in the development of a psittacine-
inspired climbing robot. Utilising the information gathered on primates and birds we
propose the design of some semi-structured environments on which we can further
test the capabilities of these specialised arboreal animals.
3 A Comparative Survey of Climbing Robots and Arboreal Animals … 33
Fig. 3.1 a An example of a reticular structure, a common Electricity Transmission Tower, Mel-
bourne, Australia. b A common tower node. c Two connecting planar sections
3.2 Climbing Robots
To our knowledge, the first known study on the automation of transmission tower
maintenance was carried out by the Electricité de France transmission network [21]
in 1993. This work included the feasibility of using a teleoperated manipulator arm
to help with both tower and line maintenance procedures. At the time, the robot being
considered was essentially a large hydraulic manipulator arm on the back of a 20-ton
wide-bodied truck that would be operated remotely. It was dismissed as infeasible
due to number of constraints, the foremost being that only 30% of towers would be
accessible with a truck of such size. In 2004, Golightly [22] presented a mobile robot
for inspecting electricity transmission support towers. Unfortunately, the preliminary
work produced only a robot capable of climbing on a simplified planar representation
of a tower and was never deployed on an actual tower. To the best of our knowledge
no further transmission tower specific attempts have been made since.
In the realm of climbing robots, one concept has gained significant traction. When
it comes to reticular structures, the inchworm concept (Fig. 3.2) appears to have seen
more development than any other [1–10]. Impressive as the development has been,
this concept is mostly limited to inspection tasks only. This is due to it having a
maximum of two points-of-contact with the substrate at any one time. Should the
robot be using one of those points of contact to perform maintenance tasks, the robot
is left in a risky and potentially unstable state. Falling from the substrate could cause
significant damage to any robot or animal alike, and as such more redundancy is
required when performing loaded tasks, such as maintenance. From the single can-
tilever support, it is also difficult to design lightweight contact points that are capable
of reacting the moment generated when the centre-of-mass (CoM) is a significant
distance from the substrate. Typically, the adhesion mechanism on these robots are
some of the heaviest parts due to this fact, and due to their weight, they in turn
increase the load on the opposing side of the robot. Careful optimisation has to be
conducted during the design to ensure the increasing mass does not feedback on
itself. The benefits of using the inchworm approach is that you can truly optimise for
low degrees-of-freedom (DoF). This is paramount when trying to reduce the overall
mass of the robot. From a control perspective, this single serial chain is also easier
to implement than a more complex multi-limbed system.
Assuming that the goal is to develop robots that are autonomously capable of
maintaining transmission towers (and other reticular structures), and going on the
assumption that the inchworm concept can only take us so far, it stands to reason
that new inspiration is needed. In the following sections, we will explore vertebrate
morphologies and climbing techniques, in Primates and Psittaciformes to draw new
inspiration.
3.3 Results and Discussion
3.3.1 Primates
Primates are an order of animal that include all the greater and lesser apes, old and
new world monkeys, as well as lemurs, lorises and an assortment of other small
quadrupedal mammals. All primates share an innate ability to climb, with several
species classified as arboreal. Amongst primates, Gibbons are perhaps the most
impressive when it comes to rapidly traversing reticular structures. Gibbons are
one of the few animals capable of sustained brachiation and have been described
as the only true brachiators [23]. Brachiating is described best as, “continuous bi-
manual locomotion that is not interrupted by other positional behaviours nor given
support from the tail or hind limbs” [24]. Spider monkeys share this unique form of
locomotion, except they often use their prehensile tail for additional support which
violates the definition given by Hollihn [24]. Using this mode of locomotion, Gibbons
are capable of navigating the tree tops at up to 40 km/h and appear to be doing so
almost effortlessly. So well is this known, that a significant amount of research has
been performed analysing and simulating the mechanics of their motion [25–29].
The exceptionality of the gibbon’s ability to brachiate has not gone unnoticed by the
engineering community either, with research being undertaken to understand and
mimic their abilities [30–33]. Whilst brachiating is an interesting topic for research
it does not give us significant insight into how to undertake stable and continuous
Fig. 3.2 Inchworm Climbing Robots. a CROC2 (1). b Climbot (2). c Shady3D (6). d 3DClimber
(7)
vertical climbing, which is at this stage a much more useful skill for our robots to
possess.
3.3.1.1 Climbing Gaits
In the field of biology, significant work has been done to identify the climbing char-
acteristics of diverse taxa of hominoid [34–37]. These studies give us insight into the
climbing styles of animals closely related in size and morphology to humans. When
looking just at the hominoids (lesser and greater apes), Isler [36] categorised ver-
tical climbing into two distinct styles, extended-elbow and flexed-elbow climbing.
Flexed-elbow climbing requires the animal to be able to grasp the substrate with a
single hand. Typically, once started hominoids will enter a regular quadrupedal gait
when climbing utilising either a diagonal sequence (Rh, Right hand; Lf, Left foot;
Lh, Left hand; Rf, Right foot) or a lateral sequence (RhRfLhLf). Gibbons on the other
hand fail to enter this regular gait, switching between the two gaits with instances of
Fig. 3.3 Extended-elbow

climbing
three-legged (tripedal) strides in between [35]. They exhibit a large variety of footfall
patterns with only a very slight preference for the diagonal sequence. Such irregular
climbing strategies could be a reflection of the unstructured and irregular nature of
their habitat. Unfortunately, to our knowledge no work has been done to explore this
hypothesis; so for now we can only speculate that their particular climbing style is
somehow related to their impressive mastery of locomotion in reticular structures.
Extended-elbow climbing is a climbing mode used by hominoids when climbing
trees of a large diameter. The large intermembral indices (ratio between arm and
leg length) of the apes facilitate this style of climbing and enable them to climb
trees with much larger diameters than that of smaller primates. By keeping their
elbows relatively straight and leaning back from the substrate, apes appear to use
the moment generated from holding their CoM away from the substrate to increase
friction at the contact points. From a physics perspective, this principle is easy to
understand. Figure 3.3 illustrates a static 2-dimensional idealization of the forces in
play during this kind of locomotion. This mode of climbing might be something that
we are easily able to replicate in a climbing robot, especially if we wish to support
significant payloads. It is used successfully by Gorillas with masses up to 200 kg.
As friction is proportional to the normal force applied at the contact point, if we
can prove that N ∝ d 1 , then we can show that the distance the CoM is held away from
the substrate is proportional to maximum frictional force achievable. Assuming the
gorilla in Fig. 3.3 is a rigid-body in static equilibrium, and taking moments about
point O, it can be found that:
mg
N= d1 (3.1)
h − µd2
We can therefore conclude that normal force, is proportional to the distance the
CoM is away from the substrate. Due to the success of this strategy in Gorillas,
and the easily understood physical phenomenon that enables it, this strategy could
potentially be used in robotics to carry significant payloads up a vertical columns of
large relative diameters.
3.3.1.2 Mass Distribution–Limb Inertia
It can be observed that in most animals there is a proximal (close to body) to distal
(away from body) distribution of mass along each of the limbs, with the majority of
the mass being located at the proximal end. This can also be observed in human arms
and legs, with the bigger muscle groups being located nearer to the centre of mass than
the smaller. Whilst it is observed in most animals, this distribution is exaggerated
in the gibbon limb [29]. In gibbons, this is observed especially in the forearms,
with the muscles for finger actuation being located significantly further away from
the wrist compared with other hominoids. The radial shafts in the gibbon forearm
are also sagittally (toward the centre plane) thicker, which is a feature seen only
in gibbons, even compared to other brachiating animals such as the spider monkey
[23]. This distribution is present in far more than just mammals however. The limbs
of arthropods, birds, fish all seem to follow this morphology. Take the hind limbs
of a grass hopper as another example. The legs are characteristically thin compared
with the body, and upon close inspection a similar distribution can be observed.
Looking at the caudal fin of a fish, the tail tapers down before expanding out into
the large lightweight fin meant for pushing water. For most animals, locomotion
takes places by constantly accelerating and decelerating their appendages or some
portion of their body. To make this as energy efficient as possible the moment of
inertia of the appendage should be reduced as much as is practicable to reduce the
energy that is wasted in accelerating limbs. This allows the animal to move with
more speed whilst consuming less energy. From an evolutionary perspective, this is
beneficial as it allows the individual to spend more energy raising young or allows
more time between meals. It should be noted that we are not claiming that this is
the major reason for this evolutionary adaptation, only that it can be observed in
most animals. In robotics, the implementation of this principle may result in longer
operating times, quicker response characteristics, and lower torque motors without
sacrificing performance; giving ultimately greater efficiency.
3.3.1.3 Strength-to-Mass Ratio
Whilst it is important to reduce the inertia of rotating components for the sake
of efficiency, the mass of the vehicle and its ability to accelerate should also be
properly considered. Whilst a high power-to-mass ratio might correlate with good
dynamic abilities, its appropriateness will ultimately depend on the intended function
of the robot. Regardless of the intended function however, mobile robots need greater
power-to-weight ratios than their static industrial cousins. It will only be when we’re
capable of developing a truly lightweight and agile robot that we will need to perform
an optimisation of power-to- weight versus intended function. For now, we need to
push forth and simply continue to develop robots that might one day match or exceed
the performance characteristics of arboreal animals.
The major requisites that allow animals to be so light and strong are their muscles,
and the lack of a significant stiffness constraint. Natural muscle has a very advanced
design when compared with actuator technology used in modern robotics. They
are light, strong, have adjustable compliance, built in sensory feedback, and are
very scalable. The human body alone has approximately 650 skeletal muscles [38],
and several million micro-muscles distributed on the skin and hair follicles. These
muscles work synergistically to actuate the human skeleton in the 244 DoF present,
although bio-mechanists will often represent this in simpler models when analysing
internal forces or simulating movement [39]. This level of complexity in a robot is
unheard of, primarily because the mechanical design, control complexity, and overall
system would be impracticably complex. Even if a mobile robot of this kind were
attempted with our current actuator technology, it is likely the robot would end up
being too heavy to have substantial locomotion capabilities.
This principle of having a high power-to-weight ratio is even more important for
climbing robots as it is in other mobile robots. This is because a climbing robot
must be capable of accelerating its mass directly against gravity. This is reflected in
nature by the very rare occurrence of heavy arboreal animals. The largest animals
that are capable of climbing we have encountered are the Lowland Gorillas, with the
weight the largest found weighing 211 kg [40] to our knowledge. Although these
Gorillas are capable climbers they do not do so without great need, are relatively
slow at it, and can hardly be considered arboreal spending the majority of their time
on the ground. Isler [36] noted the pronounced difficulties that adult male Gorillas
had when attempting to climb, so it might be reasonable to assume that these large
gorillas are approaching physical limitations when it comes to climbing mammals.
More arboreal animals like the Pongo Pygmaeus Abelii (Orangutan) are on average
much lighter, with the average weight of the adult animal in one study being around
50 kg [36]. Highly arboreal animals such as the Gibbon who live their whole lives
in trees, and who depend on being capable climbers to reach food sources are much
lighter. The average mass of the gibbons in one study was just 8.3 kg [29]. Whilst the
discussion from an evolutionary perspective on why these animals are lighter than
their cousins is out of scope in the engineering discipline, biologists have noted that
“the lightweight gibbons climb more easily than the heavy great apes” [36].
3.3.2 Birds
Birds have existed since the late Jurassic period some 150 million years ago. Arguably
beginning with Archaeopteryx (“original bird”), birds have utilised their abilities
to climb and navigate trees to give them significant evolutionary advantage since
before major diversification and radiation of mammals. Interestingly Burnham and

Feduccia [41] described these primitive birds as being more functionally analogous
with modern day mammals than with modern day birds. It is also noted that climbing
abilities in birds with specialised flight capabilities is almost always connected with
specialised scansorial foraging. In our own observations when foraging birds make
small motions, where flying or hopping may not be energetically efficient when
compared with walking/climbing. This of course depends on the size and mass of
the bird. One such bird that spends the majority of its time foraging on tree trunks is
the Treecreeper, as documented by Norberg [42]. This bird, which is closely related
to the woodpecker, spends most its days hopping up tree trucks searching for food,
typically using its wings simply to glide to the bottom of the next tree. This is
described as being the most energetically efficient method for this bird in finding
food [43]. The utility of both legs and tail is notable in the climbing gait of this bird.
The tail is utilised as a third leg to help support and counteract the moment generated
by the centre of mass (CoM) being offset from the substrate. During the aerial phase
of locomotion, this bird uses a quick flick of its tail against the substrate to begin an
angular acceleration opposing the pitch-back that was generated when pushing off
the trunk in the power phase. Interestingly birds are not the only animals that utilise
their tail in this way. An active tail use has also been found in rapid vertical running
of geckos as they overcome perturbations whilst climbing [44].
Whilst studies on the mechanics of vertical climbing in birds are scarce, we
hypothesise that such studies could have significant impact in the engineering con-
text for the development of climbing robots scaling structures with discontinuous
supports. The primary reason for this is that compared with climbing mammals,
birds utilise less DoFs to perform comparable manoeuvres. Birds tend to have fewer
bones, bones that are reduced in size, and a fused hip structure when compared with
mammals [45]. This is advantageous for robotics as typically increased DoFs are
correlated with increases in mass, which directly affects the feasibility of a climbing
robot with modern actuators. Control complexity is also made more complex with
greater DoF, and so it stands to reason that a robot inspired by bird morphology may
have advantages over one inspired by quadrupedal mammals.
One order of birds we have not been able to find any existing documentation on
their climbing mechanics is the Psittaciformes (Parrots). In our observations, we
have found that parrots display unique climbing styles when compared with small
birds like the Treecreeper. Smaller birds tend to do a lot more hopping and flying
for small distances, whilst the larger heavier parrots tend to do a lot more walking
and climbing. This may be due to the significant increase in energy needed for flight
in large birds compared with small birds. Similarly, with robots the energy required
to accelerate appendages increases cubically or faster with their length [46]. As any
useful maintenance robot will likely be an order of magnitude larger than some of
the larger parrots, we speculate that analysing the gaits of a range of parrots sizes
will help us to determine the appropriateness of their particular climbing strategy in
a robot. It is difficult to say that the gaits these birds utilise can be reproduced by
a psittacine-inspired robot of a larger scale, however, the parrot’s abilities do give
us a fresh starting point in the development of multilimbed climbing robots. To give
Fig. 3.4 Ground-to-structure transfer of an Ara chloropterus (Red-and-Green Macaw). In this

figure, the bird transfers from the ground to the structure. Notice the use of the tail in the initial
transfer in frames c and d . From Frame c–d the bird utilises only its beak and tail for support whilst
moving both its feet up to the substrate simultaneously. This happened again from frame f–g. It is
noteworthy that every time the bird releases its beak from the substrate it appears to be pushing the
tail against the substrate. More data needs to be gathered to confirm this. Once on the cage structure
the bird changes from using two feet each stride to single steps, moving the beak position after each
step. If we count the tail as a support, a minimum of 2 points of contact are used at all times. Source
https://youtu.be/BGYuX2eWx6o [47]
Fig. 3.5 A parrot performing vertical rope climbing. In this figure, we observe a parrot climbing
vertically on a non-rigid chain. Whilst the member is non-rigid it does not appear to deflect signif-
icantly under the mass of the bird, though it does sway and twist as the bird climbs. Climbing is
facilitated by first moving beak, and then by moving a foot either above or below where the beak
was placed. After the foot is placed the beak is again moved further up the chain. This style of
climbing is not unlike the inchworm method, with the exception that a minimum of two points of
contact are utilised at all times. We notice in frame d the bird using its wing to actively redistribute
its mass, which appears to be assisting in balance. Source https://youtu.be/WEXSY_YbxG0 [48]
some insight into what these parrots are capable of we have included some of our
preliminary observations on their climbing abilities and styles in Figs. 3.4 and 3.5.
3.3.3 Analysis of Gaits In Situ
Drawing from general principles of animal locomotion and observations of how

arboreal specialists (e.g. bird or a primate) climb complex natural structures, we can
design semi-structured tracks to study their behaviours in greater detail, in order
to derive biologically inspired solutions for scaling reticular structures resembling
a transmission tower. In this spirit, we can assess how an animal might approach
specific scaled down sections of a transmission tower based on the study and obser-
vations presented in Sects. 3.3 and 3.4. Braking down the transmission tower into
base components and analysing what transitions and manoeuvres are possible on
each section, we can speculate what method an animal would utilise to perform the
manoeuvre. If we treat climbing a full tower as a sequence of smaller sections we can
Table 3.1 Basic manoeuvres for transmission tower climbing

ID Description Figure
M1 Motion on a solitary beam 2
a Parallel to beam axis 2a
b Radial to beam axis 2b
M2 Beam and plane transfers 3
a Beam-to-beam transfer 4
b Beam-to-plane transfer 4
c Plane-to-plane transfer 5
M3 Moving on a 2D plane 5
M4 Free movement utilising and available beams within reach 6
begin to make predictions on effective strategies for how an animal (or robot) might
be able to climb around this structure. The manoeuvres in Table 3.1 are each per-
formed on specific sections of the tower. These sections and manoeuvres are further
defined below.
Manoeuvres designated as M1 (Fig. 3.6a) describe motion along a solitary beam
of any orientation (e.g. any inclination or rotation). The animal must move along this
beam without contacting any other member of the tower. To complete this manoeuvre,
it is expected that a primate could use either flexed-elbow or extended-elbow climbing
depending on whether a table hold is achievable with a single hand. For a bird
depending on the orientation it might either use a side-stepping gait, or something
more similar to the behaviour shown in Fig. 3.5.
M2a to M2c (Fig. 3.6c–e) describes various types of transfers, i.e. from plane to
plane, from beam to beam, etc. To perform these manoeuvres a primate would almost
certainly have to use flexed-elbow climbing, which means that it would have to be
sufficiently large in proportion to the structure. A bird might use the gait shown in
Fig. 3.4 or similar.
M3 describes simply moving along a plane that has the majority of the surface
removed. Just as with the transfers a primate would likely use flexed-elbow climbing
and would need to be sufficiently large compared to the structure in order to grasp
the members. Depending on the orientation a bird might use its vertical climbing
gait or a gait similar to frames c–d of Fig. 3.4, whereby both feet are simultaneously
moved while the beak provides the majority of the support.
M4 manoeuvres are those that did not fit into out simplified models, and deal
with truly 3D locomotion. Where M1 deals with motion primarily in one dimension,
M3 deals with motion primarily in two dimensions, M4 deals with motion in 3-
dimensional space. Based on the literature presented it is hard to speculate exactly
how either animal would tackle these issues although almost certainly the primate
would strive to retain a minimum three points-of-contact (depending on the species,
and the bird would strive to maintain a minimum of one to two points-of-contact
with support from the tail where possible.
Fig. 3.6 Visual depiction of

manoeuvres
3.4 Conclusion
There are numerous challenges associated with discovering the mechanisms by which
arboreal specialists move in their natural habitats. By designing structured tests that
approximate some of the complexity of the habitats in which they have evolved (e.g.
discontinuous supports, limited number of perches, non-rigid/compliant), we can
begin to uncover specific stereotypical climbing behaviours, strategies and transi-
tions. This set of bio-mechanic measurements can in turn inform our models and
advance our understanding of the salient features of their climbing dynamics for the
understanding of how control of locomotion could be simplified. In this work we
have presented some of the most current robots for traversing reticular structures
with discussion on their benefits and shortcomings. We concluded that the current
robots do not enable us to do significant maintenance work in reticular structures.
This is not a function of the technology being “up to scratch”, but with the fundamen-
tal morphology of these robots. As such we begin to explore what we can learn from
several climbing animals, looking specifically at primates and birds. Birds appear
to be the better bet when it comes to producing a functional climbing robot due
to their lower degrees-of-freedom. Though, the psittacine-inspired concept is still
significantly more complex than the inchworm robot and is underexplored. Though,
the prospect for its development in coming years looks promising.
Acknowledgements We would like to thank Dr. Wesley Au for the time spent revising the Motion
Primitives for Transmission Towers. We also thank the Tokyo Electric Power Company, the Centre
for Autonomous Systems and the University of Technology Sydney for funding this research.
References
1. Yang, C.J., Paul, G., Ward, P., Liu, D.: A path planning approach via task-objective pose
selection with application to an inchworm-inspired climbing robot. In: 2016 IEEE International
Conference on Advanced Intelligent Mechatronics (AIM), pp. 401–406 (2016)
2. Guan, Y., et al.: Climbot: a modular bio-inspired biped climbing robot. In: 2011 IEEE/RSJ
International Conference on Intelligent Robots and Systems, pp. 1473–1478 (2011)
3. Guan, Y., et al.: Climbot: a bio-inspired modular biped climbing robot—system development,
climbing gaits, and experiments. J. Mech. Robot. 8(2), 21026 (2016)
4. Balaguer, C., Gimenez, A., Jardón, A.: Climbing robots’ mobility for inspection and mainte-
nance of 3D complex environments. Auton. Robots 18(2), 157–169 (2005)
5. Balaguer, C., Giménez, A., Pastor, J.M., Padrón, V.M., Abderrahim, M.: A climbing
autonomous robot for inspection applications in 3D complex environments. Robotica 18(03),
287–297 (2000)
6. Yoon, Y., Rus, D.: Shady3D: a robot that climbs 3D trusses. In: Proceedings 2007 IEEE
International Conference on Robotics and Automation, pp. 4071–4076 (2007)
7. Tavakoli, M., Marjovi, A., Marques, L., de Almeida, A.T.: 3DCLIMBER: a climbing robot for
inspection of 3D human made structures. In: IEEE/RSJ International Conference on Intelligent
Robots and Systems, 2008, IROS 2008, pp. 4130–4135 (2008)
8. Kotay, K.D., Rus, D.L.: Navigating 3d steel web structures with an inchworm robot. In: Pro-
ceedings of the 1996 IEEE/RSJ International Conference on Intelligent Robots and Systems’
96, IROS 96, vol. 1, pp. 368–375 (1996)
9. Lam, T.L., Xu, Y.: A flexible tree climbing robot: Treebot-design and implementation. In: 2011
IEEE International Conference on Robotics and Automation (ICRA), pp. 5849–5854 (2011)
10. Pagano, D., Liu, D., Waldron, K.: A method for optimal design of an inchworm climb-
ing robot. In: 2012 IEEE International Conference on Robotics and Biomimetics (ROBIO),
pp. 1293–1298 (2012)
11. Luk, B.L., Collie, A.A., Billingsley, J.: Robug II: an intelligent wall climbing robot. In:
1991 IEEE International Conference on Robotics and Automation, 1991, Proceedings,
pp. 2342–2347 (1991)
12. Luk, B.L., Collie, A.A., Piefort, V., Virk, G.S.: Robug III: a tele-operated climbing and walking
robot. In: UKACC International Conference on Control’96 (Conference Publication Number
427), vol. 1, pp. 347–352 (1996)
13. Bretl, T., Rock, S., Latombe, J.-C., Kennedy, B., Aghazarian, H.: Free-climbing with a multi-use
robot. In: Experimental Robotics IX, pp. 449–458. Springer (2006)
14. Clark, J., et al.: Design of a bio-inspired dynamical vertical climbing robot. In: Robotics:
Science and Systems (2007)
15. Provancher, W.R., Jensen-Segal, S.I., Fehlberg, M.A.: ROCR: an energy-efficient dynamic
wall-climbing robot. IEEE/ASME Trans. Mechatron. 16(5), 897–906 (2011)
16. Faina, A., Souto, D., Deibe, A., Lopez-Pena, F., Duro, R.J., Fernández, X.: Development of a
climbing robot for grit blasting operations in shipyards. In: IEEE International Conference on
Robotics and Automation, 2009, ICRA’09, pp. 200–205 (2009)
17. Haynes, G.C., et al.: Rapid pole climbing with a quadrupedal robot. In: IEEE International
Conference on Robotics and Automation, 2009, ICRA’09, pp. 2767–2772 (2009)
18. Chu, B., Jung, K., Han, C.-S., Hong, D.: A survey of climbing robots: locomotion and adhesion.
Int. J. Precis. Eng. Manuf. 11(4), 633–647 (2010)
19. Blob, R.W., et al.: Morphological selection in an extreme flow environment: body shape and
waterfall-climbing success in the Hawaiian stream fish Sicyopterus stimpsoni. Integr. Comp.
Biol. 48(6), 734–749 (2008)
20. Liu, D.K., Dissanayake, G., Miro, J.V., Waldron, K.J.: Infrastructure robotics: research chal-
lenges and opportunities. In: ISARC, Proceedings of the International Symposium on Automa-
tion and Robotics in Construction, vol. 31, p. 1 (2014)
21. Soler, R., Guillet, J.: Robotic maintenance of the EDF transmission (63 to 400 kV) network:
feasibility study and effects on tower design. In: Sixth International Conference on Trans-
mission and Distribution Construction and Live Line Maintenance, 1993, Proceedings from
ESMO-93, pp. 459–468 (1993)
22. Golightly, I.: Development of a robot for inspecting electricity transmission support towers. In:
Mechatronic Systems 2004: A Proceedings Volume from the 3rd IFAC Symposium, Sydney,
Australia, 2005, vol. 1, 6–8 Sept 2004
23. Takahashi, L.K.: Morphological basis of arm-swinging: multivariate analyses of the forelimbs
of Hylobates and Ateles. Folia Primatol. 54(1–2), 70–85 (1990)
24. Hollihn, U.: Bimanual suspensory behavior: morphology, selective advantages and phylogeny.
In: Lesser Apes: Evolutionary and Behavioral Biology, pp. 85–95 (1984)
25. Bertram, J.E.A., Chang, Y.: Mechanical energy oscillations of two brachiation gaits: measure-
ment and simulation. Am. J. Phys. Anthropol. 115(4), 319–326 (2001)
26. Chang, Y., Bertram, J.E.A., Lee, D.V.: External forces and torques generated by the brachiating
white-handed gibbon (Hylobates lar). Am. J. Phys. Anthropol. 113(2), 201–216 (2000)
27. Bertram, J.E., Ruina, A., Cannon, C.E., Chang, Y.H., Coleman, M.J.: A point-mass model of
gibbon locomotion. J. Exp. Biol. 202(19), 2609–2617 (1999)
28. Usherwood, J.R., Larson, S.G., Bertram, J.E.A.: Mechanisms of force and power production
in unsteady ricochetal brachiation. Am. J. Phys. Anthropol. 120(4), 364–372 (2003)
29. Michilsens, F., Vereecke, E.E., D’Août, K., Aerts, P.: Functional anatomy of the gibbon fore-
limb: adaptations to a brachiating lifestyle. J. Anat. 215(3), 335–354 (2009)
30. Fukuda, T., Hosokai, H., Kondo, Y.: Brachiation type of mobile robot. In: Fifth International
Conference on Advanced Robotics, 1991. Robots in Unstructured Environments, 91 ICAR,
pp. 915–920 (1991)
31. Saito, F., Fukuda, T., Arai, F.: Swing and locomotion control for a two-link brachiation robot.
IEEE Control Syst. 14(1), 5–12 (1994)
32. Fukuda, T., Saito, F.: Motion control of a brachiation robot. Robot. Auton. Syst. 18(1–2), 83–93
(1996)
33. Nakanishi, J., Fukuda, T., Koditschek, D.E.: A brachiating robot controller. IEEE Trans. Robot.
Autom. 16(2), 109–123 (2000)
34. Yamazaki, N., Ishida, H.: A biomechanical study of vertical climbing and bipedal walking in
gibbons. J. Hum. Evol. 13(7), 563–571 (1984)
35. Isler, K.: Characteristics of vertical climbing in gibbons. Evol. Anthropol. Issues News Rev.
11(S1), 49–52 (2002)
36. Isler, K.: 3D-kinematics of vertical climbing in hominoids. Am. J. Phys. Anthropol. 126(1),
66–81 (2005)
37. Nakano, Y., Hirasaki, E., Kumakura, H.: Patterns of vertical climbing in primates. In: Human
Origins and Environmental Backgrounds, pp. 97–104. Springer (2006)
38. Zimmermann, K.A.: Muscular System: Facts, Functions & Diseases (2011). http://www.
livescience.com/26854-muscular-system-facts-functions-diseases.html
39. Maita, D., Venture, G.: Influence of the model’s degree of freedom on human body dynamics
identification. In: 2013 35th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society (EMBC), pp. 4609–4612 (2013)
40. Zihlman, A.L., McFarland, R.K.: Body mass in lowland gorillas: a quantitative analysis. Am.
J. Phys. Anthropol. 113(1), 61–78 (2000)
41. Burnham, D.A., Feduccia, A., Martin, L.D., Falk, A.R.: Tree climbing–a fundamental avian
adaptation. J. Syst. Palaeontol. 9(1), 103–107 (2011)
42. Norberg, R.Å.: Treecreeper climbing; mechanics, energetics, and structural adaptations. Ornis
Scand. 191–209 (1986)
43. Norberg, R.Å.: Why foraging birds in trees should climb and hop upwards rather than down-
wards. Ibis (Lond. 1859) 123(3), 281–288 (1981)
44. Jusufi, A., Kawano, D.T., Libby, T., Full, R.J.: Righting and turning in mid-air using appendage
inertia: reptile tails, analytical models and bio-inspired robots. Bioinspir. Biomim. 5(4), 45001
(2010)
45. McCauley, B.: Bird Skeletons (2016). http://brianmccauley.net/bio-6a/bio-6a-lab/chordates/
bird-skeletons
46. Waldron, K.J., Hubert, C.: Scaling of robotic mechanisms. In: IEEE International Conference
on Robotics and Automation, 2000, ICRA’00, Proceedings, vol. 1, pp. 40–45 (2000)
47. Macaw, T.: Macaw climbing back to cage (2013). https://youtu.be/BGYuX2eWx6o
48. Rakholiya, K.: Climbing parrot (2016). https://www.youtube.com/watch?v=Tgu0_eLQHNw
Chapter 4
A Service Baxter Robot in an Office
Environment
Gu Fang and Brendan Cook
Abstract Service robotics in an office environment has been an active research

topic in recent years. This is driven by the need to increase productivity and improve
worker safety. In this paper, a Baxter robot on an omnidirectional mobility base is
used as a service robot to navigate in an office environment. In particular, the robot
is able to plan its path according to a pre-existing map and avoid obstacles during
operation. This robotic system is also able to access the elevator system by using
the manipulator on Baxter to press elevator buttons just as a human user will do.
Experimental results show that the developed system is able to navigate in office
corridors and call and enter an elevator without modification to the existing office
environment.
Keywords Service robot · Baxter robot · Elevator access · Robot in office

environment
4.1 Introduction
Robot manipulators have played an important role in enhancing productivity and

in improving safety in industrial settings. To achieve similar goals in the domestic
environment, in recent years, the service robot is becoming a very active research
field [1]. Research in this area focuses on different aspects associated with mobile
robots, such as navigation in an office type of environment with human around [2],
obtain 3D mapping of an office environment [3]. Research has also been conducted
into the human-robot interaction in this context [4]. As noted in [2], the service robot
is required to address the same challenges as usual mobile robots.
In addition, service robots are required to operate in the environment that is mainly
designed for human users. For example, it will be desirable for a service robot to
access the elevator system as a normal human user [5]. To achieve this type of
G. Fang (B) · B. Cook

School of Computing, Engineering and Mathematics, Western Sydney University, Penrith, NSW
2751, Australia

48 G. Fang and B. Cook
objective and for the service robot to be able to manipulate objects as required, it
should have some manipulating capacities. Many different types of robots that have
both mobile base and manipulators have been introduced in the past few years. For
example, the Asimo [6] is a humanoid robot with two legs and two arms. On the other
hand, PR2 by Willow Garage is a wheeled mobile robot with two manipulators (arms)
[7]. While these robots are attractive in their appearances and functionalities, their
prices are prohibitively high. The Baxter robot [8] introduced by Rethink Robotics
on a mobile base, on the other hand, is an affordable alternative to Asimo and PR2. In
addition, the software environment used in Baxter, Robot Operating System (ROS)
[9], is the same as that used in PR2. Therefore, many existing algorithms can be
readily incorporated into the Baxter robot system to test new developments.
In this paper, a Baxter robot on an omnidirectional mobile base from Dataspeed
[10] is to be used as a service robot. A platform is developed to integrate existing
methods in path planning and obstacle avoidance. A modified localisation algorithm
is introduced for the sensors and software used. The developed system is also able
to use its manipulator to interact with a standard elevator button panel for the Baxter
robot to gain access into the elevator. This application of a Baxter robot is unique
and to our best knowledge it is the only such system being introduced.
The paper is organised into five sections. In the next section, the software envi-
ronment for the project together with methods used to achieve path planning, path
tracking and obstacle avoidance are briefly reviewed. Also introduced in Sect. 4.2 is
the image processing method used for the robot to interact with the elevator button
to gain access. Some results are shown in Sect. 4.3 and conclusions are given in
Sect. 4.4.
4.2 Methods
4.2.1 Software Framework
As mentioned above the Baxter robot and the mobile base are all operated using the
ROS framework. Since MATLAB has a Robotic System Toolbox (RST) [11] which
we are familiar with, it was decided that the implementation of path planning, path
tracking, obstacle avoidance, and image processing are all carried out in MATLAB.
The information obtained in MATLAB is then passed onto the robot using network
communications with the ROS. The ROS is responsible for interacting with the robot
to obtain sensory data and pass on these data to MATLAB for processing if required.
ROS is also responsible for controlling the robot operation using the data obtained
from the MATLAB RST. Due to the missing odometry software required for the
path tracking in RST, the open-source ROS package Robot Localization [12] was
implemented on Dataspeed Mobile Base. The overall software framework can be
seen in Fig. 4.1.
4 A Service Baxter Robot in an Office Environment 49
MATLAB Robot Operating System Baxter robot

LAN +
(Robotic System Toolbox) (ROS) Dataspeed
Wi-Fi Mobile base
Path Planning (PRM) Mobile base odometry
Path Tracking (PPC) Baxter joint control
Obstacle avoidance (VFH) Baxter camera operation
Localisation (AMCL) Laser scanner operation USB
Image processing Hokuyo Laser
Fig. 4.1 The overall software framework
4.2.2 Robot Control
The service robot system developed is able to autonomously navigate through an

office area with a known map from an arbitrary starting position to a chosen des-
tination. The robot is also required to avoid obstacles, both stationary and moving,
during its motion. To achieve this requirement, the following control strategies are
chosen.
4.2.2.1 Path Planning
There are many different path planning methods available that can plan paths for
a robot to follow provided a pre-existing map is available [13]. In MATLAB RST,
the probabilistic roadmap (PRM) [14] is used as a path planning method. It requires
the input of a binary occupancy grid map of the office. For this purpose, a map of
the office environment in one of the buildings in our university is generated for this
purpose. Using the PRM method in RST, a path can be generated.
4.2.2.2 Path Tracking and Obstacle Avoidance
To control the robot to move along the pre-planned path from PRM, the pure pursuit
controller (PPC) [15] is adopted as it is an available option in RST. The PPC is
designed to follow a point on the planned path. The distance between this point and
the robot is usually dependent on the path conditions such as tight corners and is
coined the “look-ahead distance”. The controller uses this point to geometrically
determine the curvature that will direct the robot towards it. For front wheel steered
vehicles this is ideal, while for omnidirectional or differential drives the curvature
is not necessary. Both these types of drives can turn on a single point, so while the
curvature might not be necessary the method still produces the heading angle for the
robot to follow the path.
As the PPC is only using the way points generated from PRM off-line, to control
the robot to avoid obstacles during the run time, the Vector Field Histogram (VFH)
[16] is used. The VFH uses the range sensor to construct a polar histogram to show
obstacles. It is similar to the potential field method [17].
4.2.2.3 Multi-rate Localisation
For the controller to work properly, the location of the robot needs to be estimated
accurately. Clearly, using dead reckoning alone will not achieve sufficient accuracy
over a meaningful operation time. Since RST has an Adaptive Monte Carlo Local-
isation (AMCL) [18] method built in, it is reasonable to use this as the localisation
method.
However, during the implementation of the robot system, it was found that due
to the slow processing of the MATLAB program, the localisation method cannot
provide the velocity commands to the robot fast enough for it to follow the path. To
avoid the implementation of localisation in ROS, a ‘multi-rate’ localisation method
is devised. In this method, the AMCL is only used to update the location of the robot
when the robot has moved a significant amount of distance or heading. When the
robot location is not updated using the AMCL, dead reckoning is used for updating
the location information. This approach can be shown in the flowchart in Fig. 4.2.
After this implementation, the localisation can be at a frequency that is adequate for
the robot to follow a path smoothly.
Fig. 4.2 Multi-rate

localization
4.2.3 Elevator Button Recognition and Manipulation
For the service robot to operate in an office building environment, gaining access to
the elevator system is an inevitable requirement. In this paper, this process is divided
into two parts, i.e., elevator button recognition and manipulation; and elevator access.
The elevator operation panel in the building can be seen in Fig. 4.3. For the robot
to recognise and operate the button, the robot was moved to a position that is close
enough to the button panel to allow the robot to acquire the image of the panel and
to be able to touch the button.
Once the robot is in position, the image processing allows the Baxter to get an
accurate position on the elevator button. It isolates the black ring around the button
and calculates the coordinates on the centroid in the frame. Baxter uses this to know
what position to move the limb to. The height of the button is constant relative to the
Baxter’s frame, so Baxter is able to move its limb to a predefined position and only
adjust to the horizontal position.
The image is first received as a ROS colour image type and then passed onto
MATLAB, converted to an unassigned 8-bit integer matrix and converted again into
a grayscale image for easier image processing. As the location height of the button
is fixed, the rough location of the button in the image can be approximated. This
information allows the image to be cropped to isolate the region of the button. After
the cropping, a threshold is performed to convert the image to only show the ring of
the button in white:

1 Ix,y < T
f x,y =
0 Ix,y ≥ T
Fig. 4.3 The elevator

operation panel
where f x,y , and I x,y are the resulting binary image and the pixel intensity of the original
image at location (x, y) respectively, and T is the threshold intensity used.
After the threshold, the ring is filled with white colour for the calculation of the
centroid location of the button (x c , yc ) in both horizontal and vertical directions. The
distance, d, from the Baxter hand to the button can be obtained using the IR sensor
on Baxter’s limb. Using both the calculated x c and d the movement command for
the Baxter can be issued. A built in Inverse Kinematics service on Baxter is used
to calculate the Baxter’s joint angles based on the given Cartesian coordinate of the
endpoint and orientation.
4.2.4 Elevator Access
For the part, the robot is required to move to the right position, waiting for the elevator
door to open, and then enter the elevator.
At the first stage, the robot is required to move back from the “button pressing
position” (X) to a pre-defined waiting position (O). It is shown in Fig. 4.4 that the
robot is required to move from X to O.
To detect the door opening, the laser scanner’s distance measurement is used. If
a large enough increase of distance is observed while waiting in position (O), then
the door is deemed to be open. It is also possible, if the lift has a mirror, to use
image processing to find the robot image in the elevator mirror to detect both the
door opening and the lift occupancy status.
To move into the elevator, a simple forward velocity command is given since the
robot is already in position to enter the elevator. To take into account any errors in
locating directly in front of the elevator and possible human traffic, the VFH is used
to adjust the robot position in the X axis as shown in Fig. 4.4.
Fig. 4.4 Translation of the

mobility base into the
elevator waiting position
V
x
4.3 Results
4.3.1 System Setup and the Operating Environment
To test the developed system, a Baxter robot on a Dataspeed mobile base is used
as the service robot (Fig. 4.5). The robot has a Hokuyo URG-04LX laser range
finder mounted at the waist-height of the robot. A binary grid map of the operating
environment is shown in Fig. 4.6. This map has been dilated to take into account the
robot’s radius.
Fig. 4.5 Baxter and mobility base
Fig. 4.6 Dilated binary grid map of the building floor including the robot radius
4.3.2 Experimental Results
4.3.2.1 Robot Control
The planned path using PRM can be seen in Fig. 4.7. Figure 4.8a shows the robot in
motion while Fig. 4.8b shows the location results using AMCL after 10 iterations.
During the motion, the AMCL is updated based on the robot’s movement distance
and Z rotation. VFH is implemented through the motion to guide the robot to avoid
obstacles.
4.3.2.2 Elevator Access
In Fig. 4.9a, the robot is shown to be in a position to manipulate the elevator button.
Figure 4.3 above shows the Baxter arm is a position to acquire the image and press the
button. The image obtained by the camera on the Baxter hand is shown in Fig. 4.9b.
The cropped image together with the processed images are shown in Fig. 4.10. The
status of the robot entering the elevator after the door opens is shown in Fig. 4.11.
Fig. 4.7 PRM generated path in green colour
Fig. 4.8 a The service robot in motion. b The AMCL localisation after 10 iterations
Fig. 4.9 a The robot is at a position to manipulate the elevator button. b The image obtained by
the robot
Fig. 4.10 a The cropped image. b Thresholded image. c Filled image
Fig. 4.11 The robot

entering into the elevator
4.4 Conclusion
In this paper, a service robot system is developed by using a Baxter robot on a

Dataspeed omnidirectional mobile base. The system uses a variety of software frame-
works under a networked structure. While the ROS framework is used to communi-
cate between the different subsystems and control the hardware such as the Baxter,
its mobile base and various sensors, MATLAB is used to perform the control and
processing activities such as path planning, path tracking, obstacle avoidance, local-
isation, and image processing. The developed system is able to achieve a good out-
come by adopting a multi-rate localisation that updates the AMCL results only when
required. This has allowed the robot to be able to operate even while the MATLAB
is running on a slow computer. Image processing is also introduced together with

manipulation method to allow for the robot to press the elevator button as a normal
human user. Test results show that the developed system is able to move in a normal
office environment and gain access into an elevator. All the source codes for MAT-
LAB is available at https://github.com/runbren/Baxter-Delivery-Service-Robot.
Future work will include the navigation to press buttons inside the lift. In addition,
the interaction with human users for the elevator is also not considered in this work.
References
1. Pineda, L.A., Rodriguez, A., Fuentes, G., Rascon, C., Meza, V.: Concept and functional struc-
ture of a service robot. Int. J. Adv. Robot. Syst. 12(6), 1–15 (2015)
2. Araujo, A.R., Caminhas, D.D., Pereira, G.A.: An architecture for navigation of service robots
in human-populated office-like environments. IFAC-PapersOnLine 48–19(2015), 189–194
(2015)
3. Tamas, L., Goron, L.C.: 3D semantic interpretation for robot perception inside office environ-
ments. Eng. Appl. Artif. Intell. 32, 76–87 (2014)
4. Brandl, C., Mertens, A., Schlick, C.M.: Human-robot interaction in assisted personal services:
factors influencing distances that humans will accept between themselves and an approaching
service robot. Hum. Factors Man. 26, 713–727 (2016)
5. Kang, J.G., An, S.Y., Oh, S.Y.: Navigation strategy for the service robot in the elevator environ-
ment. In: International conference on control, automation and systems 2007, 17–20 Oct 2007,
Seoul, Korea, pp. 1092–1097 (2007)
6. Sakagami, Y., Watanabe, R., Aoyama, C., Matsunaga, S., Higaki, N., Fujimura, K.: The intel-
ligent ASIMO: system overview and integration. In: IEEE/RSJ International Conference on
Intelligent Robots and Systems, 2002, vol. 3, pp. 2478–2483 (2002)
7. Cousins, S.: ROS on the PR2 [ROS Topics]. IEEE Robot. Autom. Mag. 17(3), 23–25 (2010)
8. Guizzo, E., Ackerman, E.: How rethink robotics built its new baxter robot worker. IEEE Spectr.
18 (2012)
9. Quigley, M., Gerkey, B., Smart, W.D.: Programming Robots with ROS A Practical Introduction
to the Robot Operating System. O’Reilly Media (2015)
10. Dataspeed: Mobility base repository, cited on 3 July 2017. Available from https://bitbucket.
org/account/user/DataspeedInc/projects/MB (2017)
11. MathWorks: Design and test algorithms for robotics applications, cited on 5 July 2017. Avail-
able from https://au.mathworks.com/products/robotics.html (2017)
12. Mac, T.T., Copot, C., Tran, D.T., De Keyser, R.: Heuristic approaches in robot path planning:
A survey. Robot. Auton. Syst. 86, 13–28 (2016)
13. ROS wiki: Robot Localization, ROS, cited on 20 July 2017. Available from http://wiki.ros.org/
robot_localization (2017)
14. Kavraki, L.E., Svestka, P., Latombe, J.-C., Overmars, M.H.: Probabilistic roadmaps for path
planning in high-dimensional configuration spaces. IEEE Trans. Robot. Autom. 12(4), 566–580
(1996)
15. Coulter, R.: Implementation of the Pure Pursuit Path Tracking Algorithm. Carnegie Mellon
University, Pittsburgh, Pennsylvania (1990)
16. Ulrich, I., Borenstein, J.: VFH: reliable obstacle avoidance for fast mobile robots. In: Proceed-
ings. 1998 IEEE International Conference on Robotics and Automation, 1572–1577 (1998)
17. Dinham, M., Fang, G.: Time optimal path planning for mobile robots in dynamic environments.
In: International Conference on Mechatronics and Automation, pp. 2132–2137 (2007)
18. Thrun, S., Burgard, W., Fox, D.: Probabilistic Robotics. MIT Press, Cambridge, MA (2005)
Chapter 5
The Design of a New Rotary Hexapod
with a Single Active Degree of Freedom
A. S. Fomin, V. A. Glazunov and J. K. Paik
Abstract This study provides a novel approach for designing a rotary hexapod
with a single actuator and definable motion of a platform. Hexapod is a parallel
structure mechanism with six legs and six degrees of freedom (DoF) of an end-
effector. It is not an under actuated system as it cannot completely control a motion
of a platform by less than six actuators. In the proposed study we developed a new
under actuated hexapod driven by planar linkage with singular DoF. This novel design
was reached by combining a zero-DoF structural group including six parallel chains
with a platform and planar linkage mechanism with mobility one set by a movable
base. We placed six parallel chains on the base by means of carriages coupled to the
movable levers of the planar linkage. The final system is actuated by single rotation
of a driving link set in the base.
Keywords Hexapod · Degree of freedom · Parallel mechanism · Linkage ·

Rotary hexapod
5.1 Introduction
Parallel structure manipulators are effective in industrial operations requiring high

precision, speedwork, stiffness, complete range of motions and other significant
technological criteria for efficient functioning [1, 2]. Yet, for the most part, parallel
manipulators have restricted opportunities for complete rotation of 360° of end-
A. S. Fomin (B)
Institute of Mechanical Engineering and Transport Siberian State Industrial University,
Novokuznetsk 654007, Russia
A. S. Fomin · V. A. Glazunov
Department of Mechanics and Control of Machines, Mechanical Engineering Research Institute
of the Russian Academy of Sciences, Moscow 101990, Russia
A. S. Fomin · J. K. Paik
Reconfigurable Robotics Lab (RRL), Swiss Federal Institute of Technology (EPFL) Lausanne,
1015 Lausanne, Switzerland

58 A. S. Fomin et al.
effectors around the vertical axis. Indeed, for some applications involving design of
medical robots, automotive platforms or for additive manufacturing, this property
becomes critically relevant. In [3, 4] authors presented a parallel manipulator having
complete rotation for application in car or plane simulators. Complete rotation of
a platform relative to the base is realized using a circular rail for the ground path.
Figure 5.1 presents a three-dimensional model of the first rotary manipulator [5]
developed by Belikov et al. in 1983 which serves as a foundation for many hexapod
models with circular guide.
The mobility which predetermines number of actuators for the whole system can
be calculated by the formula developed by A.P Malyshev for spatial kinematic chains
W = 6n − 5p5 − 4p4 − 3p3 − 2p2 − p1 , (5.1)
where W is mobility of a kinematic chain, defining number of its DoF, n is the number
of movable links in a kinematic chain, p5 , p4 , p3 , p2 and p1 are numbers of one-, two-,
three-, four- and five-DoF kinematic pairs. According to (5.1) the manipulator with
n = 16, p5 = 12 and p3 = 9 has mobility W = 9. Such mobility includes: (1) six
active DoF from six carriages moving on a circular rail set on a fixed link; (2) three
passive DoF from three links coupled to the platform—each of these three links
can have an independent rotation because spherical joint and two coupled rotational
joints have mutual axis. The platform in this manipulator allows movement in the
six DoF with three prismatic movements (lateral, longitudinal and vertical) and three
rotations (pitch, roll and yaw). Structural and kinematic analysis of this manipulator
was studied in detail by Janabi-Sharifi and Shchokin [6, 7].
The complete rotation of a platform is also reachable using rotopod variations
which are defined as systems that include circular rails around which six legs are
orienting the platform. These variations are investigated by Bonev et al. [8], Herrera
Fig. 5.1 Manipulator

developed by Belikov et al.
[5]
5 The Design of a New Rotary Hexapod with a Single Active Degree … 59
Fig. 5.2 Rotary hexapod

with two rails
et al. [9] and now manufactured in accordance with a US Patent [10] by the Hexel
Corporation, currently Mikrolar, USA.
The manipulator shown in Fig. 5.1 was further developed into a model without
passive DoF by Coulombe and Bonev [11] and is shown as a three-dimensional
model in Fig. 5.2. This manipulator includes two rails instead of the single rail used
in rotopod models or the original manipulator shown in Fig. 5.1. A second rail brings
compactness to the manipulator and increases its workspace. Further, rotational joints
replace the double spherical joints present in rotopod. This novel design solution led
to an increased workspace maximum vertical translation and high accuracy of this
parallel manipulator. The input motions are given for each of six carriages.
However for some applications requiring predetermined motion of a platform, six
actuators are not obligatory to have. Yet parallel structure of a designed mechanism
would allow supporting and replacing heavy objects. In this regard we aimed to
develop a rotary hexapod system with a single actuator, without passive mobility and
with six predetermined DoF of the platform in this study.
5.2 Structural Analysis of a Rotary Hexapod with Two

Rails
A rotary hexapod with two rails includes a structural group of links having zero DoF
set on six carriages that serve as driving links. The group is shown separately in
Fig. 5.3. The group of links is connected above by six rotational joints. The group
includes a platform and three paired parallel chains. Each contains 8 links connected
Fig. 5.3 Structural group

with parallel chains having
zero DoF
by 10 one-DoF kinematic pairs. The overall number of links for a group is 25 and
number of kinematic pairs is 30.
The overall mobility for the whole group can be calculated by (5.1). For the
structural group shown in Fig. 5.3 with n = 25 and p5 = 30, the mobility equals W = 0.
This result indicates it as a zero-DoF chain also known as an Assur group after Assur
[12]. Therefore if we add six internal DoF to this group of links, we get an overall
mobility equals W = 6. This solution explains rotary hexapod of Fig. 5.2. A mobility
W = 6 can also be derived from formula (5.1) when n = 31 and p5 = 36, which
proves the necessity of driving six carriages along a rail providing three rotations
and three translations of the end-effector (platform).
5.3 Mechanical Design of a Rotary Hexapod with a Single

Active DoF
To provide single actuation to the rotary hexapod (Fig. 5.2), it needs to connect a
zero-DoF group (Fig. 5.3) and a single-DoF planar linkage located at the base. This
connection design is presented in Fig. 5.4. It includes only one-DoF joints and links
moving within a circular rail. In this mechanism driving link 1 is a double-crank that
is coupled by a hinge in the center of the circular rail as well as diametrically by
means of the carriages moving rotationally around the central hinge. Driving link 1 is
connected with two-bar Assur groups (diads) RRR 2-3, 4-5, 6-7, 8-9, 10-11, 12-13,
14-15, 16-17 investigated previously in [13, 14]. The overall mobility of the ground
can be calculated in accordance with the known Chebishev formula [15] given as
W = 3n − 2 p5 − p4 , (5.2)
Fig. 5.4 Single-DoF base

mechanism with six
carriages moving on a
circular rail
Fig. 5.5 Rotary hexapod

with a single active DoF
where W is mobility of a mechanism, n is the number of movable links, p5 and p4

are numbers of one- and two-DoF joints. For the base mechanism we have n = 17
and p5 = 25, therefore the mobility W = 1.
Obviously the integration of a zero-DoF group (Fig. 5.3) and one-DoF base linkage
(Fig. 5.4) will result in a parallel mechanism with the novel aspect of having a
singular DoF that can be actuated by a single drive. Such an approach allows space
to be reduced for the actuation system and setting a single crank to translate further
motion through all links to the end-effector. This novel manipulator is presented in
Fig. 5.5. The platform still has six DoF as for the manipulators shown in Figs. 5.1 and
5.2; however the number of drives is now maximally reduced by one. This design
also allows us to set the second circular rail inside the first.
5.4 Direct Kinematics for the Base Mechanism
Kinematic analysis of a rotary hexapod with six active DoF has been carried out
by Bonev and Gosselin [16, 17]. For the rotary hexapod shown in Fig. 5.5, only a
single active DoF exists, therefore translations of carriages moving on the circular rail
depend on the rotation of the driving link. To know the positions of the carriages we
will analyze the base mechanism separately. As it is planar and all links are moving
in the plane, graph-analytical method of kinematics is suggested, which is quite
simple and illustrative. It demonstrates the driving directions of all links and allows
calculation of values of velocities and accelerations by making vector equations. Part
of the unknown parameters will be analytically found from kinematic equations of
planar movement.
Define velocities of points B1 , F 9 , G13 , K 1 , O17 and U 5 , which belong to each of
the six carriages (Fig. 5.6a). Accept angular velocity of the driving link w1 = 0.5 s−1
and lengths of links according to Table 5.1.
Velocity of point B1 is equal to V B1 = 0.25 m/s. Velocities of other links in point
B are equal to each other V B1 = V B2 = V B6 = V B . The scale of the velocity plan
(µV ) is calculated as
Fig. 5.6 Toward kinematic analysis of the single-DoF base mechanism: a top view of the base
mechanism; b velocity vector diagram
Table 5.1 Lengths of links Link Length (m)

of the ground mechanism
AB = AK = AO = AU = AF = AG 0.50
BC = BS = KL = KJ 0.36
CD = RS = LM = IJ 0.07
DE = QR = MN = HI 0.14
EF = QU = NO = GH 0.18
VB 0.25m/s
µv = = = 0.001 (m/s)/mm. (5.3)
pb 250mm
Figure 5.6a indicates the chosen position of the mechanism for the velocity plan
drawing. Velocity of point F can be found from velocities of points C and E. Equations
for them are written below

V̄C = V̄B + V̄C B ,V̄C B ⊥BC;
V̄C = V̄D + V̄C D ,V̄C D ⊥C D;
DE
VE = VC · . (5.4)
CD
When V C and V E are known, we can find V F as

V̄F = V̄E + V̄F E ,V̄F E ⊥E F
(5.5)
V̄F = V̄ A + V̄F A ,V̄F A ⊥F A.
Velocity of point K equals velocity of point B as AB = AK, but these velocities are in
opposition to each other. To find the velocity of point G, we need to know velocities
of points J and H. They can be found from the equations shown below

V̄ J = V̄K + V̄ J K ,V̄ J K ⊥J K ;
V̄ J = V̄I + V̄ J I ,V̄ J I ⊥I J ;
IH
VH = V J · . (5.6)
IJ
Velocity V G is then calculated as

V̄G = V̄H + V̄G H ,V̄G H ⊥G H
(5.7)
V̄G = V̄ A + V̄G A ,V̄G A ⊥G A.
Analogically we can find velocities of points O and U. For point O we need to know
velocities V L and V N , for point P—velocities V S and V Q . Remaining equations for
these velocities are listed below

V̄L = V̄K + V̄L K ,V̄L K ⊥K L;
V̄L = V̄M + V̄L M ,V̄L M ⊥L M;
MN
V N = VL · . (5.8)
LM

V̄O = V̄N + V̄O N ,V̄O N ⊥N O;
(5.9)
V̄O = V̄ A + V̄O A ,V̄O A ⊥AO.

V̄S = V̄B + V̄S B ,V̄S B ⊥S B;
V̄S = V̄R + V̄S R ,V̄S R ⊥RS;
Table 5.2 Velocities of six Velocity of a point Value (m/s)

carriages moving on a
circular rail V B (carriage 1B) 0.2500
V F (carriage 9) 0.2232
V G (carriage 13) 0.1620
V K (carriage 1K) 0.2500
V O (carriage 17) 0.1620
V U (carriage 5) 0.1620
QR
V Q = VS · . (5.10)
RS

V̄U = V̄Q + V̄U Q ,V̄U Q ⊥U Q
(5.11)
V̄U = V̄ A + V̄U A ,V̄U A ⊥AU.
A velocity vector diagram (Fig. 5.6b) is drawn according to the Eqs. (5.3)–(5.11) and
positions of carriages shown in Fig. 5.6a. To find numerical values of the velocities
we need to multiply the scale of the plan calculated in Eq. (5.4) on lengths of vectors
from the plan. Velocities of points A, D, I, M and R are equaled to zero and are
combined with point p on the velocity diagram. The velocities of the carriages are
indicated in Table 5.2 according to the scale of velocity diagram.
An acceleration vector diagram can be analogically drawn using analytical
equations and the same algorithm shown for velocity calculations based on
Eqs. (5.3)–(5.11).
5.5 Conclusions
The presented study shows a design method for reducing actuating link numbers
in parallel structure manipulators. The method is based on the separation of the
structural group having zero DoF and then its connection to a single-DoF linkage
mechanism used as a movable base. Edge kinematic pairs of parallel chains are
connected to the movable links of the base linkage providing single actuation of the
whole mechanism, while the end effector still has six DoF. The proposed method
has been demonstrated on a rotary hexapod where actuated links were reduced from
six to one. Parallel chains were set above movable carriages that serve as sliders
on a planar base linkage. The proposed idea is viable for other types of parallel
mechanisms having different numbers of parallel chains and actuators.
Acknowledgements The study has been carried out with the support of the Russian President
Grant according to the research project MK-2781.2019.8.
References
1. Kong, X., Gosselin, C.M.: Type Synthesis of Parallel Mechanisms, p. 272. Springer, Berlin
Heidelberg (2007)
2. Glazunov, V.A., Yu, C.A.: Development of mechanisms of parallel structure. J. Mach. Manuf.
Reliab. 3, 37–43 (2014)
3. Aleshin, A.K., Glazunov, V.A., Shai, O., Skvortsov, S.A., Lastochkin, A.B.: Infinitesimal dis-
placement analysis of a parallel manipulator with circular guide via the differentiation of
constraint equations. J. Mach. Manuf. Reliab. 45(5), 398–402 (2016)
4. Rashoyan, G.V., Lastochkin, A.B., Glazunov, V.A.: Kinematic analysis of a spatial parallel
structure mechanism with a circular guide. J. Mach. Manuf. Reliab. 44(7), 54–60 (2015)
5. Belikov, V.T., Vlasov, N.A., Zablonski, K.I., Koritin, A.M., Shchokin, B.M.: Manipulator,
USSR Patent No. SU 1049244 A (1983)
6. Janabi-Sharifi, F., Shchokin, B.: A rotary parallel manipulator: modelling and workspace anal-
ysis. In: Proceedings of the IEEE International Conference on Robotics and Automation, Now
Orleans, LA, pp. 3671–3677 (2004)
7. Shchokin, B., Janabi-Sharifi, F.: Design and kinematic analysis of a rotary positioner. Robotica
25, 75–85 (2007)
8. Bonev, I.A., Yu, A., Zsombor-Murray, P.: XY-theta positioning table with parallel kinematics
and unlimited theta rotation. In: Proceedings of the IEEE ISIE, Montreal, Québec, Canada,
pp. 3113–3117 (2006)
9. Herrera, F.J., Acuna, H.G., Lengerke, O., Dutra, M.S.: Conceptual design and kinematic anal-
ysis of a parallel rotational robot. In: Proceedings of ICIRA, vol. III, pp. 52–61 (2012)
10. Hexel Corporation, US: Systems and methods employing a rotary track for machining and
manufacturing. US Patent No. 6196081 (2001)
11. Coulombe, J., Bonev, I.A.: A new rotary hexapod for micropositioning. In: Proceedings of
the IEEE International Conference on Robotics and Automation (ICRA) Karlsruhe, Germany,
pp. 877–880 (2013)
12. Assur, L.V.: Investigation of planar mechanisms with lower parameters in terms of their struc-
ture and classification, 529 p. Publishing House of the Academy of Sciences of the USSR,
Moscow (1952)
13. Briot, S., Arakelian, V.: Complete shaking force and shaking moment balancing of in-line
four-bar linkages by adding a class-two RRR or RRP Assur group. Mech. Mach. Theory 57,
13–26 (2012)
14. Li, S., Dai, J.S.: Structure synthesis of single-driven metamorphic mechanisms based on the
augmented Assur groups. J. Mech. Robot. 4, 8 p (2012)
15. Dvornikov, L.T., Fomin, A.S.: Development of the basic conditions for division of mechanisms
into subfamilies. Procedia Eng. 150, 882–888 (2016)
16. Bonev, I.A., Gosselin, C.M.: Geometric algorithms for the computation of the constant-
orientation workspace and singularity surfaces of a special 6-RUS parallel manipulator. Pro-
ceedings of DETC’02, Montreal, QC, Canada, pp. 1–10 (2002)
17. Bonev, I.A., Gosselin, C.M.: A geometric algorithm for the computation of the constant-
orientation workspace of 6-RUS Parallel Manipulators. Proceedings of DETC’00, Baltimore,
MD, USA, pp. 1–10 (2000)
Chapter 6
Modelling and Model-Based Control
of a Four-Bar Mechanism with a Flexible
Coupler Link
Nguyen Van Khang, Nguyen Sy Nam and Nguyen Phong Dien
Abstract In the mechanisms and machines operating at high speeds, the elastic
vibration of links is inevitable. The present paper deals with the problem of modelling
and model-based controller design for a flexible four-bar mechanism. The fully cou-
pled non-linear equations of motion are derived using the Lagrange’s equations with
multipliers for constrained multibody systems. The resulting differential-algebraic
equations are solved using numerical methods. Based on the obtained modelling
results, a simple PD controller is designed to minimize the influence of elastic link
on the desired motion.
Keywords Steady-state vibration · Elastic mechanism · Lagrange’s equations ·

Vibration control
6.1 Introduction
The planar four-bar mechanism is one of the most widely used mechanisms in trans-
mission systems of reciprocating machines. It is well-known that the large base
motion of a lightweight high-speed mechanism leads to small relative vibrations of
flexible links such as the coupler [1]. Besides, small vibrations of flexible coupler can
cause transmission errors and noise [2]. Starting from the research results in dynamic
analysis of rigid multibody systems, e.g. [3, 4], a number of approaches and meth-
ods for dynamic analysis and control of elastic mechanisms has been proposed and
applied since the 90s of the last century [5–16]. Among them, steady-state vibrations
and vibration control of a planar four-bar mechanism with the flexible coupler were
the objective of some studies, e.g. [7, 9, 12]. Some methods of dynamic modelling
and vibration suppression for parallel manipulators with flexible links are presented
in [10, 11, 13], using continuum models or finite element models.
N. Van Khang (B) · N. P. Dien

Hanoi University of Science and Technology, 1. Dai Co Viet Road, Hanoi, Vietnam
N. S. Nam
National University of Civil Engineering, 55. Giai Phong Road, Hanoi, Vietnam
68 N. Van Khang et al.
In this study, a planar revolute-jointed 4-link mechanism with the flexible coupler
link is considered to investigate the problem of modelling and model-based controller
design for vibration suppression. The fully coupled non-linear equations of motion
are derived using the Lagrange’s equations with multipliers for constrained multi-
body systems. The resulting differential-algebraic equations are then solved using
numerical methods. Based on the obtained modelling results, a simple PD controller
is designed to minimize the influence of the coupler link on the desired motion of
the mechanism.
6.2 Equations of Motion of the Four-Bar Mechanism

with a Flexible Coupler Link
We consider now a planar revolute-jointed four-link mechanism with an elastic cou-

pler link as depicted in Fig. 6.1. The lengths of crank OA, coupler link AB and output
link BC are l 1 , l 2 and l 3 respectively, and distance OC is l0 . In addition, τ denotes the
external torque acting on the crank, Ox 0 y0 the fixed coordinate system and Axy the
link-fixed coordinate system. The position of mechanism in Ox 0 y0 can be defined by
rotation angles ϕ 1 , ϕ 2 and ϕ 3 as shown in the figure.
The position vector of point M on coupler link AB is given by
r M = r A + (x + u(x, t))e1 + w(x, t)e2 (6.1)
where rA is the position vector of point A, e1 and e2 are the unit vectors of the coordi-
nate system Axy. Variables u(x,t) and w(x,t) denote the longitudinal deformation and
the transverse deformation of coupler link AB at an arbitrary position x, respectively.
The boundary conditions of deformation are defined by
u(0, t) = 0; u(l2 , t) = u B (6.2)
Fig. 6.1 Diagram of the four-bar mechanism

6 Modelling and Model-Based Control of a Four-Bar Mechanism … 69
w(0, t) = w(l2 , t) = 0 (6.3)
We note that the constraint equations depend on the elastic deformations, namely,
on the longitudinal deformations for this case. From Fig. 6.1 we get the following
constraint equations
f 1 = l1 cos ϕ1 + [l2 + u(l2 , t)] cos ϕ2 − l3 cos ϕ3 − l0 = 0 (6.4a)
f 2 = l1 sin ϕ1 + [l2 + u(l2 , t)] sin ϕ2 − l3 sin ϕ3 = 0 (6.4b)
Upon assumption that elastic deformations of the coupler link are small, the lon-
gitudinal deformation and transverse deformation are independent of each other,
Euler-Bernoulli beam theory can be used for this case. Using Ritz–Galerkin method,
the elastic deformations of the coupler link can be written in the form

N1
w(x, t) = X i (x).qi (t) (6.5)
i=1

N2
u(x, t) = Yk (x). pk (t) (6.6)
k=1
where qi (t), pk (t) are the modal coordinates of transverse deformation and longi-
tudinal deformation, respectively, X i (x), Y k (x) are the mode shapes of transverse
deformation and longitudinal deformation of a pinned-pinned bar which are given
by [15] as

iπ 2k − 1 π x
X i = sin x , Yk (x) = sin (6.7)
l2 2 l2
The kinetic energy of the considered mechanism can be expressed in the form
l2
1 1 1
T = TO A + TBC + T AB = I O ϕ̇12 + IC ϕ̇32 + μṙ2M d x (6.8)
2 2 2
0
where I O and I C are the mass moments of inertia of the input and output links with
respect to the joint axes, respectively, μ denotes the mass per unit length of the
coupler link. Coordinates of point M in the fixed coordinate system are given by:
x M = l1 cos ϕ1 + (x + u) cos ϕ2 − w sin ϕ2 (6.9a)
y M = l1 sin ϕ1 + (x + u) sin ϕ2 + w cos ϕ2 (6.9b)

By differentiating Eqs. (6.9a) and (6.9b), making use of Eqs. (6.5), (6.6), and then
substituting into Eq. (6.8), the kinetic energy is then obtained as
μ
N2
μ
N2 N1 N1
1 1 μl 2 l2
T = I O ϕ̇12 + IC ϕ̇32 + 1 ϕ̇12 + bkl ṗk ṗl + ϕ̇22 m i j qi q j
2 2 2 2 k=1 l=1 2 i=1 j=1
N
μl23 2 μ N2
μ
2 N2 N1 N1
+ ϕ̇ + 2 Fk pk + bkl pk pl ϕ̇2 +
2
m i j q̇i q̇ j
6 2 2 k=1 k=1 l=1
2 i=1 j=1

N2
μl1l22
− μl1 ϕ̇1 sin(ϕ1 − ϕ2 ) Hk ṗk + ϕ̇1 ϕ̇2 cos(ϕ1 − ϕ2 )
k=1
2

N2
N1
+ μl1 ϕ̇1 ϕ̇2 cos(ϕ1 − ϕ2 ) Hk pk + μl1 ϕ̇1 cos(ϕ1 − ϕ2 ) Ci q̇i
k=1 i=1

N1
N1
N2
+ μl1 ϕ̇1 ϕ̇2 sin(ϕ1 − ϕ2 ) Ci qi − μϕ̇2 n ik qi ṗk
i=1 i=1 k=1

N1
N1
N2
+ μϕ̇2 Di q̇i + μϕ̇2 n ik pk q̇i (6.10)
i=1 i=1 k=1
where we use the following notations

l l l l
Ci = 0 2 X i d x; Di = 0 2 x X i d x; m i j = 0 2 X i X j d x; Hk = 0 2 Yk d x; Fk =
l2 l2 l2
0 xYk d x; bkl = 0 Yk Yl d x; n ik = 0 X i Yk d x.
According to [16], it follows that
l2 2 l2 2
1 ∂u 1 ∂ 2w
Π = EA dx + E I dx (6.11)
2 ∂x 2 ∂x2
0 0
where E, I, A are modulus of elasticity, area moment of inertia and cross-sectional

area of the coupler link, respectively. Substitution of Eqs. (6.5), (6.6) into Eq. (6.11)
yields
1 N2 N2
1 N1 N1
Π= EA gkl pk pl + E I k i j qi q j (6.12)
2 k=1 l=1
2 i=1 j=1
l l
where ki j = 0 2 X i X j d x and gkl = 0 2 Yk Yl d x.
Recall that Lagrange’s equations with multipliers for constrained holonomic
multibody systems are defined by [4]
2
d ∂T ∂T ∂Π ∂ fk
− =− − λk + Qj (6.13)
dt ∂ ṡ j ∂s j ∂s j k=1
∂s j
where s j are the generalized coordinates which include rigid body coordinates ϕ 1 ,
ϕ 2 , ϕ 3 as well as elastic modal coordinates qj and pk , f k the constraint equations,
λ1 and λ2 Lagrange’s multipliers. Qj denote the generalized forces. By substituting
Eqs. (6.4a), (6.4b), (6.10), (6.12) into Eq. (6.13), we obtained the governing equations
for the considered mechanism as follows:

μl1l22 N2
IO + μl12 l2 ϕ̈1 + ϕ̈2 cos(ϕ1 − ϕ2 ) + μl1 ϕ̈2 cos(ϕ1 − ϕ2 ) Hk pk
2 k=1

N1
N2
+ μl1 ϕ̈2 sin(ϕ1 − ϕ2 ) Ci qi − μl1 sin(ϕ1 − ϕ2 ) Hk p̈k
i=1 k=1

N1
μl1l22 2
+ μl1 cos(ϕ1 − ϕ2 ) Ci q̈i + ϕ̇ sin(ϕ1 − ϕ2 )
i=1
2 2

N2
+ 2μl1 ϕ̇2 cos(ϕ1 − ϕ2 ) Hk ṗk
k=1

N2
N1
+ μl1 ϕ̇22 sin(ϕ1 − ϕ2 ) Hk pk + 2μl1 ϕ̇2 sin(ϕ1 − ϕ2 ) Ci q̇i
k=1 i=1

N1
− μl1 ϕ̇22 cos(ϕ1 − ϕ2 ) Ci qi = l1 sin ϕ1 λ1 − l1 cos ϕ1 λ2 + τ (6.14)
i=1

l22 N2 N1
cos(ϕ1 − ϕ2 ) + cos(ϕ1 − ϕ2 ) Hk pk μl1 ϕ̈1 + μl1 ϕ̈1 sin(ϕ1 − ϕ2 ) C i qi
2 k=1 i=1
⎛ ⎞ N
l 3 N1 N1 2 N2 N2
+⎝ + 2
m i j qi q j ⎠μϕ̈2 + 2 Fk pk + bkl pk pl μϕ̈2
3 i=1 j=1 k=1 k=1 l=1

N1
N2
N1
N1
N2
N1
N1
−μ n ik qi p̈k + μ Di q̈i + μ n ik q̈i pk + 2μϕ̇2 m i j q̇i q j
i=1 k=1 i=1 i=1 k=1 i=1 j=1
N
2
N2
N2
μl1l22 2
+ 2μϕ̇2 Fk ṗk + bkl ṗk pl − ϕ̇ sin(ϕ1 − ϕ2 )
k=1 k=1 l=1
2 1

N2
N1
− μl1 ϕ̇12 sin(ϕ1 − ϕ2 ) Hk pk + μl1 ϕ̇12 cos(ϕ1 − ϕ2 ) C i qi
k=1 i=1
= (l2 + u B ) sin ϕ 2 .λ1 − (l2 + u B ) cos ϕ2 .λ2 (6.15)
IC ϕ̈3 + l3 sin(ϕ3 )λ1 − l3 cos(ϕ3 )λ2 = 0 (6.16)

N2
N1
μl1 Ci ϕ̈1 cos(ϕ1 − ϕ2 ) + μDi + μ n ik pk ϕ̈2 + μ m i j q̈ j
k=1 j=1

N2
N1
N1
− μl1 ϕ̇12 sin(ϕ1 − ϕ2 )Ci + 2μϕ̇2 n ik ṗk − μϕ̇22 mi j q j + E I ki j q j = 0
k=1 j=1 j=1
(6.17)

N1
N2
− μl1 ϕ̈1 sin(ϕ1 − ϕ2 )Hk − μϕ̈2 n ik qi + μ bkl p̈l − μl1 ϕ̇12 cos(ϕ1 − ϕ2 )Hk
i=1 l=1

N1
N2
N2
− 2μϕ̇2 n ik q̇i − μ Fk + bkl pl ϕ̇22 + E A gkl pl
i=1 l=1 l=1
+ (λ1 cos ϕ2 + λ2 sin ϕ2 )αk = 0 (6.18)
where i = 1, 2,…, N 1 , k = 1, 2,…, N 2 and

1 when k = 2n + 1, n = 1, 2, . . .
αk =
−1 when k = 2n, n = 1, 2, . . .
A set of (3 + N 1 + N 2 ) differential Eqs. (6.14)–(6.18) and two algebraic constraint

Eqs. (6.4a), (6.4b) denotes motion equations of the mechanism with the elastic cou-
pler link. Thus, this set consists of (5 + N 1 + N 2 ) differential–algebraic equations
with the same number of variables that can be expressed in compact matrix form as
the form presented in [3, 4]
M(s, t)s̈ + Ts (s,t)λ = −C(s, ṡ)ṡ − g(s) + τ (t) (6.19)
f(s, t) = 0 (6.20)
where s = [ϕ1 ϕ2 ϕ3 q1 . . . q N 1 p1 . . . p N 2 λ1 λ2 ]T , λ is the vector of Lagrange multipli-

ers, f is the vector of constraint equations.
6.3 Dynamic Analysis
In the next investigation, the Lagrange’s multiplier partition method is used to solve
Eqs. (6.19) and (6.20) numerically [3, 4]. The parameters of the four-bar mechanism
are given in Table 6.1.
The torque on the input link is given by:
Table 6.1 Parameters of Constant [unit] Description Value

four-bar mechanism
l 0 [m] Length of the ground link 0.4064
l 1 [m] Length of the input link 0.0635
l 2 [m] Length of the coupler link 0.3048
l 3 [m] Length of the output link 0.3048
I O [kgm2 ] Moment of inertia of the 7.466 × 10−6
input link
I C [kgm2 ] Moment of inertia of the 2.002 × 10−3
output link
μ [kg/m] Mass per unit length 0.2237
E [N/m2 ] Modulus of elasticity 2.06 × 1011
A [m2 ] Cross-sectional area of the 8.19 × 10−6
coupler link
m1 [kg] Mass of the input link 0.0142
m2 [kg] Mass of the coupler link 0.0682
m3 [kg] Mass of the output link 0.0682

τ0 sin(2π t/Tm ) t ≤ Tm
τ (t) = (6.21)
0 t > Tm
where τ 0 is the peak torque and T m is the time duration of the torque (Table 6.1).
The coordinates of the real position of the four-bar mechanism with a flexible
coupler link can be expressed in the following way.
ϕ1 = ϕ1d + y1 , ϕ2 = ϕ2d + y2 , ϕ3 = ϕ3d + y3 , qi = y4 ,

y5 , . . . y N1 +3 , pk = y N1 +4 , . . . , y N1 +N2 +3 (6.22)
where ϕ 1d , ϕ 2d , ϕ 3d represent the trajectory of rigid mechanism, y1 , y2 , y3 are devi-

ation of flexible trajectory versus rigid trajectory.
It is assumed that y1 , y2 ,…, y N1 +N2 +3 are small. The initial conditions are selected
as follows: input angle ϕ 10 = π /2, angular velocity of input link ϕ̇10 = 0, elastic
deformations qi0 = 0, pk0 = 0 and elastic deformations velocity q̇i0 = 0, ṗk0 = 0.
Using Newton-Raphson method for solving the contraint Eqs. (6.4a) and (6.4b)
with these initial conditions, the values of initial angles ϕ 20 , ϕ 30 are obtained as
ϕ20 = 0.6752 (rad) and ϕ30 = 2.1564 (rad). By differentiating Eqs. (6.4a), (6.4b)
and solving the new obtained equations, we get initial conditions of angular velocity
of coupler link and output link as ϕ̇20 = 0, ϕ̇30 = 0.
Figures 6.2 and 6.3 show the responses of rigid and flexible mechanism with a
peak torque magnitude of τ 0 = 0.005 (Nm) and T m = 1 (s). As mentioned in the first
section, the rigid body coordinates (e.g. input and output link angular displacements)
are affected by the elastic deformation of the coupler link due to the fully coupled
Fig. 6.2 Crank angle, …

rigid, flexible, τ 0 = 0.005
Nm
Fig. 6.3 Output angle, …

rigid, flexible, τ 0 = 0.005
Nm
nature of the equations. However, this effect is negligible when the peak of torque is
small.
Another set of simulations is carried out with a peak torque magnitude of τ 0
= 0.03 (Nm), T m = 1 (s) and the numerical simulations are performed during the
period from 0 to 5 (s). The responses for the flexible and rigid models of the four-
bar mechanism are shown in Figs. 6.4, 6.5, 6.6 and 6.7. The effect of flexibility is
now more pronounced, since the larger torque causes larger elastic deformations. In
Figs. 6.8 and 6.9 the transverse and longitudinal deformation of the coupler link are
depicted.
Fig. 6.4 Crank angle, …

rigid, flexible, τ 0 = 0.03 Nm
Fig. 6.5 Output angle, …

rigid, flexible, τ 0 = 0.03 Nm
Fig. 6.6 Crank angular

velocity, … rigid, flexible, τ 0
= 0.03 Nm
Fig. 6.7 Output angular

velocity, … rigid, flexible, τ 0
= 0.03 Nm
Fig. 6.8 Transverse

deformation of flexible
coupler link at the midpoint,
τ 0 = 0.03 Nm
Fig. 6.9 Longitudinal

coupler link, τ 0 = 0.03 Nm
6.4 Model-Based Control
To investigate the possibility to suppress vibrations of the flexible link using a control
torque applied to the input link, a simple control strategy using a PD controller is used
for the purpose of reducing values y1 , y2 , y3, … y N1 +N2 +3 approaching to zero when
the time duration approaches infinity or large enough. It follows from Eqs. (6.19)
and (6.20) that for this case
M(s, t)s̈ + Ts (s,t)λ = −C(s, ṡ)ṡ − g(s) + τ (t) + τC (t) (6.23)
where the control torque applied on the input link is given by:
τc = −k P y1 − kd ẏ1 (6.24)
where k p , k d are the PD controller gains, y1 = ϕ1 − ϕ1d , ẏ1 = ϕ̇1 − ϕ̇1d . Thus,
the total torque acting on the input link is τ + τ c . The following relation for the
numerical simulation are used
τc (ti+1 ) = −k P y1 (ti ) − kd ẏ1 (ti ), i = 0, 1, 2, . . . (6.25)
The parameters used in the following control simulations are given in Table 6.1.
The controller gains (trial and error) are chosen as k p = 0.15 and k d = 0.08, the
peak torque magnitude τ 0 = 0.03 (Nm) and T m = 1 (s). Some calculating results
are shown in Figs. 6.10, 6.11, 6.12 and 6.13. It can be seen that, the controller is
able to suppress the vibrations and control the link angular motions. The transverse
deformation is suppressed within 1.2 (s) as shown in Fig. 6.14, the longitudinal
Fig. 6.10 Crank angle with

PD controller, … rigid,
flexible
Fig. 6.11 Output angle with

PD controller, … rigid,
flexible
Fig. 6.12 Crank angular

velocity with PD controller,
… rigid, flexible
deformation is suppressed within 0.85 (s) in Fig. 6.15. Finally, Fig. 6.16 shows the
value of control torque τC in time.
6.5 Conclusions
The present study deals with modelling and model-based control of a four-bar mech-
anism with the flexible coupler link. It is assumed that the longitudinal deformation of
the coupler link is not dependent on its transverse deformation. Resulting differential-
algebraic equations are solved numerically to simulate the system behaviour. A sim-
ple PD controller was designed without requiring measurement of the elastic defor-
Fig. 6.13 Output angular

velocity with PD controller,
… rigid, flexible
Fig. 6.14 Transverse

coupler link at the midpoint,
using PD controller
mations. It has been shown that this controller can be efficient in suppressing the
vibrations of the flexible link as well as controlling the rigid body motion.
Acknowledgements This paper was completed with the financial support of the Vietnam National
Foundation for Science and Technology Development (NAFOSTED).
Fig. 6.15 Longitudinal

coupler link, using PD
controller
Fig. 6.16 Control torque of

the PD controller
References
1. Bremer, H.: Elastic Multibody Dynamics—A Direct Ritz Approach. Springer, Berlin (2008)
2. Geradin, M., Cardona, A.: Flexible Multibody Dynamics—A Finite Element Approach. Wiley,
Chichester (2001)
3. De Jalon, J.G., Bayo, E.: Kinematic and Dynamic Simulation of Multibody Systems. Springer,
New York (1994)
4. Khang, N.V.: Dynamics of Multibody Systems, 2nd edn in Vietnamese. Science and Technics
Publishing House, Hanoi (2017)
5. Sung, C.K., Chen, Y.C.: Vibration control of the elastodynamic response of high-speed flexible
linkage mechanisms. ASME J. Vib. Acoust. 113, 14–21 (1991)
6. Beale, D.G., Lee, S.W.: The applicability of fuzzy control for flexible mechanisms. ASME
Des. Eng. Tech. Conf. 84(1), 203–209 (1995)
7. Liao, W.H., Chou, J.H., Horng, I.R.: Robust vibration control of flexible linkage mechanisms
using piezoelectric films. Smart Mater. Struct. 6, 457–463 (1997)
8. Karkoub, M., Yigit, A.S.: Vibration control of a four-bar mechanism with a flexible coupler. J.
Sound Vib. 222(2), 171–189 (1999)
9. Xianmin, Z., Changjian, S., Erdman, A.G.: Active vibration controller design and comparison
study of flexible linkage mechanism systems. Mech. Mach. Theory 37, 985–997 (2002)
10. Wang, X., Mills, J.K.: A FEM model for active vibration control of flexible linkages. In:
Proceedings of the IEEE International Conference on Robotics and Automation, New Orleans
(2004)
11. Zhang, X., Mills, J.K., Cleghorn, W.L.: Coupling characteristics of rigid body motion and
elastic deformation of a 3-PRR parallel manipulator with flexible links. Multibody Syst. Dyn.
21, 167–192 (2009)
12. Hill, D.: Dynamics and control of a rigid and flexible four bar coupler. J. Vib. Control. 20(1),
131–145 (2012)
13. Zhang, Q., Mills, J.K., Cleghorn, W.L., Jin, J., Zhao, Ch.: Trajectory tracking and vibration
suppression of a 3-PRR parallel manipulator with flexible links. Multibody Syst. Dyn. 33,
27–60 (2015)
14. Liang, D., Song, Y., Sun, T., Jin, X.: Rigid-flexible coupling dynamic modeling and investi-
gation of a redundantly actuated parallel manipulator with multiple actuation modes. J. Sound
Vib. 403, 129–151 (2017)
15. Inman, D.J.: Engineering Vibrations, 2 edn. Prentice Hall, Upper Saddle River, NJ (2001)
16. Reddy, J.N.: Energy Principles and Variational Methods in Applied Mechanics, 2nd edn. Wiley,
Upper Saddle River, NJ (2002)
Chapter 7
Structural Synthesis of 2-DOF, 7-Link,
2-Basic Loops Simple and Multiple Joint
Mechanisms for Robotics
V. I. Pozhbelko and E. N. Kuts
Abstract The paper proposes a new approach to structural synthesis of all possible
2-DOF, 7-link, 2-basic (independent) loops both non-fractionated and fractionated
planar closed kinematic chains with simple as well as multiple joints. Then, various
examples of mechanical design of robots and manipulators based on synthesized
simple and multiple joint 2-DOF, 7-link, 2-basic loops kinematic chains are illustrated
to show the effectiveness of the structural synthesis method. The results of this
work provide mechanism designers the necessary data bank for generation of all
possible topological structures of multi-DOF mechanisms in the conceptual stage of
mechanical design.
Keywords Structural synthesis · Multiple joint · Degree of freedom · Manipulator
7.1 Introduction
Structural synthesis is the first step in the creation of various mechanisms and rep-
resents one of the most difficult tasks in mechanisms and machine theory [1, 2].
There exist a plethora of methods for structural synthesis of kinematic chains.
The most known methods are founded on Franke’s condensed notation, graph theory,
contracted graphs, Baranov trusses, group theory, permutation groups and matrices
[3–22].
On the one hand a large amount of studies have been devoted to develop methods
for structural synthesis of simple joint kinematic chains.
V. I. Pozhbelko
Department of Engineering Mechanics, South Ural State University (National Research
University), Chelyabinsk 454080, Russia
E. N. Kuts (B)
Department of Construction Production and Theory of Structures, South Ural State University
(National Research University), Chelyabinsk 454080, Russia

84 V. I. Pozhbelko and E. N. Kuts
On the other hand mechanisms can include kinematic chains with multiple joints,
that provides more simple construction and minimizes the space requirement. It
is especially important in robots design. However, structural synthesis of multibody
mechanisms with multiple joint kinematic chains is very complicated [23–30]. There-
fore, the structural synthesis of multiple joint kinematic chains is not fully studied
and so the development of new methods of structural synthesis is very important.
The main aim of this paper is to present the new approach to structural synthesis
of all possible 2-DOF, 7-link, 2-basic loops closed kinematic chains. Moreover, this
paper enunciates a new mobility equation and the method for design of simple and
multiple joint mechanisms for robots and manipulators.
7.1.1 Notations
ñ total number of links in the kinematic chains: ñ = n 2 + n 3 + · · · + n i ;

ni the number of links having i joints with other links in the kinematic chain;
νj the number of multiple joints (j ≥ 2);
ν2 the number of two-pin joints (denoted by sign “j2 ” in figures);
V resulted number of multiple joints (V = (j − 1)ν j );
V total multiple joint factor, V = 0—simple joint kinematic chains, V > 0—mul-
tiple joint kinematic chains;
V max the max value of resulted number of multiple joints;
K the number of independent closed loops in kinematic chains;
F the number of degrees of freedom of mechanical system with the fixed link
(F = DOF);
Li independent closed loop;
L0 the number of sides in the peripheral loop in the kinematic chain.
7.2 Basic Concepts
Before the discussion, we need to introduce the following basic concepts.
7.2.1 Link Assortment
The link assortment [L.A.] array of closed kinematic chain presents the type of links
(by number of i joints incident to the link) and the number of links in the kinematic
chain that are necessary to form the kinematic chain with given input parameters (F,
K and V ). [L.A.] can be expressed as:
7 Structural Synthesis of 2-DOF, 7-Link, 2-Basic Loops … 85
[L .A.] = [n 2 , n 3 , n 4 , . . . , n i ], (7.1)
where n2 , n3 , n4 ,…, ni are the number of binary, ternary, quaternary,…, i-nary links,
etc., in order, having i joints incident to the link.
7.2.2 Multiple Joint Assortment
Multiple joint assortment [M.J.A.] of closed kinematic chain is a set of numbers

consisting of the number of two-pin joints, three-pin joints, …, j-pin joints in the
chain. [M.J.A.] can be expressed as [28]:

[M.J.A.] = ν2 • ν3 • . . . • ν j , (7.2)
where ν 2 , ν 3 ,…, ν j are the number of j-pin joints (j ≥ 2).
7.2.3 Loop Assortment
Loop assortment [L α ] of closed simple or multiple joint kinematic chain is a set of

numbers consisting of numbers of links participating on the periphery of any loop
(α 3 , α 4 , α 5 ,…, α m , in order respectively) in the chain, including the peripheral loop
L 0 . [L α ] can be expressed as [30]:
[L α ] = [α3 − α4 − α5 − . . . − αm ] = [3 − 4 − 5 − . . . − m]. (7.3)
7.2.4 Total Multiple Joint Factor of the Kinematic Chain
Total multiple joint factor (V ) for closed kinematic chains can be expressed by
following equation:
V = ( j − 1)ν j = ν2 + 2ν3 + . . . + ( jmax − 1)ν jmax ≤ 2(K − 1). (7.4)
7.2.5 General Mobility Equation
Using the unified structure theory [30] for planar and spatial mechanical systems,
we introduce a new general mobility equation (degrees of freedom F):
F = (3 − i)n i − V − 3 = (n 2 − 3) − [n 4 + 2n 5 + . . . + (3 − i max )n imax ] − V.

(7.5)
Note that proposed general mobility equation Eq. (7.5) includes the specified total
multiple joint factor 0 ≤ V ≤ V max combined with type of links and number of links
ni (1 ≤ i ≤ imax ). Hence, it can be used for quick mobility calculation and type
synthesis of various simple and multiple joint kinematic chains.
7.3 Structural Synthesis Method for Simple and Multiple

Joint Closed Kinematic Chains
Our algorithm allows to synthesize closed planar simple and multiple joint kinematic
chains with specified degree of freedom F, the number of links ñ and the number of
independent closed loops K.
A general algorithm for structural synthesis of planar simple and multiple joint
kinematic chains with the minimum number of links participating on the periphery
of loop 4 and 5 (α min ≥ α 4 ) is based on addition method. This method uses different
sets of various numbers of links (N 1 , N 2 ) in closed (N 1 ) and open (N 2 ) kinematic
chains, where ñ = N1 + N2 = 7 = const.
The proposed algorithm (shown below in Fig. 7.1) of structural synthesis of 2-
DOF, 7-link, 2-basic loops generalized loops simple and multiple joint closed kine-
matic chains consists of the following steps:
Step 1. 3-link open kinematic chain is added on to quaternary link of the 4-link closed
kinematic chain (consisting of three adjacent binary links and one quaternary link).
As a result of structural synthesis, the 7-link fractionated simple joint kinematic chain
with [L.A.] = [6, 0, 1], [L α ] = [4 – 4 – 8], L 0 = 8 and one inversion of kinematic
chain with the peripheral loop L 0 = 4 are obtained.
Step 2. Add 3-link open kinematic chain to 2 adjacent ternary links of the 4-link
closed kinematic chain consisting of 2 adjacent binary links and 2 adjacent ternary
links. As a result of structural synthesis, obtain the 7-link non-fractionated simple
joint kinematic chain with [L.A.] = [5, 2], [L α ] = [4 – 5 – 7], L 0 = 7 and two
inversions of kinematic chain with the peripheral loops L 0 = 5 and L 0 = 4.
Step 3. Add 3-link open kinematic chain to 2 non-adjacent ternary links of the 4-link
closed kinematic chain (consisting of 2 non-adjacent binary links and 2 non-adjacent
ternary links). As a result of structural synthesis, obtain the 7-link non-fractionated
simple joint kinematic chain with [L.A.] = [5, 2], [L α ] = [4 – 6 – 6], L 0 = 6 and one
inversion of kinematic chain with the peripheral loop L 0 = 4.
Step 4. 2-link open kinematic chain is added on to 2 non-adjacent ternary links of
the 4-link closed kinematic chain (consisting of 3 binary links and 2 non-adjacent
ternary links). As a result of structural synthesis, obtain the 7-link non-fractionated
simple joint kinematic chain with [L.A.] = [5, 2], [L α ] = [5 – 5 – 6], L 0 = 6 and one
inversion of kinematic chain with the peripheral loop L 0 = 5.
Fig. 7.1 The process to form all possible 2-DOF, 7-link, 2-basic loops closed simple and multiple
joint planar kinematic chains with α min ≥ α 4 for mechanical engineering design
Step 5. 3-link open kinematic chain is added on to ternary link and adjacent joint of
the 4-link closed kinematic chain (consisting of 3 adjacent binary links and 1 ternary
link). As a result of structural synthesis, two 7-link multiple joint non-fractionated
kinematic chains are obtained:
• with one multiple joint (j2 ), where [L.A.] = [6, 1], [L α ] = [4 – 4 – 7], L 0 = 7 (V
= 1);
• with two multiple joints (j2 , j2 ), where [L.A.] = [7, 0], [L α ] = [4 – 4 – 6], L 0 = 6
(V = 2).
Step 6. 3-link open kinematic chain id added on to ternary link and non-adjacent
joint of the 4-link closed kinematic chain (consisting of 3 adjacent binary links and
1 ternary link). As a result of structural synthesis, two 7-link multiple joint non-
fractionated kinematic chains are obtained:
• with one multiple joint (j2 ), where [L.A.] = [6, 1], [L α ] = [4 – 5 – 6], L 0 = 6 (V
= 1);
• with two multiple joints (j2 , j2 ), where [L.A.] = [7, 0], [L α ] = [4 – 5 – 5], L 0 = 6
(V = 2).
Step 7. 2-link open kinematic chain is added on to ternary link and non-adjacent
joint of the 5-link closed kinematic chain (consisting of 4 adjacent binary links and
1 ternary link). As a result of structural synthesis, the 7-link multiple joint non-
fractionated kinematic chains are obtained:
• with one multiple joint (j2 ), [L.A.] = [6, 1], [L α ] = [5 – 5 – 5], L 0 = 5 (V = 1);
• with two multiple joints (j2 , j2 ), two inversions (kinematic chain with [L.A.] = [4,
2], [L α ] = [4 – 5 – 5] with L 0 = 4 (V = 2) and accordingly kinematic chain with
[L.A.] = [4, 2], [L α ] = [4 – 4 – 6]) with L 0 = 4 (V = 2).
Based on the results of structural synthesis of 2-DOF, 7-link, 2-basic loops closed
kinematic chains with α min = α 4 (Fig. 7.1) the existence of 4 types of simple joint
kinematic chains, 3 types of kinematic chains with one multiple joint (j2 ) and two
types of kinematic chains with 2 multiple joints (j2 , j2 ) is established.
7.4 Mobility Analysis of Simple and Multiple Joint

Kinematic Chains for Robotic Mechanisms
From the point of industrial application of synthesized 2-DOF, 7-link, 2-basic loops
simple and multiple joint closed kinematic chains (presented in Fig. 7.1) for mechan-
ical design of robots and pin-joint manipulators (presented in Fig. 7.2) it can be useful
to make mobility analysis.
So we propose to use new general mobility equation Eq. (7.5) for mobility anal-
ysis of synthesized closed simple and multiple joint kinematic chains, presented in
Fig. 7.1.
Fig. 7.2 Industrial applications of synthesized 2-DOF, 7-link, 2-basic loops simple (No. 1–8) and
multiple (No. 9–16) joint closed kinematic chains (presented in Fig. 7.1) for mechanical design of
robots and pin-joint manipulators
The detailed examples of mobility analysis of simple and multiple joint closed
kinematic chains are presented below.
Example 1 For closed simple-jointed kinematic chains No. 1 and No. 5 (single-arm
and dual-arm robots) shown in Fig. 7.2, we have original set of link assortment [L.A.]
= [n2 , n3 , n4 ] = [6, 0, 1], multiple joint assortment [M.J.A.] = [0] and loop assortment
[L α ] = [α 4 – α 4 – α 8 ] = [4 – 4 – 8], and the peripheral loop is L 0 = 8. Hence for
given structural parameters n2 = 6, n3 = 0, n4 = 1, V = 0 Eq. (7.5) yields the number

of degrees of freedom F = (n2 – 3) − n4 − V = (6 − 3) – 1 − 0 = +2.
and dual-arm robots) shown in Fig. 7.2, we have original set of link assortment [L.A.]
= [n2 , n3 ] = [5, 2], multiple joint assortment [M.J.A.] = [0] and loop assortment [L α ]
= [α 4 – α 5 – α 7 ] = [4 – 5 – 7], and the peripheral loop is L 0 = 7. Hence for given
structural parameters n2 = 5, n3 = 2, n4 = 0, V = 0 Eq. (7.5) yields the number of
degrees of freedom F = (n2 − 3) − n4 − V = (5 − 3) – 0 − 0 = +2.
robots) shown in Fig. 7.2, we have original set of link assortment [L.A.] = [n2 , n3 ]
= [5, 2], multiple joint assortment [M.J.A.] = [0] and loop assortment [L α ] = [α 4 –
α 6 – α 6 ] = [4 – 6 – 6], and the peripheral loop is L 0 = 6. Hence for given structural
parameters n2 = 5, n3 = 2, n4 = 0, V = 0 Eq. (7.5) yields the number of degrees of
freedom F = (n2 − 3) − n4 − V = (5 − 3) – 0 − 0 = +2.
robots) shown in Fig. 7.2, we have original set of link assortment [L.A.] = [n2 , n3 ]
= [5, 2], multiple joint assortment [M.J.A.] = [0] and loop assortment [L α ] = [α 5 –
α 5 – α 6 ] = [5 – 5 – 6], and the peripheral loop is L 0 = 6. Hence for given structural
parameters n2 = 5, n3 = 2, n4 = 0, V = 0 Eq. (7.5) yields the number of degrees of
freedom F = (n2 – 3) − n4 − V = (5 − 3) – 0 − 0 = +2.
Example 5 For closed multiple-jointed kinematic chains No. 9 and No. 10 with
one multiple joint (j2 ) (single-arm and dual-arm robots) shown in Fig. 7.2, we have
original set of link assortment [L.A.] = [n2 , n3 ] = [6, 1], multiple joint assortment
[M.J.A.] = [ν 2 • ν 3 • ν 4 ] = [1 • 0 • 0] and loop assortment [L α ] = [α 4 – α 4 – α 7 ]
= [4 – 4 – 7], and the peripheral loop is L 0 = 7. Taking into account Eq. (7.4) we
get V = ν 2 = 1. Thus for given structural parameters n2 = 6, n3 = 1, n4 = 0, V =
1 Eq. (7.5) yields F = (n2 − 3) − n4 − V = (6 − 3) – 0 − 1 = +2.
Example 6 For closed multiple-jointed kinematic chain No. 11 with one multiple
joint (j2 ) shown in Fig. 7.2 (single-arm robot), we have original set of link assortment
[L.A.] = [n2 , n3 ] = [6, 1], multiple joint assortment [M.J.A.] = [ν 2 • ν 3 • ν 4 ] = [1 •
0 • 0] and loop assortment [L α ] = [α 4 – α 5 – α 6 ] = [4 – 5 – 6], and the peripheral
loop is L 0 = 6. Taking into account Eq. (7.4) we get V = ν 2 = 1. Thus for given
structural parameters n2 = 6, n3 = 1, n4 = 0, V = 1 Eq. (7.5) yields the number of
degrees of freedom F = (n2 – 3) − n4 − V = (6 − 3) – 0 − 1 = +2.
Example 7 For closed multiple-jointed kinematic chain No. 12 with one multiple
joint (j2 ) shown in Fig. 7.2 (single-arm robots), we have with original set of link
assortment [L.A.] = [n2 , n3 ] = [6, 1], multiple joint assortment [M.J.A.] = [ν 2 • ν 3
• ν 4 ] = [1 • 0 • 0] and loop assortment [L α ] = [α 5 – α 5 – α 5 ] = [5 – 5 – 5], and the
peripheral loop is L 0 = 5. Taking into account Eq. (7.4) we get V = ν 2 = 1. Thus
for given structural parameters n2 = 6, n3 = 1, n4 = 0, V = 1 Eq. (7.5) yields the
number of degrees of freedom F = (n2 − 3) − n4 − V = (6 − 3) – 0 − 1 = +2.
Example 8 For closed multiple-jointed kinematic chains No. 13 and No. 14 with two
multiple joints (j2 , j2 ) (pin-joint manipulators) shown in Fig. 7.2, we have original
set of link assortment [L.A.] = [n2 , n3 ] = [7, 0], multiple joint assortment [M.J.A.]
= [ν 2 • ν 3 • ν 4 ] = [1 • 0 • 0] and loop assortment [L α ] = [α 4 – α 4 – α 6 ] = [4 – 4 –
6], and the peripheral loop is L 0 = 6. Taking into account Eq. (7.4) we get V = ν 2
= 2. Thus for given structural parameters n2 = 7, n3 = 0, n4 = 0, V = 2 Eq. (7.5)
yields F = (n2 − 3) − n4 − V = (7 − 3) – 0 – 2 = +2.
Example 9 For closed multiple-jointed kinematic chains No. 15 and No. 16 with two
multiple joints (j2 , j2 ) (pin-joint manipulators) shown in Fig. 7.2, we have original
set of link assortment [L.A.] = [n2 , n3 ] = [7, 0], multiple joint assortment [M.J.A.]
= [ν 2 • ν 3 • ν 4 ] = [2 • 0 • 0] and loop assortment [L α ] = [α 4 – α 5 – α 5 ] = [4 – 5 –
5], and the peripheral loop is L 0 = 5. Taking into account Eq. (7.4) we get V = ν 2
= 2. Thus for given structural parameters n2 = 7, n3 = 0, n4 = 0, V = 2 Eq. (7.5)
yields F = (n2 − 3) − n4 − V = (7 − 3) – 0 − 2 = +2.
In such a way, based on synthesized kinematic chains with α min = α 4 (Fig. 7.1)
all simple-jointed and multiple-jointed mechanisms of various robots and manipula-
tors (Fig. 7.2) are constructed. These mechanisms are efficient and compact 2-DOF
mechanical systems for application in different branches of engineering.
7.5 Conclusions
In this work, we have introduced novel structural synthesis method based on addition
method and the basic concepts of link assortment, multiple joint assortment and loop
assortment.
On the basis of proposed structural synthesis method, the existence of all the
possible synthesized 9 types of simple and multiple joint 2-DOF, 7-link, 2-basic
loops kinematic chains (presented in Fig. 7.1) is fully ascertained and illustrated.
The examples of structural and mobility analysis of various 2-DOF, 7-link, 2-basic
loops simple and multiple joint closed kinematic chains for industrial applications
in robotics and manipulators are provided (in Fig. 7.2 for case α min ≥ α 4 ) and
demonstrate the efficiency of proposed new mobility equation for mobility analysis
of different mechanisms.
The results of this work provide mechanism designers the necessary data bank
for the generation of all possible topological structures of 2-DOF, 7-link multi-DOF
mechanisms in the conceptual stage of mechanical design.
Acknowledgements The work was supported by Act 211 Government of the Russian Federation,
contract No 02.A03.21.0011.
References
1. Norton, R.L.: Machine Design. Upper Saddle River, Prentice-Hall, NJ (1966)

2. Hunt, K.H.: Kinematic Geometry of Mechanisms. Oxford University Press, Oxford (1978)
3. Erdman, A.G., Sandor, G.: Mechanism Design: Analysis and synthesis. Englewood Cliffs, NJ
(1984)
4. Jensen, P.W.: Classical and Modern Mechanisms for Engineers and Inventors. Marcel Dekker,
New York (1991)
5. Yan, H.S.: Creative Design of Mechanical Devices. Springer, Singapore (1998)
6. Ghosal, A.: Robotics: Fundamental Concepts and Analysis. Oxford University Press, Oxford
(2006)
7. Crossley, F.R.E.: A contribution to Gruebler’s theory in the number synthesis of plane mecha-
nisms. J. Eng. Ind. 86(1), 1–8 (1964)
8. Mruthyunjaya, T.S., Raghavan, M.R.: Structural analysis of kinematic chains and mechanisms
based on matrix representation. J. Mech. Des. 100(3), 488–494 (1978)
9. Davies, T.H., Crossley, F.R.E.: Structural analysis of plane linkages by Franke’s condensed
notation. Mech. Mach. Theory 1(2), 171–183 (1966)
10. Manolescu, N.I.: A method based on Baranov trusses and using graph theory to find the set of
planar jointed kinematic chains and mechanisms. Mech. Mach. Theory 8(1), 3–22 (1973)
11. Mruthyunjaya, T.S.: Structural synthesis by transformation of binary chains. Mech. Mach.
Theory 14(4), 221–231 (1979)
12. Yan, H.S., Hwang, Y.W.: Number synthesis of kinematic chains based on permutation groups.
Math. Comput. Model. 13(8), 29–42 (1990)
13. Yan, H.S.: A methodology for creative mechanism design. Mech. Mach. Theory 27(3), 235–242
(1992)
14. Hwang, W.M., Hwang, Y.W.: Computer-aided structural synthesis of planar kinematic chains
with simple joints. Mech. Mach. Theory 27(2), 189–199 (1992)
15. Lee, H., Yoon, Y.: Automatic method for enumeration of complete set of kinematic chains.
JSME Int. J. 37(4), 812–818 (1994)
16. Tuttle, E.R.: Generation of planar kinematic chains. Mech. Mach. Theory 31(6), 729–748
(1996)
17. Mruthyunjaya, T.S.: Kinematic structure of mechanisms revisited. Mech. Mach. Theory 38(4),
279–320 (2003)
18. Sunkari, R.P., Schmidt, L.C.: Structural synthesis of planar kinematic chains by adapting a
Mckay-type algorithm. Mech. Mach. Theory 41(9), 1021–1030 (2006)
19. Yan, H.S., Chiu, Y.T.: On the number synthesis of kinematic chains. Mech. Mach. Theory
89(7), 128–144 (2015)
20. Ding, H.F., Huang, Z.: A new theory for the topological structure analysis of kinematic chains
and its applications. Mech. Mach. Theory 42(10), 1264–1279 (2007)
21. Ding, H., Hou, F., Kecskemethy, A., Huang, Z.: Synthesis of a complete set of contracted
graphs for planar non-fractionated simple-jointed kinematic chains with all possible DOFs.
Mech. Mach. Theory 46(11), 1588–1600 (2011)
22. Yan, H.S., Chiu, Y.T.: An algorithm for the construction of generalized kinematic chains. Mech.
Mach. Theory 62, 75–98 (2013)
23. Yan, H.S., Hsu, C.H.: Contracted graphs of kinematic chains with multiple joints. Math. Com-
put. Model. 10(9), 681–695 (1988)
24. Jin-Kui, C., Wei-Qing, C.: Systemics of Assur groups with multiple joints. Mech. Mach. Theory
33(8), 1127–1133 (1998)
25. Hsu, C.H.: Enumeration of basic kinematic chains with simple and multiple joints. J. Frankl.
Inst. 329(4), 775–789 (1992)
26. Pozhbelko, V.I.: Edinaya teoriya struktury, strukturnyj sintez i analiz staticheski opredelimyh
mekhanicheskih sistem na osnove novoj formuly podvizhnosti (A complete theory of structure,
structural synthesis and analysis statically determinate mechanical systems on base of a new
degrees of freedom equation). Teoriya mekhanizmov i mashin 22(2), 15–37 (2013)
27. Chu, J., Zou, Y.: An algorithm for structural synthesis of planar simple and multiple joint
kinematic chains. Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 228(12), 2178–2192 (2014)
28. Pozhbelko, V., Ermoshina, E.: Number structural synthesis and enumeration process of all
possible sets of multiple joints for 1-DOF up to 5-loop 12-link mechanisms on base of new
mobility equation. Mech. Mach. Theory 90(8), 108–127 (2015)
29. Ding, H., Yang, W., Zi, B., Kecskemethy, A.: The family of planar kinematic chains with two
multiple joints. Mech. Mach. Theory 99(5), 103–116 (2016)
30. Pozhbelko, V.: A unified structure theory of multibody open, closed loop and mixed mechanical
systems with simple and multiple joint kinematic chains. Mech. Mach. Theory 100(6), 1–16
(2016)
Chapter 8
Numerical Synthesis of Stephenson
Six-Bar Mechanism Using a CAD
Geometric Approach
Yanhua Zhang and Chunhui Yang
Abstract Geometric construction and analytical calculation are recognised as two

basic approaches for planar mechanism design. However the geometric construction
is only available for relatively simple cases while the analytical calculation is quite
complicated and far from being visualised too. Both of them are complicated to be
used for the Stephenson six-bar mechanism design due to complexity. Therefore the
Computer-aided Design (CAD) geometric approach is developed to fulfil research
needs. In this study, an approximate position and posture synthesis for the Stephenson
six-bar mechanism is investigated via using the CAD geometric approach. Firstly
using the geometric constraint and dimension driving techniques, a primary simulated
mechanism is generated. Then based on different tasks of path and motion generations
for the dimensional synthesis, the simulated mechanisms of the Stephenson six-
bar mechanism are developed from primary simulation. The computer simulation
results on approximation dimensional synthesis of the mechanism prove that the
CAD geometric approach not only visualises the mechanism accurately and reliably
but also increases the number of prescribed positions of synthesis for mechanism.
Keywords Stephenson six-bar mechanism · CAD geometric approach ·

Path sythesis · Motion Sythesis
8.1 Introduction
The purpose of synthesis of mechanisms is to achieve motion transformation from

input to output of such mechanisms, which can generate specifically desired motions
at certain positions with specified postures [1]. The main outputs of the mechanisms
include function, path, and motion and so on.
Y. Zhang
College of Vocational Technology, Hebei Normal University, Shijiazhuang 050031, China
C. Yang (B)
2751, Australia

96 Y. Zhang and C. Yang
The approximate mechanism synthesis refers to that if the mechanism can produce
accurate positions and postures at some positions, but which can produce approx-
imate positions and postures at other positions, and the differences between actual
and provided positions and postures must be within an allowable limit. The geo-
metric method and analytic method are two basic methods commonly used [2]. The
geometric method is simple and visualised, but inaccurate, less repeatable, and pro-
vides only a few numbers of synthesis positions. The analytic method is accurate and
repeatable, but complex and not visualised [3–5]. Thus the CAD geometric approach
was developed for mechanism synthesis. This method is very useful and powerful to
research all kinds of mechanisms for researchers.
In this study, the CAD geometric approach is developed and utilised for conducting
the approximate position and motion syntheses for Stephenson six-bar mechanism
via using the CAD geometric approach as a showcase. Firstly using the geomet-
ric constraint and dimension driving techniques, a primary simulated mechanism is
constructed. Then based on different tasks of path and motion generations for the
dimensional synthesis, the simulated mechanisms of the Stephenson six-bar mech-
anism are developed from the primary simulated model. The computer model and
simulations are further used for the path synthesis and motion synthesis to extract
approximated solutions, which are further validated by comparing with those accu-
rate solutions availiable.
8.2 Simulation of Basic Stephenson Six-Bar Mechanism
The Stephenson six-bar mechanism is shown in Fig. 8.1a, which is used to construct
a mechanism synthesis at the k prescribed position via using the CAD geometric
approach. When the mechanism is at the ith position, two triangles Δ1 and Δ2 , and
four lines O1 O2 , L bi , L di and L ei constitute the mechanism. Line O1 O2 is fixed, and
line L ei is a floating bar. In order to produce an approximate orientation of the bar
L ei , the two lines M i and N i are constructed using the suitable geometric constraint
and dimension driving. In order to produce an approximate trajectory positions of
the bar L ei , the line S i is constructed using the suitable geometric constraint and
dimension driving. The multiple approximate positions and postures of bar L ei are
then produced.
First, the basic mechanism at the kth position is constructed using the geometric
constraint and dimension driving, before constructing the Stephenson six-bar simu-
lation mechanism, as shown in Fig. 8.1b. The construction process is described as
follows:
1. First, construct one line L a1 and a dimension Da1 , then the same k lines are
moved and copied using ‘Move’ and ‘Copy’ commands, where k is the prescribed
number of positions. The dimensions Dai are used as driving dimensions, setting
Dai = Da1 (i = 2, 3…k), and the dimensions Dai of the line L ai (i = 2, 3…k)
are changed as driven dimensions using the constraint function of the dimension
8 Numerical Synthesis of Stephenson Six-Bar Mechanism … 97
Fig. 8.1 Stephenson six-bar mechanism and its basic simulation
equation. In order to simplify the view, all the driven dimensions Dai (i = 2,
3…k) are hided using the ‘Hide’ command. All the driven dimensions Dai (i =
2, 3…k) change accordingly when modifying the driving dimension Da1 .
2. Second, construct two circles C a and C b with the radii Ra and Rb , respectively.
The two centres of circles are connected to the O1 using a geometric constraint
command. The two ends of the set of lines are respectively connected to the point
ai on the C a and the point bi on C b (i = 2, 3…k) using coincide command. Thus
three sets of lines Ra , Rb and L ai can made up k identical equivalent triangle Δ1i
(i = 1, 2, … k), and the triangle Δ1i can rotate with O1 .
3. Repeat Steps 1 and 2, construct a line L ci , two circles C c and C d , and respectively
dimensioned Dci , Rc and Rd , O2 as the center of the circle. Thus three sets of
lines Rc , Rd and L ci can made up k identical equivalent triangle Δ2i (i = 1, 2, …
k), and the triangle Δ2i can rotate with respect to O2 .
4. Repeat Step 1, construct three sets of lines L bi , L di and L ei (i = 1, 2, 3 … k).
5. Lastly, the two lines L ei and L di are connected to the point ei at the ith position
using the ‘Coincide’ command. The free ends of L ei and L di are connected to ai
on the C a and the point d i on C d using the ‘Coincide’ command, respectively.
The two ends of the L bi are connected to bi on the C b and the point ci on C c .
8.3 Path Synthesis of the Simulated Stephenson Six-Bar

Mechanism
The one point at the floating bar should move along the prescribed path relative
to the base following the rule of the path synthesis. According to the requirements
of the path synthesis, the six-bar mechanism of the approximate path synthesis is
constructed on the basis of the basic simulation mechanism, as shown in Fig. 8.2.
The input function is determined by the accurate angle Φ i of the crank (the triangle
Fig. 8.2 Path synthesis of the Stephenson six-bar simulated mechanism
Δ1 ). The position of the output path is determined by the approximate position of the
point ei in the floating bar L ei . At the ith position in Fig. 8.1, the end point pi of the
line S i is as the accurate position point of the output path, the other point ei of the line
S i is as the approximate position point of the output path. Obviously, the point ei is
restricted the arc C s of the radius S i , so any point on the C s may be the approximate
position of the bar L ei . Thus not only the number of the mechanism constraint is
reduced, but also the number of the prescribed synthesis position is added. The path
synthesis of the Stephenson six-bar mechanism can be achieved at more positions.
When the length of the radius S i is gradually reduced to the allowable limit, the ideal
simulation mechanism of approximate path synthesis can be obtained, as shown in
Fig. 8.2b.
The construction process is described as follows:
1. Construct one datum line B, the two ends B1 and B2 of the line B are fixed, as
shown in Fig. 8.1a.
2. Construct k lines Pi (i = 1, 2, 3 … k), the end of the line Pi is connected to point
B2 , as shown in Fig. 8.1b.
3. Construct k lines Rai (i = 1, 2, 3 … k), the two ends of line Rai are connected to
the points O1 and ai . The angle of the lines Ra1 and Rai+1 (i = 1, 2, 3 … k) is as
input function, and label prescribed value.
4. Construct k lines S i and dimension Dsi , the two ends of the line S i are connected
to the other end of the line Pi and point ei of the bar L ei .
5. Gradually reduce the dimension Ds1 until Ds1 = 1 mm, the point pi and the point
ei are almost coincident.
6. Label the prescribed value for angle θ i and length of the line Pi. The maximum
difference between the accurate k position and the actual position of the floating
bar is defined by the dimension Ds1 .
The results can be obtained from the approximate path synthesis of the six-bar sim-
ulation mechanism as follows:
• When the dimension of line S 1 reduced to zero (Ds1 = 0), the approximate path
synthesis and the accurate path synthesis of the simulated mechanism are the same,
as shown in Fig. 8.2a. When the angle Φ i is given with a prescribed value of the
input function, and the angle θ i and the length of the line Pi are given with the
prescribed value for the output path, the maximum number of positions is found
as 6 in the path synthesis of the simulated mechanism.
• When the length of the line S 1 is reduced to 1 mm (Ds1 = 1 mm), the actual path
position of the floating bar is close to accurate prescribed position, their difference
is within the limit. Thus the simulated mechanism can achieve the approximate
synthesis of the path position and orientation. Now the number k of the floating
bar can be increased to 11, all the angle Φ i , the angle θ i and the length of the
lines Pi can be given with initial values of driving dimensions. This shows that
the maximum number of prescribed positions can be increased to 11, which are
significantly more than 6 of the prescribed positions of the accurate path synthesis
of the simulated mechanism. These results show that the six-bar mechanism can
complete a path synthesis with more prescribed positions.
8.4 Motion Synthesis of the Simulated Stephenson Six-Bar

Mechanism
According to the requirements of the motion synthesis, two kinds of motion syntheses
of the six-bar mechanism are constructed for the basic simulated mechanism, as
shown in Figs. 8.3 and 8.4, respectively.
In the first motion synthesis of the simulated mechanism, the azimuth angle of
the floating bar L ei is given with the prescribed accurate value at the k prescribed
position, and the path position of the floating bar is given with an approximate value
of the allowable limit. The constructed method of the first motion synthesis of the
simulated mechanism is similar to the approximate path synthesis for the simulated
mechanism. The prescribed angle αi (i = 1, 2, 3 … k−1) between the line B and the
bar L ei will be replaced by using accurate input angles Φ i , as shown in Fig. 8.3b.
The result can be obtained from the approximate motion synthesis of the simulated
six-bar mechanism as follows:
1. When the dimension of the line S 1 is reduced to zero (Ds1 = 0), the actual path
position of the floating bar L ei is the same as the accurate path position at the k
prescribed positions, as shown in Fig. 8.3a. The maximum number of positions
is 4 in the motion synthesis of the simulated mechanism.
2. When the dimension of the line S 1 is reduced to 1 mm (Ds1 = 1 mm), the actual
path position of the floating bar L ei is very close to the accurate path position at
k prescribed position, as shown in Fig. 8.3b. Using the first six-bar simulation
mechanism of approximate motion synthesis, which shows that the maximum
Fig. 8.3 Motion synthesis of the first Stephenson six-bar simulated mechanism
Fig. 8.4 Motion synthesis of the second Stephenson six-bar simulated mechanism
number of prescribed positions can be increased to 11, which is significantly

more than 4 of the prescribed positions of the accurate motion synthesis of the
simulated mechanism.
In the second motion synthesis of the simulated mechanism, the azimuth angle of the
floating bar L ei is given prescribed approximate value at k prescribed positions, and
the path position of the floating bar is given with an accurate value of the allowable
limit. When the dimension of line S 1 reduced to zero (Ds1 = 0), the floating bar L ei
can obtain accurate position. In order to obtain approximate azimuth angle, the two
sets of lines M i and N i (i = 1, 2, 3 … k−1) are increased at the basic simulation

mechanism in Fig. 8.1, as shown in Fig. 8.4.
The construction process is described as follows:
The first end of the line M i is connected to the point ei of the floating bar L ei , and
the second end of the line M i is connected to the first end of line N i . The second end
of the line N i is connected to the point qi of the floating bar L ei (i = 1, 2, 3 … k−1),
which is equivalent to a sliding pair of planar mechanism. Because the geometric
constraints can make point qi sliding in line L ei , the line M i can rotate with point ei .
When the azimuth angle ψi between line B and line Mi is given prescribed accurate
value, the actual approximate azimuth angle Φ i (i = 1, 2, 3 … k−1) of the bar L ei can
be obtained between line B and line L ei , as shown in Fig. 8.4. The results show that
it not only reduces the constraint number of the second motion synthesis simulation
mechanism, but also increases prescribed position number.
When the dimension of line N i (i = 1, 2, 3 … k−1) are gradually reduced to
small enough, which can construct the motion synthesis of the second Stephenson
simulated mechanism as shown in Fig. 8.4. The maximum difference ψ i (i = 1,
2, 3 … k) between the angle ψ i and angle ϕ i can be obtained using the equation ψ
= |ϕ i −ψ i | = N i /M i at k prescribed positions. The motion synthesis of the second
Stephenson simulated mechanism is under ideal conditions, so long as the ψ is
less than the allowable value.
The result can be obtained from the approximate motion synthesis of the second
six-bar simulated mechanism as follows:
When the dimension of line S 1 is reduced to zero (Ds1 = 0), and the dimension of
line N 1 is reduced to small enough, and the dimension of the line M 1 is given with
an equal value (700 mm) with the floating bar, the actual path position of the floating
bar is the same as its exact position, as shown in Fig. 8.4. Based on the approximate
motion synthesis of the second six-bar simulated mechanism, the maximum number
of the prescribed positions can be increased to 11, which is significantly more than 4 of
the prescribed positions of the accurate motion synthesis of the simulated mechanism,
as shown in Fig. 8.4.
8.5 Conclusion
The path and motion simulation mechanism have been constructed by using the CAD
geometric approach. The six-bar approximate simulation mechanism obtained by
using this approach can be used for path and motion syntheses at 11 prescribed
positions, which are more than the number of the accurate simulation mechanisms.
The results show that the CAD geometric approach is equivalent to the geometric
method and the analytic method but the CAD geometric approach is more cost-
effective, visualised, accurate and repeatable. Therefore this method can be not only
used in simple mechanism, but also can be used to analyse complex mechanisms.
References
1. Mlinar, J.R., Erdman, A.G.: An introduction to Burmester field theory. J. Mech. Des. (AMSE)
122(1), 25–30 (2000)
2. Loerch, R.J., Erdman, A.G., Sandor, G.N.: On the existence of circle-point and center point
circles for three precision point Dyad synthesis. J. Mech. Des. (ASME) 100(3), 554–562 (1979)
3. Eleashy, H., Elgayyar, M.S., Shabara, M.N.: Synthesis of one degree-of-freedom 6-bar linkages
from three degree-of-freedom open 4-bar chain using structural code technique. Int. J. Eng. Res.
Appl. 2(2), 065–069 (2012)
4. Pennock, G.R., Israr, A.: Kinematic analysis and synthesis of an adjustable six-bar linkage.
Mech. Mach. Theory 44(6), 306–323 (2009)
5. Chaudhary, K., Chaudhary, H.: Kinematic and dynamic analysis of Stephenson six bar mecha-
nism using hyperWorks. In: Altair Technology Conference, pp. 1–7 (2013)
Chapter 9
Foot Reaction Analysis of Whole Body
Dynamic via Screw Theory
Ming-June Tsai and Wan-Shan Yin
Abstract Whole body motion dynamic analysis is a complex job since the required
parameters are not easy to obtain. This study proposes a method for determining the
dynamic parameters of whole body. Then the wrench screw exerts on the body ($0 )
is computed by the Newton/Euler equations. The computing of single foot reaction
wrench is trivial using screw theory. However, there is no single solution for dou-
ble feet contact. Using linear dependency among the reactions of both feet and $0 ,
there are 12 equations to solve 14 unknowns. The geometry of the reaction screws
compose a general two-system, namely a cylindroid, is characterized by two param-
eters. Therefore, an optimal foot reactions can be computed under the assumption of
minimum internal moment generated by the two contact screws.
Keywords Whole body dynamic analysis · Screw theory · Zero moment point
9.1 Introduction
Foot reaction analysis of whole body dynamic has its major application in the field
of robotics and biomechanics. The purpose of the foot reaction in biomechanical
analysis is to estimate the force supported by bones or the moment generated at
joints muscle. However, the required parameters are not easy to obtain. In particular,
body segment parameters (BSP) are the essential inputs in dynamics analysis of body
motion, which include the mass, center of mass (CoM), moment of inertia (MOI) and
the principle axes. Traditionally, predictive equations based on cadavers or medical
scanning of live subjects is one way to estimate the BSP [1, 2]. Other methods are
based on magnetic resonance imaging (MRI) [3], and dual energy absorptiometry
[4]. A 3D body scanning method [5] is another way to determine the BSPs. Further-
M.-J. Tsai (B)

Department of Mechanical Engineering, National Cheng Kung University, Tainan, Taiwan
W.-S. Yin
Metal Industry Research & Development Centre, 1001 Kaoshiung Highway, Kaoshiung City 811,
Taiwan

104 M.-J. Tsai and W.-S. Yin
more, 3D models are often used to track human dynamic motion. Ensminger et al. [6]
determined the upper extremity net joint forces/moments during wheelchair propul-
sion using a 3D human model. Liu et al. [7] presented a 3D lower limb kinematic and
kinetic analysis using a wireless sensor system. Nakamura et al. [8] constructed a 3D
musculoskeletal model whose body segments were applied to predict the muscle and
tendon tensions. Tsai and Fang [9] postulated a data structure of 3D body geometric
model (BGM) built from the scan data. The data structure is not only used in this
study to determine BSP, but also employed to solve and animate the motion data by
an optimized inverse kinematics (IK) [10, 11].
A motivation on the field of biomechanics is to analyze the dynamics of balance.
Vukobratovic and Juricic [12] applied the position of the resultant ground reaction
force to legged machine control. They defined how the position can be computed from
legged system state and mass distribution, and named the position as zero moment
point (ZMP). Burnfield [13] found that a short period of unbalance occurs during
normal human walking as the weight of the body is transferred from the rear leg to
the forward leg. Popovic et al. [14] discussed the correlations of the three important
ground reference points: ZMP, foot rotation indicator (FRI) and centroidal moment
pivot (CMP) using a realistic human model and gait data measured from ten subjects.
Firnami and Park [15] proposed a theoretical method for analyzing the whole body
dynamic based on classic mechanical principles. The proposed method quantifies the
imbalance state by means of the distance between the center of pressure (CoP) and
the ZMP. Based on the dynamic balance, the foot reaction force for single contact
is easy to compute. However, the foot reactions are indeterminate in double support
phase [16]. Load sharing functions are proposed to describe a complete gait and
deal with the indeterminacy of the reaction loads [17]. The method provided logical
results, but it was specific to the given task.
This study proposes a method for determining the dynamic parameters of body
motion using a 3D whole human model and an approach based on the screw method
for tracking the ZMP of the body during motion. The validity of the proposed
approach is demonstrated by evaluating the whole body dynamics over the course
of a 25 s sequence of continuous motions performed by a professional martial arts
practitioner. According to the auto-calculated BSP and ZMP, the foot reactions in
double support phase are derived using the screw theory. Under balance conditions,
the simulation results show that two parameters govern the foot reactions, and can
be illustrated by the geometric of the cylindroid. With the assumption of minimum
internal mutual moment generated by the two feet reactions, an optimal solution can
be obtained. The optimal solution signifies that the actor can perform the movement
in an efficient way with less body internal strain.
9 Foot Reaction Analysis of Whole Body Dynamic via Screw Theory 105
9.2 Methods
9.2.1 3D Whole Body Model
The present study constructs a point-structure BGM to display the motion data. Model
geometric parameters of each link can be estimated using the structured points [11].
Based on Green Theory, the volume and the centroid of each segment are integrated
from the tetrahedrons constructed by the mesh points. The density of each segment
in Ref. [18] is obtained for computing the mass. Furthermore, the MOI and principle
axes can be calculated according to Ref. [19]. As shown in Fig. 9.1, a body kinematic
model (BKM) is generated from the BGM to describe the body motion by means of
joint angles. In the BGM, no joint constraint is enforced, i.e. six degrees of freedom
(DOF) exist between adjacent links. However, in the BKM, the jointed links have
less than four DOFs. Therefore, an IK procedure is performed to determine the
appropriate joint angles [10].
9.2.2 Dynamic Parameters of Whole Body
To calculate the dynamic parameters of the whole body, it is necessary to integrate

the geometric parameters of each link in the model. The mass of the whole body,
M, is computed by summing the masses of all the links, as shown in Eq. (9.1), in
which mi represents the mass of the ith link. The centroid of the whole body, G, is
calculated as shown in Eq. (9.2). Note that the centroid G is nominated as the origin
of the dynamic frame of the whole body, and pi is the centroid of the ith link.
(2,1~4) (1,1~4)
(2,5~7) (1,5~7)
Link
(2,8~9) (0,5~8)
(1,8~9)
Joint (0,1~4)
(2,10~11) (1,10~11)
(2,12) (0,0) (1,12)
End-effector (4,1~3) (3,1~3)
(4,4~5) (3,4~5)
(4,6~7) (3,6~7)
: Link, : Joint(m, i), :Base Frame,

m: Chain ID, i: Joint axis ID
Fig. 9.1 Segments of the BGM and joint assignment of the BKM

n
M= mi (9.1)
i=1

n
pi m i
i=1
G= (9.2)
M
The principal axes of the whole body are determined from Eq. (9.3) in accordance
with matrix transformation theory and the parallel theorem [20]. Note that Ri is the
rotation of the ith link with respect to the world coordinate frame, it is the principal
MOI of the ith link, and di is the skew symmetric matrix associated with the relative
position vector. Furthermore, the principal MOI of the whole body are eigenvalues
of I CoM . Moreover, the eigenvectors of I CoM are the principle axes of the dynamic
frame of the whole body.

[ICoM ] = [Ri ][Ii ][Ri ]T − m i [di ]2 (9.3)
After obtaining the dynamic frame of the whole body, O DB , the fixed angle method
[21] is used to decompose O DB into six generalized transformations, i.e., three trans-
lation components, Tx , Ty and Tz , and three rotation components, Rx , Ry and Rz , as
shown in Eq. (9.4):
⎡ ⎤
n x (t) sx (t) ax (t) px (t)
⎢ n y (t) s y (t) ax (t) p y (t) ⎥
O
D B (t) = Rz (φ)R y (θ )Rx (ψ)Tz (z)T y (y)Tx (x) = ⎢
⎣ n z (t) sz (t) ax (t) pz (t) ⎦
⎥
0 0 0 1
(9.4)
The fixed angles, i.e., yaw(ψ), pitch(θ ) and roll(φ), are used to define the rotations
of the dynamic frame about the x-, y- and z-axes of the world frame, respectively:
⎡ ⎤ ⎡ ⎤
n x (t) sx (t) ax (t) CφCθ Cφ Sθ Sψ − SφCψ Cφ SθCψ + Sφ Sψ
⎢ ⎥ ⎢ ⎥
O
R B (t) = ⎣ n y (t) s y (t) a y (t) ⎦ = ⎣ SφCθ Sφ Sθ Sψ + CφCψ Sφ SθCψ − Cφ Sψ ⎦ (9.5)
n z (t) sz (t) az (t) −Sθ Cθ Sψ CθCψ
The vectors describing the position p(t), fixed angle (t), linear velocity v(t) and
angular velocity (t) of the dynamic frame relative to the world frame are defined
respectively as
p(t) : [ px (t), p y (t), pz (t)]T (9.6)
(t) : [ψ(t), θ (t), φ(t)]T (9.7)
v(t) : [vx (t), v y (t), vz (t)]T (9.8)

(t) : [ωx (t), ω y (t), ωz (t)]T (9.9)
The linear velocity of the whole body is equal to the time derivative of the CoM
position. Meanwhile, the relationship between the differentiation of the fixed angles
φ’(t) and the angular velocity ω(t) can be found by equating the nonzero elements
in the matrices given in Eq. (9.10) [22], i.e.
⎡ ⎤ ⎡ ⎤⎡ ⎤
ωx (t) CφCθ −Sφ 0 ψ̇
⎣ ω y (t) ⎦ = ⎣ SφCθ Cφ 0 ⎦⎣ θ̇ ⎦ (9.10)
ωz (t) −Sθ 0 1 φ̇
In accordance with the moving coordinate frame concept, the linear and angu-
lar velocities of the whole body can be obtained from the velocity vector q̇(t) =
T
Ṗx , Ṗy , Ṗz , ψ̇, θ̇ , φ̇ as follows:

v(t)
= [J(q)]q̇(t) = [J1 (q), J2 (q), . . . , J6 (q)]q̇(t) (9.11)
ω(t)
where J(q) is a Jacobian matrix whose ith column vector Ji (q) has the form [23]
⎧
⎪
⎪ zi−1 × (p − pi−1 )
⎨ if ith component is rotational
zi−1
Ji (q) = (9.12)
⎪
⎪ z
⎩ i−1 if ith component is translational
0
Note that pi−1 is the position of the origin of the (i−1)th coordinate frame with
respect to the world frame, zi−1 is the unit vector along the axis of motion of ith
component, and p is the position of the whole body with respect to the world frame.
Thus, Eq. (9.11) can be expressed as
⎡ ⎤ ⎡ ⎤⎡ ⎤
vx (t) 100 0 0 0 ṗx (t)
⎢ v (t) ⎥ ⎢ 0 1 0 0 0 0⎥ ⎢ ⎥
⎢ y ⎥ ⎢ ⎥⎢ ṗ y (t) ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ vz (t) ⎥ ⎢ 0 0 1 0 0 0 ⎥⎢ ṗz (t) ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ (9.13)
⎢ ωx (t) ⎥ ⎢ 0 0 0 CφCθ −Sφ 0 ⎥⎢ ψ̇(t) ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ ω y (t) ⎦ ⎣ 0 0 0 SφCθ Cφ 0 ⎦⎣ θ̇(t) ⎦
ωz (t) 0 0 0 −Sθ 0 1 φ̇(t)
The acceleration a(t) and angular acceleration α(t) of the whole body can be
obtained by taking the time derivative of the velocity vector given in Eq. (9.11), i.e.,

a(t)
= J̇(q, q̇)q̇(t) + J(q)q̈(t) (9.14)
α(t)
According to Newton’s second law, the inertial force Fin on the whole body can be
computed as shown in Eq. (9.15). Moreover, the inertial moment M in on the whole
body is formulated as shown in Eq. (9.16).
F in =Ma (9.15)
M in = I CoM α + × I CoM (16)
9.2.3 Finding ZMP via Screw Method
Two vectors, FB and M in , form a wrench screw (FB , M in ), which exerts a body force
FB and inertial moment M in at the CoM, point G as shown in Fig. 9.2. The body
force FB represents the sum of the inertial force Fin and the gravity force. FB and
the resultant ground reaction force, FR , are always opposite and equal. The wrench
screw (FB , M in ) can be written in Plücker coordinate as $0 ’: (L’, M’, N’, P*’, Q*’,
R*’), and normalized by L’2 +M’2 +N’2 = 1. By definition, the screw $0 ’ can be
recognized as moving the origin to the point G: (xG , yG , zG ). Therefore, to express
the screw $0 ’ about the origin of the world coordinate, the screw $0 ’ is written as
a new screw $0 : (L, M, N; P*, Q*, R*). The force components (L, M, N) are the
same as (L’, M’, N’), and the moment components (P*, Q*, R*) are computed by
Eq. (9.17). Let the intersection of the screw axis $0 and plane z = 0 be denoted as
point P. In this study, the point P is defined as the zero moment point (ZMP). The
vector r indicating the ZMP position is computed using screw theory. Then the pitch
h of the screw $0 can be calculated from Eq. (9.18). Using (19), (P, Q, R) the screw
axis can be obtained.
⎧
⎪
⎪ yG z G
⎪ P = P + M N

⎪ ∗ ∗
⎪
⎪
⎪
⎨
z x
Q ∗ = Q ∗ + G G

(9.17)
⎪
⎪ N L
⎪
⎪
⎪
⎪ x G yG
⎩ R = R +
⎪ ∗ ∗
L M
h = L P ∗ + M Q∗ + N R∗ (9.18)

(P, Q, R) = P ∗ , Q ∗ , R ∗ − h(L , M, N ) (9.19)
Furthermore, x and y can be calculated by

Fig. 9.2 The body screw

formed by the total body
$0=$0
force FB and the inertial
moment M in
GMi
n
FB
O r P
FR
Fig. 9.3 The Contact points
and the support polygon
⎧
⎪ y 0
⎪
⎨ P= ⇒y= P
N
M N (9.20)
⎪ 0 x
⎪
⎩ Q = ⇒ x = −Q
N L N
9.2.4 Foot Reaction Analysis
Foot reaction forces and moments are assumed to act on the centroids calculated
by projecting contact points of soles on the ground. In the BGM, each foot has 38
structure points on the sole. Figure 9.3 illustrates that a height threshold, H, is set to
judge if any structure points of the soles touch the ground or not. The centroids of
the left foot and the right foot are drawn in red point and green point respectively. In
addition, all contact points projected on the ground construct a convex hull, which
is defined as a support polygon.
Fig. 9.4 Single foot support
Mz rs :(xsa, ys , zs )
Mx
P
My Pc
In single support phase, a body screw $0 is balanced by an opposite screw −$0

passing through the ZMP at P as shown in Fig. 9.4. Since the foot reaction force
and moment are assumed to act on the centroid Pc , the foot reaction force is FR and
the foot reaction moment is the vector sum of the moment acting on P and an extra
moment produced by rs × FR .
In double support phase, the total force/moment exerts on a body can be repre-
sented by a body screw $0 , which is balanced by two screws ($1 , $2 ) passing through
two foot contacts cantered at P1 and P2 as shown in Fig. 9.5. To simplify the con-
straint equations, a new coordinate system is constructed by locating the origin on
the ZMP position (x0 , y0 , z0 ) and putting the z axis along the body screw $0 .
The Plücker coordinates of screw $0 now is [0, 0, 1, 0, 0, h0 ], where h0 can be
computed by (18). $1 is [L 1 , M 1 , N 1 ; P1 *, Q1 *, R1 *], and $2 is [L 2 , M 2 , N 2 ; P2 *, Q2 *,
R2 *], respectively. By definition, the two screws $1 and $2 also can be written as $1 :
[L 1 , M 1 , N 1 , P1 + h1 L 1 , Q1 + h1 M 1 , R1 + h1 N 1 ] and $2 : [L 2 , M 2 , N 2 , P2 + h2 L 2 , Q2 +
h2 M 2 , R2 + h2 N 2 ]. In which, [L i , M i , N i ] are normalized direction cosines, which
follow the constraints:
L 21 + M12 + N12 = 1 (9.21)
L 22 + M22 + N22 = 1 (9.22)
The problem is to find the two screws ($1 , $2 ) that balance the body wrench screw
$0 . There are 14 unknowns (L 1 , M 1 , N 1 , P1 , Q1 , R1 , h1 , L 2 , M 2 , N 2 , P2 , Q2 , R2 , h2 )
$
Central $
axis $3
$
$ $
$
Cylindroid
P1 (x1 , y1 , z1 )
O
P2 (x2 , y2 , z2 )
O
Fig. 9.5 Geometric characteristic of a cylindroid: z3 and a direction ratio tanθ 3 = M 3 /L 3 are two
parameters characterizing $3
in total. Furthermore, the two reaction screw axes passing through the contact points
P1 (x1 , y1 , z1 ) and P2 (x2 , y2 , z2 ), so the dual parts of the screw axes can be written as

y1 z 1

P1 = = y1 N1 − z 1 M1 (9.23)
M1 N 1

z 1 x1
Q1 = = z 1 L 1 − x 1 N1 (9.24)
N1 L 1

x y
R1 = 1 1 = x1 M1 − y1 L 1 (9.25)
L 1 M1

y2 z 2

P2 = = y2 N2 − z 2 M2 (9.26)
M2 N 2

z x
Q 2 = 2 2 = z 2 L 2 − x2 N2 (9.27)
N2 L 2

x2 y2

R2 = = x2 M2 − y2 L 2 (9.28)
L 2 M2
Equations (9.21)–(9.28) provide eight constraint equations. In addition, a matrix

A is formed by the three screws in the Plücker coordinates.
⎡ ⎤
0 0 1 0 0 h0
A : ⎣ L 1 M1 N1 P1∗ Q ∗1 R1∗ ⎦ (9.29)
L 2 M2 N2 P2∗ Q ∗2 R2∗
Since the three screws are linearly dependent, any determinant of 3 × 3 sub-matrix
should be zero.
From columns 1, 3, 6:
L 2 (R1 + h 1 N1 ) + h 0 L 1 N2 − h 0 N1 L 2 − L 1 (R2 + h 2 N2 ) = 0 (9.30)
(P1 + h 1 L 1 )(Q 2 + h 2 M2 ) − (Q 1 + h 1 M1 )(P2 + h 2 L 2 ) = 0 (9.31)
(P1 + h 1 L 1 )M2 − (P2 + h 2 L 2 )M1 = 0 (9.32)
(P1 + h 1 L 1 )(R2 + h2N 2) + h 0 N1 (P2 + h 2 L 2 ) − h 0 N2 (P1 + h 1 L 1 )

− (P2 + h 2 L 2 )(R1 + h 1 N1 ) = 0 (9.33)
There are 12 constraint equations from Eq. (9.21) to (9.33) in total, but with 14
unknown parameters. Consequently, the double support phase is an indeterminate
nature. Since the three screws ($0 , $1 , $2 ) are linear dependent, they degenerates into
a two-system. In addition, the geometry of a general two-system forms a cylindroid.
A centre axis ($3 = [L 3 , M 3 , N 3 , P3 , Q3 , R3 ]), shown in Fig. 9.5, intersects all the
screw axes in the two-system orthogonally [24]. The orthogonality of two lines ($,
$’) is LL’ + MM’ + NN’ = 0, and the condition of intersection is LP’ + MQ’ + NR’
+ L’P + M’Q + N’R = 0. Accordingly, $0 intersect $3 : R3 = 0, and $0 ⊥$3 : N 3 = 0,
so that $3 = [L 3 , M 3 , 0, P3 , Q3 , 0]. Since $0 lies along the z-axis, $3 lies in a plane π
perpendicular to the axis of $0 . The plane π has the equation of z3 = 0. Then z3 and
the direction ratio tanθ 3 = M 3 /L 3 are two parameters characterizing $3 , (Fig. 9.5).
If z3 and θ3 are specified, the two balancing screws $1 and $2 can be determined.
Moreover, the $3 becomes [L 3 , M 3 , 0; −z3 M 3 , z3 L 3 , 0]. Equations (9.34)–(9.37)
describe the conditions of ($1, $3) and ($2, $3).
For $1 intersect $3 : −z3 L 1 M 3 + z3 M 1 L 3 + L 3 P1 + M 3 Q1 = 0, we got
−z 3 L 1 tanθ3 + z 3 M1 + P1 + Q 1 tanθ3 = 0 (9.34)

For $1 ⊥$3 : L 1 L 3 + M 1 M 3 = 0, we got
L 1 + M1 tanθ3 = 0 (9.35)
For $2 intersect $3 : −z3 L 2 M 3 + z3 M 2 L 3 + L 3 P2 + M 3 Q2 = 0, we got
−z 3 L 2 tanθ3 + z 3 M2 + P2 + Q 2 tanθ3 = 0 (9.36)
For $2 ⊥$3 : L 2 L 3 + M 2 M 3 = 0, we got
L 2 + M2 tanθ3 = 0 (9.37)
The mutual moment, the twofold of the virtual coefficient by Ball [25], between
screws ($1 , $2 ) should be minimum for an optimal solution of minimum internal
moment generated by the two screws through the foot contacts. That is:
Min{|L 1 P2∗ + M1 Q ∗2 + N1 R2∗ + L 2 P1∗ + M2 Q ∗1 + N2 R1∗ |} (9.38)
After obtaining a set of optimal variables by the function “NMinimize”

in Mathe-
matica
[26], two screws with magnitudes (F 1 and F 2 ) are solved by F i N i = 0 and
F i Ri * = 0 as shown in Eqs. (9.39) and (9.40). Note that F 0 is the magnitude of
the $0 , and R* component of $0 in the Plücker coordinate equal to h0 .
F0 N0 + F1 N1 + F2 N2 = 0 (9.39)
F0 h 0 + F1 R1∗ + F2 R2∗ = 0 (9.40)
9.3 Results
Figure 9.6 shows the variation of the resultant force and resultant moment acting on
the whole body over the 752 timeframes of the martial arts sequence. The results
demonstrate that the magnitudes of the forces and moments are reasonable, in accor-
dance with real conditions. In the dynamic analysis, the subject takes a continuous
motion, and a snapshot is shown in Table 9.1. His foot-contact points are centroids of
sole structure points below a threshold 30 mm. The orange, red, green and blue dash
lines represent body screw $0 , left foot-contact screw $1 , right foot-contact screw
$2 and common normal line $3 , respectively. According to the new coordinate, the
known conditions are given as follows:
ZMP : (0, 0, 0).

$0 : [0, 0, 1, 0, 0, 0.002639].
Table 9.1 Comparison of two variables are free and both constrained
Case 1: z3 and tanθ3 both are not constrained Case 2: z3 = 0.4417 m, and
tanθ3 = 10
Mutual moment 2Vc 6.3755 × 10−11 m 0.09486 m

z3 0.4417 m 0.4417 m
tanθ3 1.4102 10
F1 208.913 N 286.427 N
F2 365.082 N 327.964 N
|F1 | + |F2 | 573.995 N 614.391 N
h1 −0.539777 −0.2153
h2 0.18258 0.08867
$1 [−0.3504, 0.2484, −0.9031; [−0.4993, 0.04993, −
0.0092, −0.2391, −0.0693] 0.8650; 0.1094, −0.2337,
0.1724]
$2 [0.2005, −0.1422, −0.9693; − [0.4361, −0.04361, −
0.1501, 0.2395, −0.0661] 0.8989; −0.0955, 0.2041, −
0.1549]
F0 $0 [0, 0, 542.549; 0, 0, 1.43189] [0, 0, 542.549, 0, 0, 1.43189]
F1 $1 [−73.199, 51.900, −188.664; [−143.013, 14.301, −
41.448, −77.969, 87.341] 247.756; 31.339, −66.927,
49.370]
F2 $2 [73.19, −51.9, −353.885; − [143.013, −14.301, −
41.4481, 77.9696, −88.7728] 294.793; −31.339, 66.927,
−50.802]

Fi $i [1.46 × 10−4 , −1.11 × 10−4 , [3.63 × 10−5 , −3.53 ×
0; −1.74 × 10−5 , 1.27 × 10−4 , 10−6 , 0; 3.18 × 10−5 , 7.32
0] × 10−6 , 0]
Distance of $1 & $2 0.4308 m 0.1994 m
2Vc |F1 ||F2 | 4.8626 ×10−6 0.09486 m
Fig. 9.6 Variation of inertial force/moment on whole body
P1 : (−0.2793, 0, −0.03742).
P2 : (0.2588, 0.1466, −0.05666).
The minimum mutual moment is 6.3755 × 10−11 m if common normal line $3 is

not constrained as shown in Case 1 of Table 9.1. However, as shown in Table 9.1,
Case 2 shows if both $3 and θ3 are constrained at z3 = 0.4417 m and tanθ3 = 10,
the optimal mutual moments goes higher to 0.09486 m. As illustrated in Table 9.2,
Cases 3 and 4 has only one variable constrained, i.e. $3 is located at z3 = 1 m and
z3 = −0.5 m, the optimal mutual moments are 1.11895 × 10−3 m and 6.13337 ×
10−5 m, respectively. It is conclude that, if $3 is given more degrees of freedom, the
objective function can reach a smaller value. Please notice that the unit of mutual
moment is [meter] because all the screws have been normalized; the force unit is in
[Newton] which is the intensity of the screws.
Comparing data in the four cases, if $1 and $2 are more parallel to each other, the
resultant reaction force acting on feet |F 1 | + |F 2 | will decrease. From L, M, P*, and
Q* components of F 1 $1 and F 2 $2 , the internal force and moment can be obtained.
The components of F 1 $1 and F 2 $2 have nearly the same magnitude but different
signs. Thus, the internal forces and moments are always balanced.
9.4 Conclusions
This study has proposed a method for determining the dynamic parameters of human
motion using a 3D whole human body model. In the proposed approach the joint angle
data are derived using an IK procedure based on motion data captured from a real
human. Therefore, any errors in the captured data will cause the solved joint angles
to be unsmooth. To avoid discontinuous joint angles, the IK results are smoothed
by a cubical smoothing algorithm with five-point approximation before calculating
the dynamic parameters of the whole body. It has been shown that the dynamic
parameters are in good agreement with the inferences based on realistic conditions.
In the double support phase, there are ∞2 of solutions available. Nothing to
constrain the actor to apply the forces follow the optimal solution. Besides, the foot
reactions are assumed to act through the centroid of foot contact, which may cause
errors due to the differences between a reality and the assumption. However, a well-
Table 9.2 Cases of only one variable is constrained

Case 3: z3 = 1 m Case 4: z3 = −0.5 m
Mutual moment 2Vc 1.1190 × 10−3 m 6.1334 × 10−5 m

z3 1m −0.5 m
tanθ3 5.4043 −3.5157
F1 251.925 N 331.829 N
F2 305.885 N 305.125 N
|F1 | + |F2 | 557.81 N 636.954 N
h1 −0.4091 0.1134
h2 0.2858 −0.1258
$1 [−0.2517, 0.0465, −0.9667; [ 0.4830, 0.1374, −0.8648;
0.0017, −0.2606, −0.0130] 0.0051, −0.2596, −0.0383]
$2 [ 0.2073, −0.03835, −0.9776; [−0.5253, −0.1494, −0.8377;
−0.1455, 0.2413, −0.0403] −0.1313, 0.2466, 0.0383]
F0 $0 [ 0, 0, 542.549, 0, 0, 1.4318] [ 0, 0, 542.549, 0, 0, 1.432]
F1 $1 [−63.395, 11.731, −243.536; [160.286, 45.592, −286.95;
26.374, −70.446, 96.354] 19.885, −80.971, −45.278]
F2 $2 [63.395, −11.731, −299.013; [−160.286, −45.591, −
−26.374, 70.446, −97.786] 255.599; −19.885, 80.971,
43.846]

Fi $i [−1.13 × 10−5 , −2.75 × [3.40 × 10−5 , 9.93 × 10−6 , 0;
10−6 , 0; −2.99 × 10−6 , −8.02 −7.26 × 10−6 , −1.34 × 10−5 ,
× 10−6 , 0] 0]
Distance of $1 & $2 0.2420 m 0.006239 m
2Vc |F1 ||F2 | 86.226 6.2100
trained athlete should be able to take action with minimum internal moment to reduce
strain and fatigue during the motion. The optimal foot reaction condition may be used
as a reference to evaluate the performance of an athlete. Theoretically speaking, the
proposed method provides a systematic approach to analyse the foot reaction, and
the result shows reasonable solutions indeed. Moreover, the foot reaction can be
observed easily by the screw geometry of a cylindroid by varying the two parameters
(z3 , θ3 ).
In the future, the optimal foot reaction calculated by the proposed method will
combine inverse dynamic analysis to evaluate joint force/moment of the body model,
and the calculated results would be compared to those from the literatures and the
force plate experiments, In addition, a human motion sequence will be retargeted to
humanoid robots to perform human tasks, and the robot dynamics will be analysed
by the proposed method.
Acknowledgements The financial support from Ministry of Science and Technology of Taiwan
through the grant number: MOST 105-2221-E-006-080, is greatly appreciative. The authors want
to express our great thanks to Mr. Tseng, Chi-Juang, a grand master of Chinese Martial Art, who
contributed precious actions in video recording that make this study possible.
References
1. Yeadon, M.R., Morlock, M.: The appropriate use of regression equations for the estimation of
segmental inertia parameters. J. Biomech. 22(6), 683–689 (1989)
2. De Leva, P.: Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters. J. Biomech.
29(9), 1223–1230 (1996)
3. Pearsall, D.J., Reid, J.G., Ross, R.: Inertial properties of the human trunk of males determined
from magnetic resonance imaging. Ann. Biomed. Eng. 22(6), 692–706 (1994)
4. Lee, M.K., Koh, M., Fang, A.C., Le, S.N., Balasekaran, G.: Estimation of body segment
parameters using dual energy absorptiometry and 3-D exterior geometry. In: 13th International
conference on biomedical engineering, (pp. 1777–1780). Springer, Berlin, Heidelberg (2009)
5. Norton, J., Donaldson, N., Dekker, L.: 3D whole body scanning to determine mass properties
of legs. J. Biomech. 35(1), 81–86 (2002)
6. Ensminger, G.J., Robertson, R.N., Cooper, R.A.: A model for determining 3-D upper extremity
net joint forces and moments during wheelchair propulsion. In: IEEE 17th annual conference
on engineering in medicine and biology society, vol. 2, pp. 1179–1180 (1995)
7. Liu, T., Inoue, Y., Shibata, K., Shiojima, K.: Three-dimensional lower limb kinematic and
kinetic analysis based on a wireless sensor system. In: 2011 IEEE international conference on
robotics and automation (ICRA), pp. 842–847
8. Nakamura, Y., Yamane, K., Fujita, Y., Suzuki, I. Somatosensory computation for man-machine
interface from motion-capture data and musculoskeletal human model. IEEE Trans Robot
21(1), 58–66 (2005)
9. Tsai, M.J., Fang, J.J.: U.S. Patent No. 7,218,752. U.S. Patent and Trademark Office, Washing-
ton, DC (2007)
10. Tsai, M.J., Lung, H.Y.: Two-phase optimized inverse kinematics for motion replication of real
human models. J. Chin. Inst. Eng. 37(7), 899–914 (2014)
11. Tsai, M.J., Lee, A., Lee, H.W.: Automatic full body inverse dynamic analysis based on person-
alized body model and MoCap data. Adv. Mech., Robot. Des. Educ. Res., 305–322. Springer
International Publishing (2013)
12. Vukobratovic, M., Juricic, D.: Contribution to the synthesis of biped gait. IEEE Trans. Biomed.
Eng. 1, 1–6 (1969)
13. Burnfield, M.: Gait analysis: normal and pathological function. J. Sport. Sci. Med. 9, 353 (2010)
14. Popovic, M.B., Goswami, A., Herr, H.: Ground reference points in legged locomotion: defi-
nitions, biological trajectories and control implications. Int. J. Robot. Res. 24(12), 1013–1032
(2005)
15. Firmani, F., Park, E.J.: Theoretical analysis of the state of balance in bipedal walking. J.
Biomech. Eng. 135(4), 041003 (2013)
16. Ren, L., Jones, R.K., Howard, D.: Predictive modelling of human walking over a complete gait
cycle. J. Biomech. 40(7), 1567–1574 (2007)
17. Ren, L., Jones, R.K., Howard, D.: Whole body inverse dynamics over a complete gait cycle
based only on measured kinematics. J. Biomech. 41(12), 2750–2759 (2008)
18. Zatsiorsky, V.M.: Kinetics of Human Motion. Human Kinetics, Champaign, IL (2002)
19. Chang, C.H.: Numerical simulations for a 3D system composed of polyhedral blocks-dissection
of polyhedral blocks. Master thesis of department of civil engineering, National Central Uni-
versity, Zhongli, Taiwan (2006)
20. Kane, T.R., Levinson, D.A.: Dynamics, Theory and Applications. McGraw Hill, New York
(1985)
21. Carretero, J.A., Podhorodeski, R.P., Nahon, M.A., Gosselin, C.M.: Kinematic analysis and
optimization of a new three degree-of-freedom spatial parallel manipulator. J. Mech. Des.
122(1), 17–24 (2000)
22. Fu, K.S., Gonzalez, R.C., Lee, C.G.: Robotics, pp. 163–189. McGraw-Hill, New York (1987)
23. Whitney, D.E.: The mathematics of coordinated control of prosthetic arms and manipulators. J.
Dyn. Syst., Meas., Control. 94(4), 303–309 (1972)
24. Hunt, K.: Kinematic Geometry of Mechanisms. Clarendon Press, Oxford (1978)
25. Phillips, J.: Freedom in Machinery: Screw Theory Exemplified, vol. 2. Cambridge University
Press, Cambridge (1990)
26. http://reference.wolfram.com/language/ref/NMinimize.html
Chapter 10
Comparison of Motion/Force
Transmissibility in a 3-SPR Parallel
Manipulator and a 6-SPS Equivalent
Mechanism
M. Russo, M. Ceccarelli and Y. Takeda
Abstract The analysis of motion/force transmissibility and constrainability is a key

point in the design of parallel manipulators. Even if recent advances led to a gen-
eral formulation for a transmission index, the application to lower-mobility parallel
manipulators requires additional investigation of the constraints of the system. This
paper applies a definition of the output transmission index evaluated from the pres-
sure angle of the mechanism to a particular 3-SPR (Spherical-Prismatic-Revolute
Joint) parallel manipulator, which is characterized by a moving platform consisting
of a single point where all the three open-loop kinematic chains converge. The Out-
put Transmission Index OTI3 of the mechanism is computed through the pressure
angle definition. However, this index can only take into account the motion/force
transmissibility of the mechanism acting as a pointing device, without consider-
ing the rotational motion of the moving platform. Therefore, an equivalent 6-SPS
(Spherical-Prismatic-Spherical Joint) parallel manipulator is analyzed and its Output
Transmission Index OTI6 is evaluated. The results are then compared and discussed,
in order to highlight the similarities and differences between the two different for-
mulations.
Keywords Transmission index · Parallel mechanisms · Mechanism design ·

Kinematics
10.1 Introduction
Parallel mechanisms usually outperform serial mechanisms in accuracy, speed and

payload. However, they are characterized by a more complex structure, made of sev-
M. Russo (B) · M. Ceccarelli

Department of Civil and Mechanical Engineering, University of Cassino and South Latium, Via
G. Di Biasio 43, 03043 Cassino, FR, Italy
Y. Takeda
Department of Mechanical Engineering, Tokyo Institute of Technology, 2-12-1 Ookayama,
Meguro-Ku, Tokyo 152-8550, Japan

120 M. Russo et al.
eral parallel chains called limbs that connect a fixed platform to a moving one. The
analysis of their motion and force transmission and constraints is therefore extremely
important for their design. Among all parallel manipulators, lower-mobility mech-
anisms are characterized by less than 6-Degrees-of-Freedom and they usually have
a larger workspace but lower motion capability than 6-Degrees-of-Freedom parallel
manipulators [1–3]. The most common lower-mobility manipulators are 3-Degrees-
of-Freedom ones. In order to characterize their behaviour, a constraint analysis is
needed along the Forward or Inverse Kinematics of the system, because, as described
by Zlatanov in [4], the motion of mechanisms with reduced degrees-of-freedom
can be limited by constraint singularities. These singularities cannot be detected by
analysing the input-output relation only [5], but they require additional geometrical
considerations, as in [6] for a 3-UPU (Universal-Prismatic-Universal Joint) manip-
ulator and in [7] with a general approach, or the complete 6 × 6 Jacobian matrix of
the manipulator, as shown in [8] for the 3-UPU and the 3-RPS parallel manipulators
and in [9] for the Linear Delta robot.
Most of the indices that were proposed to characterize their performance, such
as the condition number or the manipulability index, depend on the Jacobian of
the system [1]. Thus, they evaluate only the kinematics of the mechanism. In order
to consider dynamic behaviour, the concept of transmission performance has been
introduced [10], but research groups have defined a variety of Transmission Indices
(TI) as global performance parameters. As documented in [11], Takeda and Funabashi
considered the virtual power transmitted from input to output link as a measure of
TI and related it to the pressure angle of the manipulator when all input links but one
are fixed [12]. This approach allowed the TI to be also used in singularity analysis
[13]. A general formulation was proposed by Chen and Angeles in [14], who define a
Global Transmission Index (GTI), which is based on the virtual coefficient between
the transmission wrench screw and the output twist screw. The GTI has been applied
to redundant parallel manipulators, as for example in [15], and to workspace analysis
[16]. In particular, paper [16] defined the Input and Output Transmission Indices (ITI
and OTI) as the power coefficients that describe the force transmission from the fixed
base to each limb and from each limb to the moving platform respectively. Other TI
formulations are shown in [17, 18]. The pressure angle approach has been adopted
by Brinker et al. in [19] for the Delta Parallel Robot, and by Takeda and Liang in
[20] for a lower-mobility 3-PSR mechanism.
This paper presents the constraint and force transmission analysis of a novel 3-
Degrees-of-Freedom spatial mechanism, by following the approach in [19, 20]. The
manipulator is characterized by the 3-SPR architecture introduced in [21–23], with a
moving platform that mathematically collapses to a single point. Constructive details
of the mechanisms are documented in patent applications [24, 25]. A closed-form
expression of the output transmission index for the 3-SPR manipulator is evaluated
by using the pressure angle as presented in [11]. However, this model only allows
for the computation of the OTI 3 index, which does not consider the rotational output
motion of the mechanism. For this reason, only the performance of the mechanism
as a pointing device is evaluated by this problem formulation. Therefore, following
the approach presented in [20], each limb of the mechanism is modelled as a 2-
10 Comparison of Motion/Force Transmissibility in a 3-SPR … 121
SPS architecture where the two linear motors always share the same actuation. The
transmission index evaluated by using the pressure angle definition for this equivalent
mechanism is the OTI 6 index of the manipulator, since it considers all the possible
motion modes of the end-effector, including its rotational motion. The last section of
this work compares the transmission performance of the proposed 3-SPR manipulator
analysed as a pointing device to the one of the full 6-SPS equivalent mechanism,
by comparing the OTI 3 and OTI 6 results obtained from the two different modelling
options.
10.2 Methodology
The proposed 3-SPR mechanism is shown in Fig. 10.1a with its main design param-
eters. The mechanism is characterized by three SPR limbs, and each limb is driven
by a linear actuator in the prismatic joint that controls the i-th limb length d i . The
three spherical joints on the base are placed at the vertices of an equilateral tri-
angle inscribed in a circumference with radius a, while the revolute joints on the
end-effector are centered in point H. By considering each spherical joint on the base
platform as equivalent to three intersecting non-coplanar revolute joints, it is possible
to apply the theory of reciprocal screws for the Jacobian analysis [3, 8]. In particular,
the instantaneous twist of the moving platform ξ H can be expressed as the linear
combinations of the instantaneous twists of the five 1-DoF joints of the ith limb, as
Fig. 10.1 Kinematic diagrams of the proposed manipulator with its main design parameters: a 3-
SPR mechanism; b equivalent 6-SPS mechanism
122 M. Russo et al.

ω
ξH = = θ̇ 1,i ξ 1,i + θ̇ 2,i ξ 2,i + θ̇ 3,i ξ 3,i + ḋ i ξ 4,i + θ̇ 5,i ξ 5,i (10.1)
v
where θ j,i is the intensity of the motion associated with the jth joint of the ith limb and
ξ j,i is the unit screw associated to the same joint. The twist of the moving platform is
defined by ω, which is the angular velocity of the moving platform, and by v, which
is the linear velocity of the end-effector point H shown in Fig. 10.1a. The XYZ
coordinate system fixed on the base platform in Fig. 10.1a is assumed as reference
frame. As described in [3], it is possible to identify a pair of unit screws that are
reciprocal to all the joint screws and to all the joint screws except for the actuated
joint in the ith limb, respectively. It is then possible to perform the product of both
sides of Eq. (10.1) with the reciprocal screws, producing an equation that can be
written in matrix form as
J X,i ξ H = J q,i q̇ (10.2)
where q̇ represents the velocity vector of the actuated joints and is defined as
(0, 0, 0, q̇1 , q̇2 , q̇3 )T . Equation (10.2) can be written 3 times, once for each limb,
and the resulting equations can be assembled in the form
q̇ = Jξ H (10.3)
where J is the overall Jacobian matrix of the system and can be expressed as
⎡ T ⎤
−d 1 × s4,1 T
s4,1
⎢ T ⎥
⎢ −d 2 × s4,2 T
s4,2 ⎥
⎢ ⎥
⎢ −d 3 × s4,3 T T
s4,3 ⎥
J =⎢
⎢ −d × s T
T
⎥.
⎥ (10.4)
⎢ 1 5,1
T
s5,1 ⎥
⎢ ⎥
⎣ −d 2 × s5,2 T
s5,2 ⎦
T
−d 3 × s5,3 T
s5,3
where sj,i is the unit vector related to the jth joint of the ith limb and it is defined as
the direction of the allowed motion for the prismatic joints or as the direction of the
rotation axis for the revolute joints, as shown in Fig. 10.1a. Vector d i is the position
vector of the end-effector point with regards to the spherical joint of the ith limb.
In order to obtain a closed-form expression for the Jacobian matrix in Eq. (10.4),
the inverse and direct kinematics of the structure are needed. Limb vector d i can be
expressed as
d i = r − ai (10.5)
where r is the position of the end-effector point H and ai represents the position of
the spherical joint belonging to the ith limb, as shown in Fig. 10.1a. Therefore, these
vectors can be written as
⎛ ⎞ ⎛ √ ⎞ ⎛√ ⎞ ⎛ ⎞
x − 23a 3a 0
⎜ ⎟ ⎜ 2 ⎟
⎝
r = y ; a1 = ⎝ − 2 ⎠; a2 = ⎝ − 2 ⎠; a3 = a ⎠.
⎠ a a ⎝ (10.6)
z 0 0 0
Once the limb vectors d i are defined, it is possible to evaluate unit vectors s4,i
from
di
s4,i = (10.7)
di
as
⎛ √ ⎞ ⎛ √ ⎞ ⎛ ⎞
x+ 23a x− 23a x
⎜ ⎟ ⎜ ⎟ d3
⎟; s4,3 = ⎜ ⎟
d1 d2
s4,1 = ⎜
⎝
y+ 2a ⎟; s4,2 = ⎜
⎠ ⎝
y+ 2a
⎠ ⎝
y−a
d3 ⎠ (10.8)
d1 d2
z z z
d1 d2 d3
Unit vectors s5,i can be computed thanks to geometrical considerations on the

constraints of the structure. In particular, unit vector s5,1 corresponds to unit vector
s5,2 and must be perpendicular to both limb vectors d 1 and d 2 . Thus, it is possible to
express it as
⎛ ⎞
0
d1 × d2 ⎜− z ⎟
⎜ ⎟
s5,1 = s5,2 = =⎜ z 2 +( y+ 2a )
2
⎟. (10.9)
d 1 × d 2 ⎝ y+ 2a ⎠

z 2 +( y+ 2a )
2
Unit vector s5,3 can be obtained in a similar way, since it is perpendicular to limb
vector d 3 and to unit vector s5,1 . Therefore, it is computed as
⎛ ⎞
−( y+ 2a )( y−a)−z 2

⎜ ⎟
[( y+ 2 )( y−a)+z2 ]2 +x 2 ( y+ 2a )2 +z2
a
⎜ ⎟
s5,1 × d 3 ⎜ x ( y+ 2 )
a ⎟
⎜ ⎟
s5,3 = =⎜ ⎟. (10.10)
s5,1 × d 3 ⎜ [( y+ 2a )( y−a)+z2 ]2 +x 2 ( y+ 2a )2 +z2 ⎟
⎜ ⎟
⎝ xz

⎠
[( y+ 2a )( y−a)+z2 ]2 +x 2 ( y+ 2a )2 +z2
Equations (10.8)–(10.10) solve the second column of the Jacobian expressed in

Eq. (10.4). The first three rows of the first column are easily solved as
124 M. Russo et al.
−d i × s4,i = 0 (10.11)
since the ith limb vector d i and its unit vector s4,i always share the same direction. The
missing rows of the first column of the overall Jacobian matrix of Eq. (10.4) can be
then solved. Furthermore, Eqs. (10.5)–(10.8) can be used to obtain an expression for
the Transmission Index by evaluating the pressure angle α if the rotational motion of
the moving platform is not considered. As suggested by Takeda and Funabashi [11],
it is possible to imagine the virtual motion of a single-DoF mechanism obtained by
locking all the actuators and by substituting one of the limbs with the corresponding
unit force f̂ i transmitted to the moving platform along the removed limb. The motion
of this mechanism could be described by the instantaneous velocity vi , which for the
proposed mechanism would be perpendicular to the directions of the two remaining
limbs. The angle between the unit force f̂ i and the instantaneous velocity vi can be
physically interpreted as pressure angle and its cosine value is given by
vi d j × dk
η i = cosα i = · f̂ i = ·s (10.12)
vi d j × d k 4,i
For each configuration of the parallel manipulator it is possible to determine three

different values of the pressure angle, each obtained by removing a different limb
from the mechanism [11–13]. These values can be expressed as
3az
η1 = cosα 1 = 2 (10.13)
√
d1 4z 2 + a − 3x − y
3az
η2 = cos α 2 = 2 (10.14)
√
d2 4z 2 + a + 3x − y
3az
η3 = cos α 3 = 2 (10.15)
2d 3 z 2 + y + 2a
and the output transmission index OTI 3 can then be computed as

O T I 3 = min η i ∀i = {1, 2, 3} (10.16)
This index depends both on the geometry and the configuration of the manipulator,
as both appear in Eqs. (10.13)–(10.15).
10.3 Equivalent Mechanism for OTI6 Formulation
By neglecting the rotational motion of the moving platform, the OTI 3 index cannot
be used to evaluate the motion/force transmission and constraint characteristics of
the mechanism unless it is acting as a pointing device. Therefore, the Transmission
Index OTI 6 is defined to properly evaluate the mechanism performance. A revolute
joint is physically manufactured by using a pair of bearings, which support any
load applied to it by radial and thrust reaction forces. A representation of such
a system, as presented in [20], is shown in Fig. 10.1b, where the ith limb of the
proposed manipulator is modelled as a 2-SPS mechanism. The 2-SPS kinematic
chain is characterized by two limbs d i,1 and d i,2 , which share the spherical joint on
the base platform and split onto two different spherical joints on the moving platform,
each representing a bearing and distant 2bi from the other. Point H lies in the middle
point between the bearings, and vectors bi,1 and bi,2 define the position of the two
bearings of the ith limb with respect to point H position. By assuming limbs d i,1 and
d i,2 as sharing the same actuation in each moment, this manipulator is kinematically
equivalent to the proposed 3-SPR mechanism. The main advantages of the proposed
modelling are the inclusion of the distance between bearings as a design parameter,
and a complete definition of the OTI 6 transmission index, as explained in [20].
A Jacobian matrix of the equivalent 6-SPS mechanism can be computed by using
the screw theory [3] and is given by
⎡ T ⎤
T
s1,1 b1,1 × s1,1
⎢ T T ⎥
⎢ s1,2 b × s1,2 ⎥
⎢ T 1,2 T ⎥
⎢s b2,1 × s2,1 ⎥
J 6S P S =⎢ 2,1
⎢ sT T ⎥ (10.17)
⎢ 2,2 b2,2 × s2,2 ⎥
⎢ T T ⎥
⎥
⎣ s3,1 b3,1 × s3,1 ⎦
T
T
s3,2 b3,2 × s3,2
where
d i, j
s i, j = (10.18)
d i, j
The overall Jacobian in Eq. (10.17) can be computed again in a closed-form

expression. The design parameters are the base platform radius a, the distances b1 ,
b2 and b3 of the bearing points from point H, the minimum link length d 0 and the
actuator stroke s. The last two parameters constrain limb lengths as
d 0 ≤ d i, j ≤ d 0 + s ∀i = {1, 2, 3}; j = {1, 2} (10.19)
Then, a global transmission index GTI is defined to evaluate the performance of

the equivalent manipulator as
126 M. Russo et al.
GT I = min(O T I 6 (x, y, z))∀{x, y, z} ∈ W (10.20)
where W is the target workspace for the manipulator.
10.4 OTI3 and OTI6 Comparison
The two formulations for the Output Transmission Index evaluated from the pres-
sure angle in Eqs. (10.16) and (10.20) give similar results for the analysed manipu-
lator, even if slight changes of behaviour and absolute value can be highlighted. In
this section, some numerical examples are reported in order to discuss these differ-
ences. Three different designs of the proposed 3-SPR manipulator have been used
for these examples: the first one is characterized by a = 60 mm, d 0 = 200 mm, s =
Fig. 10.2 OTI variation for the manipulator with a = 60 mm, d 0 = 200 mm, s = 100 mm: a OTI 3 ;
b OTI 6
b OTI 6
b OTI 6
Fig. 10.5 Relationship between the OTI 3 and OTI 6 indices from computed data sets defined by
different designs and configurations of the proposed 3-SPR mechanism
128 M. Russo et al.
100 mm, and the OTI variation in the workspace of this manipulator for the plane z
= 200.8 mm is shown in Fig. 10.2; the second example is defined by a = 75 mm,
d 0 = 200 mm, s = 100 mm, and Fig. 10.3 reports its OTI variation in the plane z =
197.9 mm; the third and last one is characterized by a = 75 mm, d 0 = 150 mm, s =
100 mm and its OTI values are mapped on the plane z = 154.9 mm in Fig. 10.4.
In all the examples, the OTI 3 index is smaller than the OTI 6 index. The case
reported in Fig. 10.2 shows that for the studied mechanism the OTI 3 maximum in
the workspace is equal to 0.425, while the OTI 6 maximum is equal to 0.463. This
is true even for the second case, where the OTI 3 maximum is 0.521 while the OTI 6
maximum is 0.564, and the third example, with 0.646 as OTI 3 maximum and 0.675
as OTI 6 maximum. Furthermore, the region with higher values of OTI 3 is centred
in the workspace, while the OTI 6 behaviour is less symmetric and the region with
higher values is slightly shifted along the negative y direction.
Figure 10.5 shows the relationship between the OTI 3 and OTI 6 indices obtained
from 20 computed data sets. Each data set is defined by a different manipulator design
and contains the OTI 3 and OTI 6 indices computed for over 104 different configura-
tions. The variation tendency shown in the figure is characterized by direct propor-
tionality. All the points lie above the diagonal OTI 3 = OTI 6 , plotted in Fig. 10.5.
Therefore, the Transmission Index evaluated for the 6-SPS model is always greater
than the OTI 3 of the 3-SPR model.
10.5 Conclusions
In this paper, the motion/force transmissibility and constrainability of a novel 3-SPR

parallel manipulator has been presented. The constrainability of the mechanism has
been studied through the computation of the complete Jacobian matrix of the system
by using the screw theory. In order to evaluate the Output Transmission Index of
the mechanism, an analysis based on two different modellings has been proposed:
first, the OTI 3 index of the manipulator has been obtained through the definition of
pressure angles; then, a kinematically equivalent mechanism has been modelled in
order to take into account the full mobility of the manipulator by evaluating the OTI 6
index. Finally, the results of the two different approaches have been compared to
highlight the differences and the relationship between the OTI 3 and OTI 6 indices.
Acknowledgements The first author has spent a period of research in 2017 at Tokyo Institute of
Technology under the supervision of prof. Y. Takeda, who is gratefully acknowledged.
References
1. Merlet, J.P.: Parallel Robots, vol. 74. Springer, Dordrecht (2012)

2. Ceccarelli, M.: Fundamentals of Mechanics of Robotic Manipulation, vol. 27. Springer, Dor-
drecht (2004)
3. Tsai, L.W.: Robot Analysis: the Mechanics of Serial and Parallel Manipulators. Wiley, New
York, NY (1999)
4. Zlatanov, D., Bonev, I. A., Gosselin, C.M.: Constraint singularities of parallel mechanisms. In:
IEEE International Conference on Robotics and Automation (ICRA 2002), vol. 1, pp. 496–502,
Washington, DC (2002)
5. Zlatanov, D., Bonev, I.A., Gosselin, C.M.: Constraint singularities as C-space singularities. In:
Advances in Robot Kinematics, pp. 183–192. Springer, Netherlands (2002)
6. Di Gregorio, R., Parenti-Castelli, V.: Mobility analysis of the 3-UPU parallel mechanism assem-
bled for a pure translational motion. ASME J. Mech. Des. 124, 259–264 (2002)
7. Wen, J.T., O’Brien, J.F.: Singularities in three-legged platform-type parallel mechanisms. IEEE
Trans. Robot. Autom. 19(4), 720–726 (2003)
8. Joshi, S.A., Tsai, L.W.: Jacobian analysis of limited-DOF parallel manipulators. ASME J.
Mech. Des. 124, 254–258 (2002)
9. Zoppi, M., Bruzzone, L.E., Molfino, R.M., Michelini, R.C.: Constraint singularities of force
transmission in nonredundant parallel robots with less than six degrees of freedom. J. Mech.
Des. 125(3), 557–563 (2003)
10. Tsai, M.J., Lee, H.W.: The transmissivity and manipulability of spatial mechanisms. J. Mech.
Des. 116(1), 137–143 (1994)
11. Takeda, Y., Funabashi, H.: Motion transmissibility of in-parallel actuated manipulators. JSME
Int. J. Ser. C Dyn. Control Robot. Des. Manuf. 38(4), 749–755 (1995)
12. Takeda, Y., Funabashi, H., Ichimaru, H.: Development of spatial in-parallel actuated manipu-
lators with six degrees of freedom with high motion transmissibility. JSME Int J. Ser. C 40(2),
299–308 (1997)
13. Takeda, Y., Funabashi, H.: Kinematic and static characteristics of in-parallel actuated manip-
ulators at singular points and in their neighborhood. JSME Int. J. Ser. C Dyn. Control Robot.
Des. Manuf. 39(1), 85–93 (1996)
14. Chen, C., Angeles, J.: Generalized transmission index and transmission quality for spatial
linkages. Mech. Mach. Theory 42(9), 1225–1237 (2007)
15. Xie, F., Liu, X.J., Wang, J.: Performance evaluation of redundant parallel manipulators assim-
ilating motion/force transmissibility. Int. J. Adv. Rob. Syst. 8(5), 66 (2011)
16. Wang, J., Wu, C., Liu, X.J.: Performance evaluation of parallel manipulators: motion/force
transmissibility and its index. Mech. Mach. Theory 45(10), 1462–1476 (2010)
17. Liu, H., Huang, T., Kecskeméthy, A., Chetwynd, D.G.: A generalized approach for computing
the transmission index of parallel mechanisms. Mech. Mach. Theory 74, 245–256 (2014)
18. Liu, H., Huang, T., Kecskeméthy, A., Chetwynd, D.G.: Force/motion/stiffness transmissibility
analyses of redundantly actuated and overconstrained parallel manipulators. In: Proceedings
of the 14th IFToMM World Congress, pp. 609–618 (2015)
19. Brinker, J., Corves, B., Takeda, Y.: On the motion/force transmissibility and constrainability
of delta parallel robots. In: Proceedings of the 7th International Symposium on Computational
Kinematics (2017)
20. Takeda, Y., Liang, X.: Transmission index of lower-mobility parallel mechanism case study on
3-PRS mechanism. In: The ASME 2017 International Design Engineering Technical Confer-
ence (IDETC2017) (2017)
21. Russo, M., Ceccarelli, M.: Kinematic design of a tripod parallel mechanism for robotic legs. In:
The 4th Conference on Mechanisms, Transmissions and Applications, Trabzon, Turkey (2017)
22. Russo, M., Herrero, S., Altuzarra, O., Ceccarelli, M.: Multi-objective optimization of a tripod
parallel mechanism for a robotic leg. Comput Kinemat. Mech. Mach. Sci. 50, 374–382 (2017)
23. Russo, M., Herrero, S., Altuzarra, O., Ceccarelli, M.: Kinematic analysis and multi-objective
optimization of a 3-UPR parallel mechanism for a robotic leg. Mech. Mach. Theory 120,
192–202 (2018)
24. Ceccarelli, M., Russo, M.: Device for the spherical connection of three bodies. IT Patent
Application No. 1,020,160,000,936,9 (in Italian) 19 Sep 2016
25. Russo, M., Cafolla, D., Ceccarelli, M.: Device for tripod leg. IT Patent Application No.
1,020,160,000,972,58 (in Italian) 28 Sept 2016
Chapter 11
Kinematic and Dynamic Dimensional
Synthesis of Extended Delta Parallel
Robots
J. Brinker, B. Corves and Y. Takeda
Abstract Industrial modifications of Delta-like robots impose high challenges on

the dimensional synthesis. On the one hand, purely kinematic approaches taking
into account the input and output transmission capabilities may be sufficient to opti-
mize the design of a basic Delta robot in the field of high-speed application. On
the other hand, a combined kinematic and dynamic optimization approach can be
employed to additionally include the actuation torques as highly important parame-
ters for the motor selection process. More importantly, taking into account dynamics
optimization and evaluation criteria (such as energy consumption and peak power of
a system) meets the growing demands for energy-efficient and sustainable manufac-
turing. Such combined approaches may be particularly advantageous when it comes
to additional moving masses as observed in recently modified industrial variants.
In this context, efficient kinematic models are introduced establishing the input and
output transmission indices based on the notion of pressure angles. The problem
of increased computational complexity of dynamic modelling approaches for func-
tionally extended Delta-like robots is tackled by employing an efficient approach
based on the Lagrange-d’Alembert Principle of Virtual Work. With these models,
sets of Pareto optimal candidates are identified by multi-objective optimization tech-
niques. The most efficient candidates resulting from both approaches are selected
and compared against one and another. It is shown that the combined approach helps
to further improve the design of the extended variants whereas superior candidates
for the light-weight basic Delta robots can already be obtained using kinematic per-
formance indices only.
Keywords Delta parallel robot · Kinematics · Dynamics · Dimensional

synthesis · Transmission index · Principle of Virtual Work
J. Brinker (B) · B. Corves

Department of Mechanism Theory and Dynamics of Machines, RWTH Aachen University,
Aachen 52072, Germany
Y. Takeda
Department of Mechanical Engineering, Tokyo Institute of Technology, Tokyo 152-8552, Japan

132 J. Brinker et al.
11.1 Introduction
Delta parallel robots serve a niche market for high-speed pick-and-place applications
[1]. The kinematic structure usually comprises three symmetric kinematic chains of
the type R(SS)2 with revolute actuation and a spatial parallelogram denoted by R and
(SS)2 , respectively. With the parallelogram design, the connecting rods only need to
transmit axial forces. In other variants the spherical joints are replaced by universal
joints in order to suppress the internal mobilities (i.e. the rotation of the rods along
their axes of symmetry) that do not contribute to the output motion.
Steadily increasing requirements of modern production processes have led to sig-
nificant modifications of industrial robots. The broadly diversified portfolio of Delta
parallel robots clearly demonstrates the trend of functionally extending industrial
designs to meet the demands of industry for higher efficiency and greater flexibility.
Available variants are characterized by up to three additional serial chains driving
a serial wrist joint mechanism and thus providing additional rotational degrees-of-
freedom (dof) to the basic parallel structure.
Industrial versions use, e.g.,
• one or two RUPUR chains actuated from the frame (ABB IRB 360 and Codian
D5),
• one or three RRUR chains attached to the parallelograms (FANUC M-3iA/6S and
M-3iA/6A), or
• direct drives at the platform (Kawasaki YF03).
Recent developments include two RUPUR chains coaxially arranged and actuated
from the frame (MAJAtronic RL5). This market trend of functionally extending the
original Delta parallel robot is confirmed analysing related patents. Accordingly, the
number of patents significantly increased after the expiration of Clavel’s patents in
2007. Exemplarily, some recent inventions of serial-parallel variants that are not yet
available on the market employ
• a drive shaft with special U-joint allowing linear motion (up to 6-DOF)
[US8047093 2010]
• a drive shaft attached to the parallelogram (up to 6-DOF) [US8307732 2010]
• a coaxially guided drive shaft (5-DOF) [US20150202779A1]
• a drive shaft guided through prox. link and parallelogram (up to 6-DOF)
[US20130142608]
• a coaxially guided linear-rotary drive (5-DOF) [DE102013106004A1]
• a drive shaft guided through prox. link and connecting rod used as drive shaft (up
to 5-DOF) [DE102013213057A1]
Following the concepts from academia and industry, four potential functional
extensions are proposed in [2] and further refined, modelled and analysed in [3], cf.
Fig. 11.1. Comparing the dynamic influences of the additional chains on the basic
3-dof structure and related design issues, it is shown that the telescopic motions of
the additional members (appearing in EF and EPL ) can be circumvented attaching the
11 Kinematic and Dynamic Dimensional Synthesis of Extended Delta … 133
Basic Delta EF E PL E DL EP
Fig. 11.1 Illustrations of translational 3-dof Delta robot and 6-dof variants with functional exten-
sions
additional wrist motor(s) to the distal links (EDL ). However, higher moving masses
arise increasing the actuation torques of the basic parallel structure. Even higher
parasitic effects are observed for variants including direct drives (EP ). Moreover,
from the modelling point of view, EP corresponds to the standard Delta robot with
higher platform masses and thus do represent a challenge [3].
Against this background, this contribution is concerned with the dimensional syn-
thesis of functionally extended Delta-like parallel robots in which three additional
serial chains are attached to the distal links (EDL ). Using kinematic and dynamic
performance measures, different modelling approaches are elaborated potentially
improving the optimization results for extended Delta robots compared to the appli-
cation to standard Delta robots. First, simple physically meaningful definitions of
kinematic indices based on pressure angles are proposed. Second, in order to compute
the actuation torques of the functionally extended Delta robots, a validated dynamic
model based on the Lagrange-d’Alembert Principle of Virtual Work is presented.
Finally, the proposed indices are employed for multi-objective Pareto optimizations
of the basic and the extended Delta parallel robot EDL .
11.2 Dimensional Synthesis
Based on the requirements of given handling tasks, a first design step may be to
determine the kinematic parameters of robot architectures such that a prescribed
workspace is reached. By solely using workspace and occupied space requirements
the resulting design may however suffer from poor kinematic and dynamic charac-
teristics. Thus, kinematic and dynamic performance measures are commonly taken
into account as (additional) optimization criteria.
Delta-like parallel robots have been optimized extensively during the last three
decades. In respect of purely kinematic approaches, Miller [4] proposes a weighted
objective function containing the global conditioning index (i.e., the average value
of the reciprocal condition number over the workspace) and a space utilization index
for Delta-like robots. Pierrot et al. [5] optimize a fully-parallel 4-dof Delta-like robot
by minimizing the length of a kinematic chain while constraining the maximum
actuator velocity (depending on available motor and gear types) and the maximum
condition number, where subsequent experiments are used to improve the dynamic
performance.
In respect of combined kinematic and dynamic optimization approaches, Zhang
and Song [6] propose a novel index taking into account the ratio of average power
consumption and total moving mass while constraining the range of the first natu-
ral frequency. Tackling the opposing behaviour of velocity and error transmission,
Zhang et al. [7] utilizes transmission angle constraints while optimizing the maxi-
mum actuation torque resulting from the inertial effects of all components reduced
to the proximal link. The proposed definition of the transmission angles is criti-
cally discussed in [8]. Courteille et al. [9] identify Pareto optimal solutions for Delta
robots taking into account the condition number and minimum singular value of the
global stiffness matrix as well as the volume of the related stiffness ellipsoid. Wu
et al. [10] perform a multi-objective design optimization of the novel 4-dof fully-
parallel Ragnar robot using the workspace volume, the root mean square (RMS)
torque and the actuation velocity as performance criteria. The ratio of distal and
proximal link lengths, the orientation of the moving platform, the local transmission
index (defined as the minimum of input and output transmission index throughout
the entire workspace), as well as the position and orientation errors resulting from
the structural stiffness are taken into account as constraints. In [11] the objectives are
changed to the workspace volume and the global transmission index (defined as the
average value of the minimum of input and output transmission index throughout the
entire workspace) while constraining the extreme values of link ratios and orientation
capabilities of the platform.
Delta-like robots are tailored for high-speed applications. Thus, good velocity
transmission capabilities favour the overall design of such robots. In addition to
the maximum speed criterion, two torque-related criteria (RMS and peak torque)
dominate the motor selection process as cost-relevant design phase subsequent to
the dimensional synthesis [12]. Against this background, this contribution is con-
cerned with two optimization approaches, a purely kinematic (solely optimizing
the motion/force transmission) and a combined kinematic and dynamic optimiza-
tion approach (additionally taking into account the maximum actuation torque). The
resulting Pareto candidates are evaluated based on the maximum power (combining
velocity and torque criteria into a single performance indicator). For the sake of
brevity, the geometric and fundamental kinematic relations of the Delta robot are
presumed to be known in the following, see also Fig. 11.2.
11.2.1 Kinematic Performance Indices
Jacobian-based indices (e.g., the manipulability [13] or the condition number [14])
are common measures for the performance optimization of parallel robots. However,
inhomogeneous units in case of mixed dof of the moving platform, undetectable
singularities using the input-output Jacobian of limited-dof manipulators, and the
Fig. 11.2 Geometric and kinematic relations
frame-dependency may impede flawless analyses. To overcome these problems,

motion/force transmission indices can be used.
The concepts for analysing the transmissibility of parallel robots have been revised
several times in the last decades. Recent approaches contain generally applicable
measures based on the normalized virtual coefficient called power coefficient [15].
In contrast to the simple definition of input transmission indices, analyses of the
output transmission capabilities usually require additional investigations of the con-
strainability. However, the related constraint transmission indices are not required
for parallel robots, in which the output link is supported by six links. In these special
cases, the transmission wrench screws (TWS) at the application points (i.e., spherical
joints) degenerate to transmission force lines. Then, the pose-dependent maximum
virtual coefficient (as used for normalization) is unambiguous. An example of such
simplification applied to a 6-PUS can be found in [16]. Further simplifications can
be achieved by defining the TWS at each of the six spherical joints. In this case,
normalization is not required at all. A detailed derivation and application to Delta
parallel robots including an extensive literature overview can be found in one of our
previous studies [8]. Accordingly and based on the relations shown in Fig. 11.2, the
TWS of the supporting link ( j, i) with respect to a spherical joint Ck,m are defined
as

f̂ j,i
$̂T W S, j,i = (11.1)
r P, j,i − r P,k,m × f̂ j,i
where f̂ j,i and r P, j,i denote the rod forces and joint positions on the platform,
respectively. Imagine to fix all actuators and to remove one of the six connecting
rods k of a chain m (Fig. 11.2, k = 1), the output link of the resulting single-dof
mechanism performs a virtual motion [17]. The related instantaneous motion is given
by the following output twist screw (OTS)
(k,m)

(k,m) ω
$O T S = (k,m) (11.2)
v j,i
with (k,m) ω as angular velocity and (k,m) v j,i as translational velocity at joint C j,i of
the output link. As all $̂T W S, j,i except $̂T W S,k,m represent constraint wrenches only
and no work can be applied to the output link [16], the reciprocal condition
δW j,i = $̂T W S, j,i ◦(k,m) $ O T S = 0 (11.3)
holds. Accordingly, a system of five equations needs to be solved for the unknowns
of (k,m) $ O T S , where one entry can be chosen arbitrarily while constraining the sum
of all entries of (k,m) $ O T S to 1. In order to compute the pressure angle αk,m at joint
Ck,m , the instantaneous velocity (k,m) v j,i can be extracted from the solution vector
and leads to the output transmission index

O T I = min cos αk,m ∀m = {1, 2, 3}, ∀k = {1, 2} (11.4)
with
(k,m) T
cos αk,m = v k,m f̂ k,m /(k,m) v k,m (11.5)
The input transmission index is defined as the minimum of the cosine of the
absolute pressure angles γi among all three kinematic chains
I T I = min(|cos(γi )|) ∀i = {1, 2, 3} (11.6)
The pressure angles of the input transmission are given by
cos(γi ) = v TB,i f̂ j,i /v B,i (11.7)
11.2.2 Dynamic Performance Indices
A comparative study [3] of the four types of functionally extended Delta robots
revealed the merits of 6-dof variants in which the additional wrist motor(s) are
attached to the distal link(s), cf. Sect. 11.1 and Fig. 11.1. It is also shown in [3]
that computational costs to solve the inverse dynamics problem may be reduced
applying alternative modelling techniques such as the Lagrange-d’Alembert Prin-
ciple of Virtual Work [18], Kane’s formulation [19], or the Method of the Natural
Orthogonal Complement [20]. Against this background, the Lagrange-d’Alembert
Principle of Virtual Work is applied to analyses of the dynamic effects of the exten-
sions on the actuation torques of the basic Delta robot. Based on the dynamic anal-
yses of a Stewart-Gough manipulator in [18], Zhao et al. [21] propose the inverse
dynamics model for the basic 3-dof Delta robot employing the principle of virtual
work. Accordingly, the fundamental idea is that the work due to external forces
and moments corresponding to any set of virtual movements is zero. The modelling
procedure is as follows. First, six-dimensional resultants of the applied and inertia
wrenches are determined for each body. Second, so-called link Jacobian matrices
[18] are formulated relating the platform displacement to the virtual movements
of the links. Finally, the desired driving torques can be computed for a given state
of motion [22]. Externally applied forces and moments as well as frictional forces
are neglected in the following. The six-dimensional wrenches acting on the tubular
member Q T ,i and the related wrist motor of an additional serial chain due to the
gravity g and inertial effects are

f T ,i m T,i g − ẗ T ,i
Q T ,i = = (11.8)
n T ,i−I T ,i ω̇ T ,i − ω T ,i × I T ,i ω T ,i

f M,i m M,i g − c̈i

Q M,i = = (11.9)
n M,i −I M,i ω̇ T ,i − ω M,i × I M,i ω T ,i
where f and n denote the forces and moments and m and I denote the masses
and inertia matrices
expressed
about the respective COG. With the resultants of the
proximal links Q 1,i , one of the two connecting
rods Q 2,i of the parallelogram
of the distal links, and the platform Q P , for the variation of work it follows

3
δW = δϕ 1T τ + δχ TP Q P + δχ Ts,i Q s,i = 0 (11.10)
i=1 s∈B
with δϕ 1 and δχ P as three-dimensional virtual rotations and virtual displacements,

respectively, and δχ Ts,i as the six-dimensional virtual movements of the remaining
bodies B = {1, 2, M, T}.
To compute the unknown driving torques τ , all virtual movements need to be
expressed by the virtual displacement δχ p of the platform. Therefore, link Jacobian
matrices are introduced mapping the platform velocity to the link velocities. Accord-
ingly, the velocity of the attachment point of the wrist motor and the tubular member
can be derived as

ċi = ṗ − (1 − x M ) ω2,i × l DL,i = E + (1 − x M ) l 2,i × J 2,i,ω ṗ
= J M,i ṗ (11.11)

ṫ T ,i = J M,i − [t i0 ]× [t i0 ]× E − J M,i /2 ṗ = J T ,i ṗ (11.12)
where x M as relative wrist motor position and J 2,i,ω as link Jacobian matrix relating
the platform to the angular velocity of a distal link [22]. E and []× denote a 3 × 3
identity matrix and a cross-product matrix or skew-symmetric matrix, respectively.
The angular velocity is the same for the motor and the tubular member. Thus, the
following applies

1
ω T ,i = [t i0 ]× E − J M,i ṗ = J T ,i,ω ṗ = J M,i,ω ṗ = J i,ω ṗ (11.13)
lT
The virtual movements (displacement and rotation) of the bodies are exclusively
expressed by the virtual displacement of the platform, i.e., δϕ 1 = J P δχ P , δχ 1,i =
J ∗1,i δχ P , δχ 2,i = J ∗2,i δχ P , and δχ s,i = J ∗s,i δχ P , where the asterisk denotes 6 × 3
link Jacobians. Rearranging of (11.10) and cancellation of δχ TP gives

3
τ= − J −T
P QP + J ∗T
s,i Q s,i (11.14)
i=1 s∈B
where for the bodies it follows B = {1, 2, M, T}.
11.2.3 Design Optimization Problem
The lengths of the proximal and distal links l P L and l DL as well as the radii of frame
and platform r F and r P represent four dimensional design variables (Fig. 11.2),
where symmetry is presumed. The prescribed workspace consists of a cylinder (with
radius D1 = 1.00 m and height Z 1 = 0.25 m) and a conical portion (with radius
D2 = 0.80 m and height Z 2 = 0.05 m) adjacent to it. The centre of their connecting
surface determines the relative position P 0 = [0, 0, Z0 ] of the workspace to the
origin of frame O, where Z 0 is used as design variable. Consequently, the following
five design variables are considered:

χ = l P L l DL r F r P Z 0 (11.15)
The kinematic indices are derived for 55 evaluation points W within the prescribed
workspace as illustrated in Fig. 11.2. The minimum over the workspace points is
taken into account for each transmission index (denoted by I T Imin and O T Imin ).
For the dynamic indices, an industrially-relevant path similar to [23] is defined in
accordance with the prescribed workspace size and rotated about the vertical axis in
steps of π/6. Thus, 18 trajectories Γ are analysed in respect of the resulting actua-
tion torques. Here, the maximum actuation torque over all trajectories constitutes the
dynamic optimization criterion, i.e., τmax . The underlying mass distribution is based
on [3], the mass parameters are adapted according to the changes in link lengths
during dimensional synthesis based on [23]. In the initial design phase, the designer
may be interested in determining the greatest possible design parameter space that
provides all potential robot designs. However, in order to increase the performance of
the optimization approach, the design parameter space should be smallest possible.
Thus, in addition to the customer requirements (e.g., workspace shape and size as
well as trajectories), a set of additional requirements g are taken into account. Such
requirements may arise from parametric relations based on practical implementa-
tion (e.g., maximum joint motion angle), kinematic characteristics (e.g., maximum
pressure angle), or other experience-based conditions such as maximum link length
ratio. Applying requirements based on experience may lead to unneeded and unde-
sired reductions of the solution space. The following analyses are thus solely based
on mandatory requirements. For a critical point P c = [−D1 /2, 0, Z 0 + Z 1 ], the
maximum pressure angle γi,max can for instance be used to formulate the following
constraint.
The detailed derivation of other constraints is left to be addressed in future pub-
lications.
2
−D1 /2 + r P,i − r F,i + (Z 0 + Z 1 )2

≤ l 2P L ,i + l 2DL ,i − 2l P L ,i l DL ,i cos γi,max + π/2 (11.16)
The problem can finally be stated as follows
(A)max. f A1 (χ, W ) = I T Imin , max. f A2 (χ , W ) = O T Imin

(B)max. f B1 (χ , W ) = I T Imin , max. f B2 (χ , W ) = O T Imin ,
max. f B2 (χ, W ) = O T Imin
subject to constraints g(χ, W ) (11.17)
where (A) and (B) denote a purely kinematic (solely optimizing the motion/force
transmission) and a combined kinematic and dynamic optimization approach (addi-
tionally taking into account the maximum actuation torque), respectively. Both of
which are implemented in Matlab® and solved using the built-in multi-objective
optimization function gamultiobj, which is based on the well-known Non-dominated
Sorting Genetic Algorithm NSGA-II [24]. The results are obtained for the basic 3-
dof Delta robot (no extensions) and the 6-dof Delta robot (with extensions to each
distal link, EDL ).
The optimizations are performed presuming a minimum workspace ratio of ϑ ≥ 0.3

and a minimum value for the smallest input transmission index of I T Imin ≥ 0.4.
When higher demands on the workspace ratio need to be fulfilled, the Pareto front
is shifted towards worse results for both criteria I T Imin and O T Imin and vice versa.
Figure 11.3 depicts the resulting candidates of approach (B) based on the three
objectives (i.e., I T Imin , O T Imin , τmax ), where, for clearer presentation, τmax is nor-
malized by the maximum torque of all candidates and displayed as 1 − τmax,0 . In
addition, all candidates are coloured based on the related normalized peak power
Pmax,0 . At the same time, black circles indicate the Pareto candidates of the purely
kinematic approach (A). The related Pareto front of (A) clearly boarders the Pareto
surface proving the robustness of the more complex approach (B). For both Delta
variants (3-dof and 6-dof) similar surface shapes are obtained, whereas the spread
of τmax,0 is larger in the 6-dof case. Both distributions of candidates taper for maxi-
mal I T Imin until reaching maximal peak torques. Irrespective of these torque peaks
and independent from the optimization approach, it can be seen from the colour
indication for the 3-dof Delta variants (Fig. 11.3a) that the power efficiency can
substantially be improved by optimizing the input transmissibility and, associated
therewith, slight reductions in O T Imin . Candidates with minimum peak power are
marked in red. In contrast, power efficient candidates for the 6-dof Delta variants
are observed for peak torques improvements at expense of decreasing I T Imin and
O T Imin . Accordingly and as displayed in Fig. 11.3b, superior solutions are obtained
applying the combined kinematic and dynamic optimization approach (B) instead
of the purely kinematic approach (A). This can be explained by the colour gradient
denoting the peak power distribution. For the 3-dof candidates, the peak power is
reduced for increasing I T Imin with minimal values at the tip of the surface (cf. arrow
in Fig. 11.3a). This area also includes candidates resulting from approach (A). For the
6-dof candidates, the colour gradient is evidently changed (cf. arrow in Fig. 11.3b)
so that power efficient solutions are observed along the border adjacent to the border
given by the Pareto candidates of approach (A). This also means that optimal can-
didates cannot stem from the purely kinematic approach (A). Instead of considering
O T Imin as third optimization criterion, it is commonly proposed to combine the
transmission indices taking into account the minimum or the product of I T Imin and
O T Imin . However, following approach (B), both of the latter approaches would lead
to rather small values for I T Imin as O T Imin is preferably increased to simultane-
ously allow for smaller torques (and thus, smaller I T Imin ). It should also be noted
that O T Imin can be directly influenced by the distance between the spherical joints
D S (cf. Fig. 11.2) without affecting the other two criteria. O T Imin should however
be maximized for any approach in order to improve the output transmission capabil-
ities. A comparison of the candidates with the lowest peak power as resulting from
approaches (A) and (B) with a benchmark robot as introduced in [8] shows the ben-
efits of including dynamic characteristics in the optimization process (Table 11.1).
Accordingly, for the basic 3-dof variant, the maximum instantaneous power Ptot,max
and the total energy consumption E tot of the three main drives for all trajectories
can be reduced by more than 14% with slightly better performance of (A). At the
same time, losses in the workspace ratio and the peak torque need to be accepted.
Similar results are obtained optimizing a 6-dof variant with (A). Further consider-
able improvements are obtained using the combined approach (B), while keeping
(a) Basic: 3-dof (b) E DL: 6 -dof
Fig. 11.3 Results for (A) and (B) applied to the 3-dof (a) and 6-dof (b) Delta robot variants
Table 11.1 Pareto solutions with minimum normalized peak power

(a)
r F,i r P,i l P L ,i l DL ,i Z0
Benchmark: 3-dof 0.200 0.050 0.400 0.800 –0.800
Basic: 3-dof (A) 0.202 0.070 0.460 0.986 –1.074
Basic: 3-dof (B) 0.200 0.070 0.510 0.936 –1.051
Benchmark: 6-dof 0.200 0.050 0.400 0.800 –0.800
EDL : 6-dof (A) 0.202 0.070 0.460 0.986 –1.074
EDL : 6-dof (B) 0.202 0.070 0.344 1.006 –1.072
(b)
I T Imin O T Imin ϑ Ptot,max E tot τmax
Benchmark: 3-dof 0.647 0.437 0.366 0 0 0
Basic: 3-dof (A) 0.818 0.448 0.301 –17.5% –15.2% +15.5%
Basic: 3-dof (B) 0.809 0.460 0.300 –15.4% –14.9% +16.8%
Benchmark: 6-dof 0.647 0.437 0.366 0 0 0
EDL : 6-dof (A) 0.818 0.448 0.301 –16.2% –14.1% +17.2%
EDL : 6-dof (B) 0.745 0.351 0.350 –24.1% –17.5% +2.8%
the peak torque at similarly low levels and slightly reducing I T Imin and O T Imin .
The associated decline of O T Imin can for instance be compensated increasing D S .
11.4 Conclusion
In this paper it was shown that combined kinematic and dynamic optimization
approaches can lead to better overall designs as compared to variants optimized
by kinematic indices only. In this context, a simple definition of input and output
transmission indices is presented. In addition, an efficient dynamic model for Delta
parallel robots with functional extensions is presented. In these variants, the mass
distribution is changed increasing the relevance of the dynamic characteristics. The
advantages of integrating dynamic indices into the design process are clearly empha-
sized by the results of multi-objective Pareto optimizations. While the design of the
basic light-weight 3-dof Delta robot is still mainly influenced by velocity character-
istics, the optimal design of the extended 6-dof variant (including additional masses
attached to its links) represents a compromise of velocity and torque characteristics.
Due to the strong correlation of input transmission characteristics and maximum
actuation velocity, purely kinematic design optimization approaches may lead to
similar results for the basic Delta robot. Exemplarily, the selection of candidates was
based on the minimum peak power. The final evaluation of the candidates’ fitness
needs to be assessed in close consultation with the customer.
Acknowledgements This work is supported by the German Academic Exchange Service (DAAD)
with funds from the Federal Foreign Office (FFO).
References
1. Brinker, J., Corves, B.A.: Survey on parallel robots with Delta-like architecture. In: Proceedings
of the 14th World Congress in Mechanism and Machine Science, Taipei, Taiwan (2015)
2. Borchert, G., Battistelli, M., Runge, G., Raatz, A.: Analysis of the mass distribution of a
functionally extended delta robot. Rob. Comput. Integr. Manuf. 31, 111–120 (2015)
3. Brinker, J., Funk, N., Ingenlath, P., Takeda, Y., Corves, B.: Comparative study of serial-parallel
Delta robots with full orientation capabilities. In: IEEE Robotics and Automation Letters,
Additionally Selected for Presentation at the Conference ICRA’17 (2017)
4. Miller, K.: Optimal design and modeling of spatial parallel manipulators. Int. J. Robot. Res.
23(2), 127–140 (2004)
5. Pierrot, F., Nabat, V., Company, O., Krut, S.: Optimal design of a 4-DOF parallel manipulator:
from academia to industry. IEEE Trans. Robt. 25(2), 213–224 (2009)
6. Zhang, L., Song, Y.: Optimal design of the Delta robot based on dynamics. In: Proceedings of
the 2011 IEEE International Conference on Robotics and Automation, Shanghai, China (2011)
7. Zhang, L.M., Mei, J.P., Zhao, X.M., Huang, T.: Dimensional synthesis of the Delta robot using
transmission angle constraints. Robotica 30, 343–349 (2011)
8. Brinker, J., Corves, B., Takeda, Y.: On the motion/force transmissibility and constrainability of
Delta parallel robots. In: 7th IFToMM International Workshop on Computational Kinematics
(CK2017), Futuroscope-Poitiers, France (2017)
9. Courteille, E., Deblaise, F., Maurine, P.: Design optimization of a Delta-like parallel robot
through global stiffness performance evaluation. In: Proceedings of the IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS), pp. 5159–5166, St. Louis, USA (2009)
10. Wu, G., Bai, S., Hjørnet, P.: Parametric optimal design of a parallel schönflies-motion robot
under pick-and-place trajectory constraints. In: Proceedings of the IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS), Hamburg, Germany (2015)
11. Wu, G., Bai, S., Hjørnet, P.: Multi-objective design optimization of a parallel schönflies-motion
robot. Adv. Rec. Mech. Robt. II, 657–667 (2016)
12. Han, G., Xie, F., Liu, X.J.: Optimal selection of servo motor and reduction ratio for high-speed
parallel robots. In: Liu, H. et al. (eds.) ICIRA 2015, Part II, LNAI 9245, pp. 109–120. Springer,
Cham (2015)
13. Yoshikawa, T.: Manipulability of robotic mechanisms. Int. J. Robot. Res. 4(2) (1985)
14. Gosselin, C., Angeles, J.: The optimum kinematic design of a planar three-degree-of-freedom
parallel manipulator. J. Mech. Trans. Autom. Des. 110(1) (1988)
15. Liu, H., Huang, T., Kecskeméthy, A., Chetwynd, D.G.: A generalized approach for computing
the transmission index of parallel mechanisms. Mech. Mach. T. 74, 245–256 (2014)
16. Wang, J., Wu, C., Liu, X.J.: Performance evaluation of parallel manipulators: motion/force
transmissibility and its index. Mech. Mach. T. 45, 1462–1476 (2010)
17. Takeda, Y., Funabashi, H.: Motion transmissibility of in-parallel actuated manipulators. JSME
Int. J. Ser. C 38(4), 749–755 (1995)
18. Tsai, L.W.: Solving the inverse dynamics of a Stewart-Gough manipulator by the principle of
virtual work. J. Mechan. Des. 122, 3–9 (2000)
19. Kane, T.R., Levinson, D.A.: The use of Kane’s dynamical equations for robotics. Int. J. Robot.
Res. 2, 3–21 (1983)
20. Angeles, J., Lee, S.K.: The formulation of dynamical equations of holonomic mechanical
systems using a natural orthogonal complement. ASME J. Appl. Mech. 55, 243–244 (1988)
21. Zhao, Y., Zhiyong, Y., Tian, H.: Inverse dynamics of Delta robot based on the principle of
virtual work. Trans. Tianjin Univ. 11(4), 268–273 (2005)
22. Brinker, J., Corves, B., Wahle, M.A.: Comparative study of inverse dynamics based on Clavel’s
Delta robot. In: Proceedings of the 14th World Congress in Mechanism and Machine Science,
Taipei, Taiwan (2015)
23. Brinker, J., Lübbecke, M.E., Takeda, Y., Corves, B.: Optimization of the reconfiguration plan-
ning of handling systems based on parallel manipulators with Delta-like architecture. IEEE
Robotics and Automation Letters, Additionally Selected for Presentation at the Conference
CASE’17 (2017)
24. Deb, K.: Multi-Objective Optimization Using Evolutionary Algorithms, p. 515. Wiley, Chich-
ester (2001)
Chapter 12
An Influence Based Error Identification
for Kinematics Calibration of Serial
Robotic Manipulators
D. A. Patel, T. F. Lu and L. Chen
Abstract In serial robotic manipulators, due to the nature of the coupling of links, the
influence of errors in joint parameters on pose accuracy varies with the configuration.
Kinematics parameter’s error identification in the standard kinematics calibration has
been configuration independent which does not consider the influence of kinematics
parameter on robot tool pose accuracy for a given configuration. Mutually dependent
joint parameter errors cannot be identified at the same time, and hence error of one
parameter in each pair is identified. In a pair of mutually dependent joint parameters,
the effect of error in one parameter on positional error can be more than the other
one depending on the configuration. Therefore, the error detection may be incorrect
if the influence of joint parameters is ignored during the error identification. This
research analyses the configuration dependent influences of kinematics parameters
error on pose accuracy of a robot. Based on the effect of kinematics parameters, the
errors in the kinematics parameters are identified. Kinematics model of the robot is
composed of the modified DH method and an improved DH method to avoid the
limitations of the original DH method. First, the robot is calibrated to identify errors
in 17 kinematics parameters conventionally, and then errors are detected based on
the proposed method.
Keywords Calibration · Error identification · Kinematics · Measurement · Pose

accuracy · Serial robots
12.1 Introduction
Serial robotic manipulators have numerous applications across various industries

such as automobile, manufacturing, medical, space, and so forth. Various geometric
factors such as inaccurate joint angles, joint twist, link lengths and non-geometric
factors such as joint eccentricity, joint flexibility, dynamics and control errors intro-
duce the difference between the model of a manipulator on the controller and an
D. A. Patel (B) · T. F. Lu · L. Chen

School of Mechanical Engineering, University of Adelaide, Adelaide, SA 5005, Australia

146 D. A. Patel et al.
actual mechanism and hence accuracy of a robot decreases [1]. Robot calibration
reduces this difference to improve accuracy. The geometric parameters errors can
be systematically identified and compensated whereas other errors are difficult to
model and identify. The main reason that causes the pose error is inaccurate geomet-
ric parameters used to calculate the pose [2]. So often geometric errors correction
fulfils the desired pose accuracy for many applications. Geometric errors calibration
is also known as a level 2 or a kinematics calibration. The current kinematics cali-
bration includes four steps, namely kinematics and error modelling, end-effector’s
pose measurement, identification of error sources and compensation for the errors
into a kinematic model [3]. The Level 2 calibration methods modify the kinematic
parameters of the robot to minimize the difference between the kinematics model of
a robot in the controller and actual mechanism. The modified kinematics parameters
best fit the accuracy over the selected poses or small region of the robot’s workspace.
Completeness and continuity of kinematics model, precision of measurement system,
accurate errors identification, proper error compensation scheme, and specification
of the robot determine the effectiveness of the kinematics calibration process.
Kinematics modelling using Modified DH and improved DH method, the Com-
plete and Parametrically Continuous (CPC) model, the Modified CPC model, and
POE formula results into the complete and continuous kinematics model of the robot
[4]. The large volume metrology such as Laser tracker is accurate up to 15 μm [5].
However, it is difficult to directly measure errors of individual kinematics parameter
(i.e. joint-link parameter such as joint orientation or link length). An error identifi-
cation in the contemporary kinematics calibration simultaneously approximate the
errors of all kinematics parameters from a measured pose of robot end-effector for
a given configuration using methods such as linear least squares, non-linear least
squares, pseudo-inverse, genetic algorithm, and heuristic search method [6]. This
process is repeated for few selected configurations to calculate a set of kinematics
parameters which best fit the accuracy to all selected configurations. However, dif-
ferent pose errors occur for the same individual joint parameter over various config-
urations. For example, in Fig. 12.1, θ1 , θ2 , and θ3 are mutually dependent parameters
whose errors cause positional error at end-point P. In configuration 1 (i.e. P1 ), θ1 is
more influential than θ2 , and opposite in configuration 2 (i.e. P2 ). Therefore, during
the error identification more influential parameter in each pair must be considered
at every selected configuration. However, contemporary error identification ignores
the configuration dependency of the influence of kinematics parameters on a pose
accuracy which leads to incorrect error identification at certain configurations of a
robot. Therefore, this research analyses the influence of each kinematics parameters
on the pose accuracy of a robot and proposes an influence based error identification.
Section 12.2 prepares the kinematics model. Section 12.3 contains error identifica-
tion and compensation. Section 12.4 performs experiments and Sect. 12.5 concludes
the research.
12 An Influence Based Error Identification for Kinematics … 147
Fig. 12.1 Mutually

dependent parameters with
variable influence
12.2 Kinematics
This research combined the modified DH method and improved DH method to retain
continuity of kinematics model of the robot considering the nominal values of the
kinematics parameters listed in Table 12.1. In the modified DH method, the frame
i is rigidly attached to the link i, which rotates around joint i. The transformations
i T between the frames (i − 1) and i is described with the help of two rotational
i−1
parameters αi−1 and θi , and two translational parameters ai−1 and di (Fig. 12.2).
Therefore, the homogeneous link transformation matrix i−1i T is obtained using
the following transformations as:
i−1
iT = Rot(X, αi−1 )T rans(X, ai−1 )Rot(Z , θi )T rans(Z , di ). (12.1)
Table 12.1 Kinematics parameters of Katana 450

Joint i αoi−1 ai−1 mm θoi βoi−1 di mm
1 0 0 ±169.5 – 0
2 90 0 +102/−30 – 0
3 0 190 ±122.5 0 –
4 0 139 ±112 0 –
5 −90 0 ±168 – 147.3
6 90 0 Inactive – 200
Fig. 12.2 Frames assignment
Joint 2, 3 and 4 are parallel so the improved DH method must be employed with
an additional parameter β to correlate frames 2, 3 and 4 to avoid discontinuity. The
transformation matrix is obtained using transformations:
i−1
iT = Rot(X, αi−1 )Rot(Y, βi−1 )T rans(X, ai−1 )Rot(Z , θi ). (12.2)
Equation (12.2) correlate frames 2, 3 and 4 using T 23 and T 34 . The improved H

method avoids the limitations of the modified DH method. The transformation 06 T
between the robot base and the robot end-effectors is obtained by putting values of
joint link parameters of Table 12.1 into transformation matrices as:
0
6T = 01 T · 12 T · 23 T · 34 T · 45 T · 56 T (12.3)
The homogeneous transformation matrix 06 T in (12.3) describes pose (i.e. position

and orientation) of the robot end-effectors on the robot’s nominal base. The kine-
matics model of the Katana 450 robot describes orientation of robot’s end-effector
as ZXZ Euler angles Φ, θ , and Ψ . Therefore, the pose P of robot is defined by the
coordinates X , Y , and Z and orientation angles Φ, θ , and Ψ in the form of vector
P = [X Y Z ΦθΨ ]T . The derived kinematic model of the robot has been verified
against the robot’s control software. For the calibration purpose, the positional error
vector P between the actual pose Pa and the theoretical pose Pt of the end-effectors
can be described as:
T
P = Pa − Pt = X Y Z . (12.4)
12.3 Error Identification
12.3.1 Influence of Kinematics Parameters on Positional

Accuracy
The coordinates of 118 poses are selected within the largest cube of the robot’s
workspace as per proposed performance criteria and related test methods in the
ISO 9283:1998 [7] for the robotic manipulators. A deviation of +0.05 on angular
parameters (θi . . . , αi . . . βi . . .) and +0.1 mm on linear parameters (ai . . . , di . . .) is
imposed one by one at a given data point (i.e. configuration). This simulation provided
the influence of each kinematics parameters for a given robot configuration. Same
process is repeated over 118 configurations. The error of +0.1 mm in linear param-
eters causes an absolute positional error of 0.1 mm regardless of the configuration
of the robot. However, an error of +0.05° in rotational parameters causes configura-
tion dependent error on end-effectors position as shown in Fig. 12.3. The common
understanding is the influence of rotational parameters error decreases from the base
towards end-effector in serial robot, i.e. error in θ2 has a larger impact on positional
accuracy than θ3 . However, error in θ3 can have a larger impact on positional accuracy
than θ2 for some configurations as per analysis in Fig. 12.3. Therefore, followings
section proposes influence based error identification of kinematics parameters error.
12.3.2 Influence Based Error Identification

For the calibration purpose, positional errors vector P = X Y Z is cor-
related to the kinematics parameters error vector E with the help of the mapping
matrix J as:
P = J · E. (12.5)
⎡ ∂ PX ⎤
∂θ1
. . . . ∂∂θP5X ∂ PX
∂α0
. . . ∂∂αPX5 ∂ PX
∂β2
∂ PX ∂ PX ∂ PX ∂ PX ∂ PX
∂β3 ∂a3 ∂a4 ∂d5 ∂d6
⎢ ∂ PY ∂ Py ∂ Py ∂ PY ∂ PY ∂ PY ∂ PY ⎥
where, J = ⎣ ∂θ1
. . . ∂∂θP5Y ∂ PY
∂α0
. . . ∂∂αPY5 ∂β2 ∂β3 ∂a3 ∂a4 ∂d5 ∂d6 ⎦
∂ PZ ∂ Py ∂ PZ ∂ PZ ∂ PZ ∂ PZ
∂θ1
. . . ∂∂θP5Z ∂ PZ
∂α0
. . . ∂∂αPZ5 ∂ PX
∂β2 ∂β3 ∂a3 ∂a4 ∂d5 ∂d6
and
Fig. 12.3 Influence of kinematics error on position accuracy
E = [θ1 . . . θ5 α0 . . . α5 β2 β3 a3 a4 d5 d6 ]T .
Equation (12.5) correlates the kinematics parameters error vector E with the
positional error vector P. Firstly, the kinematics parameters’ errors are identified
using the unique least square estimation [8] as:
JT
E = · P (12.6)
J · JT
Equation (12.6) is iteratively used at each pose to correct the kinematics param-
eters error. In each iteration, a new E is obtained which is compensated in (12.5)
to obtain new P. This process is repeated till the positional error is detectable by
the measurement equipment being used (i.e. above 0.01 mm in this case) for the
calibration. The same procedure identifies the kinematics parameters’ errors for all
poses. From the sets of errors in the kinematics parameters of all poses, a set of
kinematics parameters is calculated that best fit the accuracy to all measured poses.
However, the error detection may be incorrect if influence is not considered dur-
ing the error identification. For example, in the configuration θ1 , θ2 , θ3 , θ5 = 0 and
◦
θ4 = 80 influence of θ4 is larger than θ3 . The positional error P can be corrected
by correcting θ3 or θ4 . Even if influence of θ4 is larger than θ3 for that configura-
tion, the conventional identification may identify larger error of θ3 instead of smaller
error in θ4 for the same positional error P. This incorrect identification of large
error in θ3 at this configuration would affect the set of best fit parameters in the end.
Additionally, incorrect identification at few configurations may lead to significant
positional error at uncalibrated points. Therefore, this research employs coefficient
C < 1 in (12.8) to increase the numbers of iterations for errors identification at each
pose. At each pose, in each of the iteration, error vector E is multiplied influence
vector k = [k1 · · · k17 ] as:
P = J.(Ek), (12.7)
and subsequent error vector E is calculated as:
JT
E = C
· P. (12.8)
J JT
where, k = [k1 . . . k17 ] is obtained from the influence of kinematics parameters at a

pose as explained in the Sect. 3.1. For example, assume that the parameters influence
at one of the configuration is like average influence of kinematics parameters shown
in Fig. 12.3. In this case θ2 is the most influential with nearly 0.34 mm error leads
to k2 = 1. For this configuration, 0.34 mm is considered as 100%, and values
for the remaining ks in that configuration can be found with reference to k2 . Like
k8 = 0.59 for α2 . For some of the configurations, where only θ1 changes, vector k
remains same, otherwise changes with the configurations. The proposed approach for
error identification increases the computational cost, however, with the availability
of low cost and faster computing power, an accurate error identification is desired.
The following section performs experiments and identifies the errors in kinematics
parameters with both conventional and influence based error identification approach
with intensive experiments.
12.4 Experimental Results
The measurement setup includes a five DOF Katana 450 robot, an Optotrack system
with a volumetric resolution of 0.01 mm, active vibration isolation table, and a
computer to control the robot. The end link of the Katana 450 robot is 118 mm long
gripper, which is replaced with the 200 mm long and 0.5 kg tailored attachment. The
attachment imitates maximum payload of the robot, provide the ease for attaching
the measurement targets, and amplify the joint errors due to a larger length. The
robot is controlled with the MATLAB using Katana Native Interface language for
the calibration, and modification of the kinematic parameters after the calibration.
The digitising probe shown in the top-left corner of Fig. 12.4 used by the Optotrack
system, it is easy to establish the global coordinate system for the measurements.
The system measures Cartesian coordinates of three active markers on the established
global coordinates system at the structural base of the robot. The coordinates of three
markers are used to calculate the position as well as the orientation of the robot’s end-
Fig. 12.4 Experimental

setup
effector on the structural base of the robot as shown in Fig. 12.4. The translational
transformation of [55 55 201.5]’ mm transforms the coordinates of the structural base
to the robot’s nominal base as per design specification of the robot. The measurements
are sequenced such that all five joints angle change when moving from one pose to
another. Tables 12.2 and 12.3 lists errors of 17 kinematics parameters identified
with standard method and influence based approach respectively, and Table 12.4
compares the improvement in pose accuracy in term of various pose parameters.
The overall positional accuracy improves significantly using proposed method for
error identification. The current identification could reduce average positional error
from 1.21 mm to 0.38 mm whereas influence based identification reduced error
from 1.21 mm to 0.21 mm. Even though the orientation errors are not identified, the
measurements show improvement in orientation accuracy as well.
Table 12.2 Standard simultaneous identification

1 0 – −0.061 – –
2 89.92 – 0.0232 – –
3 0.003 190.003 −0.057 0.0021 –
4 0.007 139.01 0.0641 0.0013 –
5 −90.01 – −0.121 – 147.302
6 90.03 – – – 200.001
Table 12.3 Proposed influence based identification

1 0 – −0.061 – –
2 90.053 – 0.0341 – –
3 −0.016 190.00 0.066 0.005 –
4 0.062 139.00 0.0231 −0.003 –
5 −89.82 – −0.01 – 147.300
6 90.01 – – – 200.000
Table 12.4 Calibration results

Average over 118 positions
Pose parameter Before the calibration Simultaneous Influence based
identification identification
|X | 0.63 0.28 0.18
|Y | 0.44 0.13 0.10
|Z | 0.85 0.25 0.16
|P| 1.21 0.38 0.21
|Φ|◦ 0.27 0.096 0.088
|θ|◦ 0.17 0.027 0.022
|Ψ |◦ 0.26 0.084 0.079
12.5 Conclusion
The proposed approach for the identification of kinematics parameters errors has
proven to be effective compared to the standard one. Consideration of influence of
kinematics parameters during an error identification improved positional accuracy
of a robot by nearly 14%. This approach can be further developed for improving the
dynamic pose accuracy of the serial robotic manipulators.
Acknowledgements We would like to thank School of Mechanical Engineering, Adelaide for

providing the resources for this research.
References
1. Conrad, K.L., Shiakolas, P.S., Yih, T.: Robotic calibration issues: accuracy, repeatability and
calibration. In: Proceedings of the 8th Mediterranean Conference on Control & Automation,
Rio, Patras, GREECE (2000)
2. Renders, J.M., Rossignol, E., Becquet, M., Hanus, R.: Kinematic calibration and geometrical
parameter identification for robots. IEEE Trans. Robot. Autom. 7, 721–732 (1991). https://doi.
org/10.1109/70.105381
3. Mooring, B.W., Roth, Z.S., Driels, M.R.: Fundamentals of Manipulator Calibration. Wiley, New
York (1991)
4. Chen-Gang, C., Tong, L., Ming, C., Qing, J., Xu, S.: Review on kinematics calibration technology
of serial robots. Int. J. Precis. Eng. Manuf. 15, 1759–1774 (2014)
5. Nubiola, A., Bonev, I.A.: Absolute calibration of an ABB IRB 1600 robot using a laser tracker.
Robot. Comput.-Integr. Manuf. 29:236–245 (2013). https://doi.org/10.1016/j.rcim.2012.06.004
6. Wu, Y., Klimchik, A., Caro, S., Furet, B., Pashkevich, A.: Geometric calibration of industrial
robots using enhanced partial pose measurements and design of experiments. Robot. Comput.-
Integr. Manuf. 35, 151–168 (2015). https://doi.org/10.1016/j.rcim.2015.03.007
7. Johnsrud, V.: Improvement of the Positioning Accuracy of Industrial Robots. Norwegian Uni-
versity of Science and Technology, Masters (2014)
8. Roth, Z., Mooring, B., Ravani, B.: An overview of robot calibration. IEEE J. Robot. Autom. 3,
377–385 (1987). https://doi.org/10.1109/jra.1987.1087124
Chapter 13
Autonomous Welding Seam Detecting
and Tracking Using Vision and Sound
Sensors in Robotic Gas Metal Arc
Welding
Chao Chen, Gu Fang, Yanling Xu, Na Lv and Dinham Mitchell
Abstract At present, robotic welding requires the robot to be taught by a human

operator using teach and playback methods. When the weldment or seam is changed,
the prior off-line programming cannot be reused. The paper proposed an autonomous
welding seam detecting and tracking system. The system can detect realistic weld
joints and determine their position in the robot workspace by extracting seam feature
with image processing algorithms. The robotic trajectory is planned based on the
extracted seam feature. During the metal-inert gas (MIG) welding process, the system
can locate the welding position and track seam path by processing multi-sensor
information combined of dual- microphone array sound signals and passive vision
image from a CCD camera. When the deviation between welding pool centre and
seam centre line is detected, the system can eliminate the deviation by adjusting
the welding torch position. The experiments show that the proposed path planning
method can identify butt weld joints autonomously regardless of the base material,
surface finish and surface imperfections. Meanwhile, the multi-sensor system showed
its potential merit in quality monitoring and control for the MIG welding process,
whose deviation range of welding seam tracking accuracy is within ±0.17 mm.
Keywords Welding seam detection and path planning · Seam tracking · Sound
localization · Metal-inert gas robotic welding
C. Chen · Y. Xu (B) · N. Lv (B)

School of Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China
N. Lv
G. Fang · D. Mitchell
School of Computing, Engineering and Mathematics, Western Sydney University, Sydney,
Australia

156 C. Chen et al.
13.1 Introduction
Industrial robots have been widely used in welding manufacturing. It is always a

significant challenge to improve the automation level of robotic welding, which
can enhance the manufacturing efficiency and ensure welding quality in the end.
At present, the most common method to implement robotic welding is ‘teach and
playback’ technique by human operators. Although ‘teach and playback’ method has
unique advantages in standard mass production, when it comes to low to medium
volume manufacturing or even repair and maintenance work, it is very inefficient
because the process of ‘teach and playback’ takes a lot of setup time and is expense
to program welding path and adjust parameters. In addition, the programmed robot
is impossible to self-rectify any variation during the welding process. Therefore, it
is necessary to develop a new technique to enhance the automation level of robotic
welding, in order to decrease the expense and time of robotic offline planning and
self-rectify the deviation during the welding process.
The methods such as automatic welding path detection and planning, seam local-
ization and tracking are the key technique to solve the above problem. As for auto-
matic welding path detection and planning, the computer vision is a commonly used
method. In [1–3], the combined method of median filters, smoothing and binary
threshold segmentation techniques are applied to weld seam extraction during alu-
minium welding. However, different to aluminium which has an obvious colour
difference compared to the background, these methods are unsuitable for mild steel
plates because of the low contrast between workpiece and background. In [4], the
weldment was segmented out from the image through comparing the image with
and without the weldment in the same position. After image filtering, the noise left
from the subtraction process is eliminated. The welding seam is detected via using
the method introduced in [1]. Another method to detect the welding seam is the
application of pre-defined region of interest (ROI). By using the ROI, the algorithm
is able to focus on the welding seam and neglect the other useless information. In [5,
6], a ROI is placed in the centre of where the weldment is placed, so the background
can be ignored. The advantage of ROI is weakening the effect of the background.
However, for the robotic welding process, it’s not possible to place the weldment in
a fixed region due to occlusions from tooling and access limitations.
After the welding seam is detected and welding path is programmed, it is a key
point to ensure welding quality that the robot can follow the welding path. As the
welding path is easily influenced by factors such as parts distortion, spreading heat,
variable gap and thermal deformation etc. When the welding path changed, it is
very difficult to ensure the welding quality because the robot cannot follow the
changed path. So, to solve the problem, various sensors are used to monitor the
welding process, such as arc sensors [7], acoustic sensors [8, 9], ultrasonic sensors
[10] and vision sensors [11–13]. The most widely applied sensors are vision sensors
because the image contains huge information and the image processing method is
mature. In [11], a computer vision system using composite filtering technology was
established, and the clear welding pool images were captured during robotic-pulsed
13 Autonomous Welding Seam Detecting and Tracking Using Vision … 157
gas tungsten arc welding. In [12], a real-time seam tracking control technology base
on passive vision system was applied in robotic gas tungsten arc welding. In [13]
a robust algorithm was put forward for weld seam detecting and tracking based on
prior knowledge of weld seam in robotic gas tungsten arc welding.
Although vision sensors have many advantages in seam tracking, it still has its
limitations such as, the huge size of the camera may not be suitable for the complex
weld joint, the reliable 3D information may not be easy to obtain from the weld-
ing images. Therefore, it is necessary to use other sensors as an alternative or an
improvement.
There is a lot of researches about the acoustic sensor during the welding process.
In [14–16], Arata et al. probed to extract the arc sound feature and figure out the
influences of welding process parameters on arc sound. Many researchers tried to
find the relationship between the sound feature and welding penetration state. In
[17], a purpose-built real-time sensing system was designed to identify different
penetration via arc sound signals. In [18], a new feature extraction method for arc
sound signal during welding process was developed, which could extract 128 features
to identify the welding dynamic states in TIG welding. However, the sound signal
is a complex signal which may include various noise and useless information. Most
researches are based on single microphone which could acquire the part information
about welding process. We have considered to use more microphones to collect sound
signals and implement the seam localization.
In this paper, an autonomous welding seam detecting and tracking system is pro-
posed, which is capable of automatically detecting and planning welding path before
welding, and locating and tracing seam during welding process. Before the start of
welding, the system is capable of detecting realistic weld joints and calculating their
positions in the robot workspace with minimal human interaction. This is accom-
plished using a pair of calibrated robot mounted stereo cameras combined with image
processing algorithms. During the welding process, the system can locate welding
position and track seam path during MIG welding by processing multi-sensor infor-
mation combined of dual-microphone array sound signal and CCD camera passive
vision image.
The paper is organized into five sections. The system design is given in Sect. 13.2
which introduces the robotic welding system for MIG, by which we implement
welding seam detecting and path planning before welding and realize the welding
seam locating and tracking. The autonomous welding path detecting and planning
function was demonstrated in detail in Sect. 13.3 and the welding seam locating and
tracking function was elaborated in Sect. 13.4. Finally, the conclusion was provided
in Sect. 13.5.
158 C. Chen et al.
13.2 The Robotic Welding System for MIG
The setup of the whole experiment is shown in Fig. 13.1 [20]. It consists of a Fanuc
ArcMate 100iC six axis industrial robot, a welding system, an industrial PC, an
exhaust fan, a Blue Yeti Pro USB dual-microphone acoustic sensor, two eye-in-hand
uEye USB CCD camera with a resolution of 1280 × 1024 pixels mounted to the
(a)
Welding system Welding system
Welding machine
Welding Robot Exhaust fan
Welding process control

Noise generation
PC
Signal collection
Signal collection
Workpiece
Microphone Microphone
#1 Image collection #2
Visual sensor
(b)
Visual
Welding
Exhaust
Welding
Weldment
Fig. 13.1 The autonomous welding path detecting and planning system: a Schematic diagram
b physical diagram
The visual sensor used

for detecting welding seam
The visual sensor used

for tracing welding seam
Fig. 13.2 The visual sensor used in robotic welding system for MIG
robot wrist. The dual-microphones are placed in diagonal of work bench and are
fixed in the two sides of the centre of the diagonal and have a distance of 300 mm
to the centre. The welding exhaust fan comes from Ozone Pollution Technology.
The welding system includes the LINCON ELECTRIC, POWER WAVE F355i and
AUTODRIVE 4R90. The setup of visual sensor is shown in Fig. 13.2. The USB CCD
camera used for tracking the seam during welding is placed under the pair of stereo
cameras. The function of welding seam locating and tracking system is implemented
by multi-sensor, providing the welding vision information combined with the sound
signal from the dual-microphone array.
13.3 Welding Path Detecting and Planning
This section demonstrates the method introduced in [19], which automatically detects
the butt-jointed seam through computer vision. The method can be divided into three
parts as shown in Fig. 13.3.
1. Pre-processing is used to filter and smooth image and detect edge information.
Fig. 13.3 The welding seam

identification principle
160 C. Chen et al.
(a) (b)
Weldment Weldment
Workbench Workbench
Fig. 13.4 a Original image b the image after median filtering
2. The Hough Transform is applied to eliminate the background of the weldment.

3. The welding seam is extracted and reserved through remove other redundant part
in the image.
13.3.1 Pre-processing
The procedure of the image pre-processing can be divided into three part. The first part
is Gray transform which convert the colour image to a grayscale image using standard
conversion methods [19]. The second part is image filtering using a median filter,
which is utilized to eliminate the noise, such as scratches, salt and pepper noise and
other imperfect surface noise. The effect of the median filters is shown in Fig. 13.4
[19]. The last part is edge identification. In the paper, Sobel edge identification
algorithm is used for edge identification. The result is shown in Fig. 13.5 [19].
13.3.2 Background Segmentation
As demonstrated in Sect. 13.3.1, there is two method to segment the foreground from
the background. In [4], high contrast between the foreground and background was
used to segment images. In [5, 6], a pre-define ROI was utilized to extract the fixed part
of the image. In this paper, an automatically detecting foreground method is proposed
to segment the weldment from the background. The method is complemented through
the Hough lines. The idea of Hough transform is that a point in the original image
coordinate system corresponds to a straight line in the parameter coordinate system.
A straight line of the same parameter coordinate system corresponds to a point in
the original coordinate system, and then the original coordinate system assumes all
Fig. 13.5 Pre-processing

edge image
Weldment
Workbench
points on a straight line, their slope and intercept are the same, so they correspond
to the same point in the parameter coordinate system. So that after all points in
the original coordinate system are projected into the parametric coordinate system,
if there is aggregation point in the parameter coordinate system, the aggregation
point corresponds to the straight line in the original coordinate system. After Hough
transform is completed, the maximum aggregation point accumulator needs to be
counted to determine the threshold that is used to decide which Hough lines should
be reserved. This is demonstrated in Eq. 13.3.1.
T1 = β Hmax (13.3.1)
where T1 is the threshold, β is the threshold multiplier as the percentage of the

maximum aggregation point accumulator value Hmax . After several experiments, the
best value of β is 0.22. When β = 0.22, the result of the Hough transform is shown
in Fig. 13.6 [19].
However, there are unavoidable redundant lines reserved, as shown in Fig. 13.7
[19]. Some lines pass through the background. In order to eliminate the redundant
lines, a search window is used to determine which lines surround the weldment. Each
Hough line is utilized as a path where a series of sliding windows f and g of size
(w × w) are placed, as shown in Fig. 13.8 [19]. The search windows are utilized to
compare the grayscale value on the two sides of the Hough lines. It is obvious that
the more the grayscale difference is, the more possibility it is the line that surround
the weldment. The Eqs. 13.3.2 and 13.3.3 are used to calculate grayscale value on
either side of the Hough lines.

w
w
w w
f avgn = Ig (u n − − 1 + i, vn − − 1 + j)/(w2 ) (13.3.2)
i=−w j=−w
2 2
162 C. Chen et al.
Fig. 13.6 The result of

Hough transform (β = 0.22)
Weldment
Hough Lines
Workbench
Fig. 13.7 The segmenting

result of Hough lines
Redundant Lines
Weldment
Workbench

w
w
w w
gavgn = Ig (u n + + 1 + i, vn + + 1 + j)/(w2 ) (13.3.3)
i=−w j=−w
2 2
where u n and vn are the horizontal and vertical coordinates of Nth point on the Hough
lines respectively. w is the length of search windows. f avgn and gavgn are the average
value of f n windows and gn windows respectively. Then f avgn and gavgn need to be
normalized as the percentage of 255, which is the highest value of grayscale. The
normalized equation is as Eqs. 13.3.4 and 13.3.5. At last, the normalized value need
to be compared with each other through calculating the follow equations to determine
which lines should be reserved.
Fig. 13.8 Search windows

along with Hough line
Hough Line

N
f nor m n
n=1
F= (13.3.4)
N

N
gnor m n
n=1
G= (13.3.5)
N
D = |F − G| (13.3.6)

T r ue D > T2
line = (13.3.7)
False D ≤ T2
where N is the total number of the points on the Hough line. T 2 is the threshold that
is used to determine which line should be reserved. If the line is judged as true, it
will be reserved and vice versa. In the paper, a value of T2 = 10 can obtain good
effect. The grayscale value outside the reserved lines will be set to zero. The result
is shown in Fig. 13.9 [19].
13.3.3 Weld Joint Detection
As shown in Fig. 13.10, the operation of welding joint detection can be divided into
five sections. The Sobel edge detection algorithm is used again and its result is shown
in Fig. 13.11 [19]. The boundary can be found through applying a contour tracing
algorithm such as Moore Neighborhoods. The grayscale value of the boundary is set
to zero and is shown in Fig. 13.12 [19]. Due to the influence of reflect light, scratches
164 C. Chen et al.
Weldment
Fig. 13.9 Background elimination
Fig. 13.10 The principle of welding seam identification
or noise, the detected welding seam line may be discontinued. The image needs to
be processed by Close Operation, that is, to be dilated and then eroded. However
there are some small clusters of pixels caused by reflect light, scratches and image
noise. In order to decrease and eliminate the effect of these small clusters of pixels,
the method of Area filtering is applied to process the image. The redundant small
clusters of pixels can be eliminated by comparing their area to a threshold. If the
area is less than the threshold, the grayscale value the area will be set to zero. The
equation of area filters is as Eq. 13.3.8.
Fig. 13.11 The weldment

after the Sobel edge
detection
Noise
Seam
Boundary
Fig. 13.12 The image after

eliminating the boundary
Eliminating the boundary

1 Ar ea > T3
I = (13.3.8)
0 Ar ea ≤ T3
Area is the connected number of white pixels in the binary image. It is obvious
that the small white area will be removed by the Area Filters and its result is shown
as Fig. 13.13 [19]. The seam line is the longest line in the image and it can be
distinguished by calculating and comparing the connected number of white pixels
in the binary image. The final result is shown in Fig. 13.14 [19]. The welding seam
can be completely reserved and extracted as the base of the welding path planning.
166 C. Chen et al.
Fig. 13.13 The image after

area filtering
Eliminating the noise
Fig. 13.14 Detected

welding path
Extracting the seam
13.4 The Welding Seam Locating and Tracking System
In this session, the welding seam locating and tracking function is implemented
by a multi-sensor system, which is the combination of sound locating system and
visual tracking system. During the welding process, the planned path will change
due to parts distortion, spreading heat, variable gap, thermal deformation etc. The
combined sound locating and visual tracking system will assist robots to self-rectify
path according to the real-time weld seam condition.
13.4.1 Sound Locating
In [20], the implementation of sound locating function mainly consist of two parts.
The first part is the feature extraction of the welding sound signals from dual-
microphones. The second part is the establishment of the relationship model between
the sound signal feature and welding position. By using the method, the welding posi-
tion can be located through the analysis of the sound signals.
13.4.1.1 Signal Process and Feature Extraction
Because the two microphones were placed on the two side of centre point of diagonal,
when the welding position change, there are different changes on the sound signals
received by different microphones. As shown in Fig. 13.15 [20], when the welding
start in position A and end in position B, the detected welding sound pressure from
different microphone change into different trend. As the welding position gets close
to microphone 1, it gets far away from microphone 2. So the welding sound pressure
from the microphone 1 gets larger and the microphone 2 s gets smaller.
After several experiments and sound signals processing, the arc sound energy is
chosen as sound feature as there is a strong relationship between the welding position
and the sound energy as shown in Fig. 13.16 [20].
The increasing trend

Microphone 1
Sound Pressure /v
0.4
0.2
0
-0.2
-0.4
0 5 10 15 20 25 30
Time /s
The decreasing trend
Microphone 2
Sound Pressure /v
0.4
0.2
0
-0.2
-0.4
0 5 10 15 20 25 30
Time /s
Fig. 13.15 The change of sound pressure from dual-microphone

168 C. Chen et al.
200
Microphone 1
Microphone 2
150
The fitted lines
Energy 100
50
0
0 10 20 30
Time /s
Fig. 13.16 Arc sound energy from the dual-microphone
200 The five experiment data
150
Energy 2
100
50
200
0 100
10 5 0 y1
-5 -10 0 erg
Distance En
Fig. 13.17 The five experiment data about arc sound energy and welding position
13.4.1.2 Relation Model of Sound Feature and Welding Position
In order to obtain a more precise relationship between arc sound energy and welding
position, five experiments were conducted and analyzed. Five sets of data of arc
sound energy from dual-microphone, combined with welding position information, is
expressed in a 3D visualization as shown in Fig. 13.17 [20]. To obtain a relation model
of sound feature and welding position, the five data were combined by calculating
the average of each arc sound energy on specific distance through the Eq. 13.4.1.
The result is shown in Fig. 13.18 [20].

5
si ( j)
j=1
si = (13.4.1)
5
Fig. 13.18 The average The combined experiment data

value of the five experiment
data
S i (j) is the jth arc sound energy signal from the ith microphone and Si is the ith
average of the arc sound energy.
Through the analysis of Fig. 13.18, it is obvious that there is a strong space curve
relationship among energy 1, energy 2 and welding position. Therefore, in the paper,
a general 3D space line function is applied to fit these data points in Fig. 13.18:
y = β0 + β1 x1 + β2 x2 (13.4.2)
The value of β0 , β1 and β2 can be calculated through the method of linear regres-
sion. The result can be obtained as follows:
D(i) = −11.2439 + 0.2819S1 (i) − 0.1264S2 (i) (13.4.3)
where D(i) is the welding position. S1 (i) is the arc sound energy on the position
i from the sound signal of microphone 1 and S2 (i) is from the microphone 2’s.
β0 = −11.2439, β1 = 0.2819 and β2 = −0.1264.
Similarly, a quadratic Eq. (13.4.4) obtained by multiple regression analysis is
applied to fit the data as a further analysis of the validity of this prediction model.
The result shows that the both model can obtain good effect. The predicted result is
shown in Fig. 13.19 [20] and the error analysis table is shown in Table 13.1. From
the analysis of the predicted error, it is shown that the linear model has a better effect
than the quadratic model. Therefore, the linear model is more suitable for the welding
localization.
D(i) = −18.1589 + 0.6621S1 (i) − 0.2375S2 (i) − 0.0025S12 (i) + 0.0004S22 (i)
(13.4.4)
170 C. Chen et al.
The real welding seam
The error
The predicted position
Fig. 13.19 The predicted result of the two regression model
Table 13.1 Prediction error of the two models

First order Micro #1#2 (mm) Quadratic Micro #1#2 (mm)
Absolute error 0.6708 Absolute error 1.1901
Maximum error 5.6703 Maximum error 3.652
Minimum error 0.3458 Minimum error 1.9575
13.4.2 Visual Tracking
The implementation of visual tracking can be summarized in two parts, the extrac-
tion of the feature parameters and the adjustment of robot trajectory, which is demon-
strated in [21]. The extraction of the feature parameters mainly depends on the fast
and accurate image processing method. The adjustment of robot trajectory needs
real-time and smooth robot control algorithm.
13.4.2.1 The Extraction of the Feature Parameters
The aim of feature extraction based on image processing algorithm is to obtain the
weld pool centre O(ax , a y ) and the weld seam centreline equation y = kx + b. The
definition of the feature parameter is shown in Fig. 13.20 [21].
By the observation of the welding pool grey-scale map, it is found that there is
an obvious change of grayscale value along the welding pool edge. Therefore, the
Fig. 13.20 The definition of the feature parameters (centre point & centre line)
edge of welding pool can be extracted through a series of image processing method.
Further, the welding pool centre can be obtained by the analysis of the welding pool
edge. The flow of welding edge processing is shown in Fig. 13.21 [21]. After the
coordinate points of welding edge fitting are confirmed, the leftmost and rightmost
coordinate points can determine the welding pool centre by calculation the Middle
Point of the two end points. In a similar way, the flow of welding seam processing
is shown in Fig. 13.22 [21]. After the coordinate points of welding seam fitting are
confirmed, the two line equation are fitted to the left-edge and right-edge respectively,
as shown in Eq. 13.4.5.

fle f t (x) = k1 x + b1
(0 ≤ x ≤ n width ) (13.4.5)
fright (x) = k2 x + b2
Further, the seam centre line can be obtained as the equation y = k1 +k

2
2
x + b1 +b
2
2
.
According to the above image process, the welding pool centre point coordinate
and the seam centre line can be obtained. The deviation between the welding pool
centre and the seam centre line can be calculated as the parameter of robot trajectory
adjustment. The calculating formula of deviation is as Eq. 13.4.6.
k +k
1 2 ax − a y + b1 +b2
d(t) = 2

2
(13.4.6)
k1 +k2 2
2
+ (−1) 2
172 C. Chen et al.
Median filter Restoration ROI Window1
Improved Canny Removing false edge Edge scan
Edge fitting
Fig. 13.21 The flow of welding edge processing
Median Restoration ROI
Window2
Improved Canny Removing false edge Edge scan
Seam center fitting Edge fitting
Fig. 13.22 The flow of welding seam processing
From the above analysis, the deviation from welding pool centre to welding seam
centre line can be calculated real-timely and accurately. According to the deviation,
the robot trajectory can be self-rectify timely and precisely.
13.4.2.2 The Adjustment of Robot Trajectory
So as to adjust the robot welding position smoothly without affecting the welding
process, the method of piecewise interpolation was applied to adjust the robot tra-
jectory planning. The robot trajectory was segmented with an interval of 3–5 mm.
When the deviation between welding pool centre and seam centre line is detected,
the coordinate of next segment point will be recalculated to ensure the pool centre is
kept on the seam centre line. When the welding robot was shifted on the seam tracing
mode, the robot trajectory is segmented with a 3–5 mm interval and is controlled by
computer, as shown in Fig. 13.23 [21].
Every time the deviation occurs, it’s not a good idea to change the robotic trajectory
immediately because it will affect the smooth and accuracy of robot welding. A time
point that updates robotic trajectory position must be determined and before the time
point, the deviation need to be accumulated into the data segment which has not
moved yet, so as to ensure the smoothness of robot movement. In Fig. 13.24 [21],
assuming that the current position of the robot is at P0 , the (i − 1)th segment point

is P1 , the ith segment point is P2 , and the P0 is the Projection Point of P0 on the
original trajectory of robotic movement.

P P1
The formula M = P02 P1 is defined as the relationship function between robotic
trajectory and teaching trajectory. The calculating formula of M is as Eq. 13.4.7.
When M is less than 0, it indicates that the robot has not arrived the P1 yet and the
next position of the robot path shouldn’t be updated at this state. When 0 < M < 1,
it indicates that the robotic position is between the P0 and P1 . The next position point
should be updated according to the real deviation in this state. In this way, the robotic
movement can keep in a smooth and continuous pace.
(x0 − x1 )(x2 − x1 ) + (y0 − y1 )(y2 − y1 ) + (z 0 − z 1 )(z 2 − z 1 )

M= (13.4.7)
(x2 − x1 )2 + (y2 − y1 )2 + (z 2 − z 1 )2
Step
Start point End point
Fig. 13.23 The segment of robotic trajectory
The current position of robot
The segment point The segment point
The projection position
Fig. 13.24 Sketch map of welding position

174 C. Chen et al.
(a) (b)
teaching trajectory
welded
welding
Fig. 13.25 a Welding seam tracking experiments, b result after welding
0.3
X: 244
The deviation between wire and seam
Y: 0.2142
0.2
center d(t) / (mm)
0.1
-0.1
X: 279
Y: -0.09483
-0.2
0 50 100 150 200 250 300
Sample point
Fig. 13.26 The deviation between wire and seam centre
13.4.2.3 Verification Experiment
The verification experiment was designed as shown in Fig. 13.25a [21]. The teaching
trajectory is set to deviate 20 mm from the real weld seam. Through the developed
seam tracing algorithm, the real deviation from the welding path to seam is less
than 20 mm as shown in Fig. 13.25b [21]. According to the quantitative analysis of
experiment data, the maximum range of deviation is within ±0.21 mm, as shown in
Fig. 13.26 [21].
13.5 Conclusions
In this paper, an autonomous welding path detecting and tracking system in metal-
inert gas robotic welding is introduced. The following conclusions can be made:
(1) The proposed welding seam detecting method can detect welding seam of var-
ious shapes and sizes without any prior knowledge of the geometry or location
of the seam. The experiments show that the method can work well even in the
condition that the weldments have imperfect surfaces such as scratches and mill
scale. The system can plan the robot trajectory according to the extracted weld
seam feature.
(2) By analysing signals from dual-microphone array, it is found that the arc sound
energy can reflect the movement trend of welding position. According to the
features of arc sound energy, the relation model between the sound signal and
welding position can be established to implement welding localization. Then the
linear regression function and quadratic regression function were established to
predict the relation between the arc sound energy and welding position. The
absolute error can be controlled within the range of 0.67 mm. The experiments
show that the proposed sound localization method can locate the welding posi-
tion accurately.
(3) The image feature of weld pool centre point and weld seam centre line can
be extracted by a developed image process method. Their deviation can be
calculated accurately in real. According to the calculated deviation, the robotic
trajectory can be adjusted and self-rectified. The proposed seam tracking method
has obtained a good effect and can self-rectify the deviation between the taught
trajectory and the actual welding seam. The maximum range of final deviation
is controlled to be within ±0.21 mm.
Acknowledgements This work is partly supported by the Australian Research Council under
project ID LP0991108 and the Lincoln Electric Company (Australia), the National Natural Science
Foundation of China under the Grant No. 51405298 and 61401275, the State Key Laboratory of
Smart Manufacturing for Special Vehicles and Transmission System (GZ2016KF002).
References
1. Kong, M., Shi, F.H., Chen, S.B., Lin, T.: Recognition of the initial position of weld based on
the corner identification for welding robot in global environment. In: Tarn, T.J., et al. (eds.)
Robotic Welding Intelligence and Automation, LNCIS, 362, pp. 249–255. Springer Verlag,
Berlin, Heidelberg (2007)
2. Micallef, K., Fang, G., Dinham, M.: Automatic seam identification and path planning in robotic
welding. In: Tarn, T.J., Chen, S.B., Fang, G. (eds.) Robotic Welding Intelligence and Automa-
tion, LNEE, 88, pp. 23–32. Springer Verlag, Berlin, Heidelberg (2011)
3. Pachidis, T.P., Lygouras, J.N.: Vision-based path generation method for a robot based arc
welding system. J. Intell. Robot. Syst. 48(3), 307–331 (2007)
4. Chen, X.B., Chen, S.B., Lin, T.: Recognition of macroscopic seam for complex robotic welding
environment. In: Tarn, T.J., et al. (eds.) Robotic Welding, Intelligence and Automation, LNCIS,
vol. 362, pp. 171–178. Springer Verlag, Berlin, Heidelberg (2007)
176 C. Chen et al.
5. Dinham, M., Fang, G., Zou, J.J.: Experiments on automatic seam identification for a MIG
welding robot. In: Deng, H., et al. (eds.) Artificial Intelligence and Computational Intelligence,
LNAI 7003, pp. 390–397. Springer Verlag, Berlin, Heidelberg (2011)
6. Ryberg, A., Ericsson, M., Christiansson, A.K., Eriksson, K., Nilsson, J., Larsson, M.: Stereo
vision for path correction in off-line programmed robot welding. In: Proceedings of the 2010
IEEE International Conference on Industrial Technology, pp. 1700–1705 (2010)
7. Xu, Y.L., Zhong, J.Y., Ding, M.Y., Chen, H.B., Chen, S.B.: The acquisition and processing of
real-time information for height tracking of robotic GTAW process by arc sensor. Int. J. Adv.
Manuf. Technol. 65, 1031–1043 (2013)
8. Estochen, E.L., Neuman, C.P.: Application of acoustic sensors to robotic seam tracking. Ind.
Electron. 3, 219–224 (1984)
9. Lv, N., Xu, Y.L., Zhang, Z.F., Wang, J.F., Chen, B., Chen, S.B.: Audio sensing and modeling
of arc dynamic characteristic during pulsed Al alloy GTAW process. Sens. Rev. 33, 141–156
(2013)
10. Mahajan, A., Figueroa, F.: Intelligent seam tracking using ultrasonic sensors for robotic weld-
ing. Robotica 15, 275–281 (1997)
11. Chen, S.B., Zhang, Y., Qiu, T., Lin, T.: Robotic welding systems with vision sensing and
self-learning neuron control of arc weld dynamic process. J. Intell. Rob. Syst. 36, 191–208
(2003)
12. Xu, Y.L., Yu, H.W., Zhong, J.Y., Lin, T., Chen, S.B.: Real-time seam tracking control technology
during welding robot GTAW process based on passive vision sensor. J. Mater. Process. Technol.
212, 1654–1662 (2012)
13. Ye, Z., Fang, G., Chen, S.B., Dinham, M.: A robust algorithm for weld seam extraction based
on prior knowledge of weld seam. Sens. Rev. 33, 125–133 (2013)
14. Arata, Y., Inoue, K., Futamata, M., Toh, T.: Investigation on welding arc sound(Report I)—effect
of welding method and welding condition of welding arc sound. Trans. JWRI 8(1), 25–31 (1979)
15. Arata, Y., Inoue, K., Futamata, M., Toh, T.: Investigation on welding arc sound(Report
II)—evaluation by hearing acuity and some characteristics of arc sound. Trans. JWRI 8(2),
33–38 (1979)
16. Arata, Y., Inoue, K., Futamata, M., Toh, T.: Investigation on welding arc sound(Report
III)—effects of current waveforms on TIG welding arc sound. Trans. JWRI 9(2), 25–30 (1980)
17. Liu, L.J., Lan, H., Wen, J.L., Yu, Z.W.: Feature extraction of penetration arc sound in MIG
welding via wavelet packet frequency-band energy. Trans. China Weld. Inst. 31, 45–50 (2010)
18. Wang, J.F., Yu, H.D., Qian, Y.Z., Yang, R.Z., Chen, S.B.: Feature extraction in welding pene-
tration monitoring with arc sound signals. Proc. Inst. Mech. Eng. Part B: J. Eng. Manuf. 225(9),
1683–1691 (2011)
19. Dinhamn, M.: Autonomous Weld Joint Detection and Localisation Using Computer Vision
in Robotic Arc Welding. School of Computing, Engineering and Mathematics University of
Western Sydney (2013)
20. Na, L.V., Fang, Gu, Yan-ling, Xu, Hui, Zhao, Shan-ben, Chen, Ju-jia, Zou: Real-time monitoring
of welding path in pulse metal-inert gas robotic welding using a dual-microphone array. Int. J.
Adv. Manu. Technol. 90, 2955–2968 (2017)
21. Yanling, Xu, Na Lv, Gu, Du Fang, Shaofeng, Zhao, Wenjun, Ye, Zhen, Chen, Shanben: Welding
seam tracking in robotic gas metal arc welding. J. Mater. Process. Tech. 248, 18–30 (2017)
Chapter 14
Applications of a 3-Revolute Orientation
Sensing Mechanism (3-ROSM)
in Controlling a Camera
Shasa A. Antao, Vishnu S. Nair and Rajeevlochana G. Chittawadigi
Abstract In controlling the orientation of an object, it is always desired to have a

more intuitive form of controller or an input device as it helps the user adapt to it
at a faster pace, hence making it more productive. This paper focuses on the devel-
opment of a prototype and application of an intuitive serial chain mechanism to
measure orientation. This user driven mechanism, referred to as 3-Revolute Orienta-
tion Sensing Mechanism (3-ROSM), proposed elsewhere by the authors, is a serial
chain mechanism with three intersecting axes and has rotary encoders to measure the
joint angles. It is also gravity balanced to prevent reorientation of its links due to the
action of gravity. This reduces the effort needed by the user to move its links and also
its dependence on load bearing servo motors. It also emulates the joint system in a
human wrist and reduces the overall load acting on the user’s hand while gripping the
end-effector. The gravity compensation is achieved through the placement of coun-
terweights along and orthogonal to the joint axis under consideration. This makes
the 3-ROSM more adaptable to desirable or undesirable load variations. The addi-
tional features of this device are its simplicity, low cost, energy efficiency and wide
available workspace with which the user can operate this device. To demonstrate its
usability and versatility, an easy to assemble physical prototype of the 3-ROSM was
designed and developed, which has encoders mounted on its joint axes and was used
in two different applications. First, in controlling a 3-axis motorized camera mount
and second, navigating the position and orientation of the view-camera in a CAD
software.
Keywords Gravity compensation · Serial mechanism · Orientation control device
S. A. Antao · V. S. Nair · R. G. Chittawadigi (B)

Department of Mechanical Engineering, Amrita School of Engineering, Bengaluru, Amrita
Vishwa Vidyapeetham, Bengaluru 560035, India

178 S. A. Antao et al.
14.1 Introduction
In most 6-axis industrial robot arms, the first three axes can be actuated to bring the
arm to the desired position and the last three axes control the orientation of the wrist.
Therefore, it is easier to decouple the control of position and orientation separately
to reduce the complexity in using the controller. One type of frequently used position
and orientation controller is the teach pendant which incorporates the use of buttons
to increase or decrease the value of change in joint angles. Although this method
is simplistic, it lacks intuitiveness which makes the coordinated motion of the end-
effector difficult and also, there exists a high probability of operator disorientation
[1]. Therefore, an intuitive device to control the orientation is desirable.
Another type of orientation control is obtained with the use of parallel mecha-
nisms, in which the user can achieve high accuracy in orienting the end-effector.
Long and Collins [2] proposed the use of a six-DOF generalized master hand con-
troller using three pantograph linkages which provided high mobility and stiffness
in a compact design [2]. This was also demonstrated by Yoon and Ryu [3] in the
design of a haptic mechanism which utilized three servomotor driven pantograph
mechanisms to drive an end-effector widening the effective workspace of the device
[3]. An original design proposed by Birglen et al. [4] incorporated the use of spher-
ical links and revolute joints with the joint axes intersecting at a central point in the
end-effector [4]. This device achieves pure rotation around a point located in the
user’s hand but has a restricted workspace. However, parallel mechanisms in general
have complex kinematics and insurmountable kinematic singularities [5]. In contrast,
serial mechanisms have simple constructions and the forward kinematic problem is
easier to solve.
A successful input device should not only achieve accurate position and orientation
control but should also neutralize moment acting on the links which may be caused
by moving links or by gravity. This can be achieved through static balancing, also
known as gravity compensation which is achieved when the total force acting on
the fixed link due to the entire mechanism does not produce any torque under static
conditions [6]. This can be achieved passively through the use of counterweights
or springs [7–11], or actively through the use of pneumatic or hydraulic actuators.
Gravity compensation also reduces the effort required to move the links making it
easier to control while conserving energy by reducing dependency on the external
actuators.
A modified Gimbal mechanism, shown in Fig. 14.1a, having three intersecting
joints can be used to measure the orientation of the end-effector. This can be achieved
if the individual joint angles can be measured when the user rotates the end-effector
sphere about its center. However, once the user ceases to hold the end-effector, the
mechanism would move to its most stable configuration under the action of gravity.
Hence, a gravity balancing method to avoid this is desirable.
The first and the last author proposed the 3-ROSM (3 Revolute Orientation Sensing
Mechanism), in which passive gravity balancing was achieved for the mechanism
shown in Fig. 14.1a by modifying two links and placing counterweights at appropriate
14 Applications of a 3-Revolute Orientation Sensing Mechanism … 179
Fig. 14.1 CAD model of 3-ROSM (Antao et al. [12])
places [12]. As shown in Fig. 14.1b, the links of the mechanism denoted as Link1,
Link2 and Link3 move about a fixed link (Link0 or Ground) and the axes of its
respective joints (denoted as Joint1, Joint2 and Joint3). The three joints intersect at a
point located at the center of the spherical gripper attached to Link3, also called the
end-effector (EE). When the configuration of the device is changed, the centre of the
sphere on the EE remains at the same position while its orientation changes according
to the change in joint angles. The main idea was to make the net moment about a
joint axis as zero such that for any joint angle, a particular joint would not have any
motion due to gravity. As Link3 is axis symmetric, there was no need to modify
its shape. Link1 and Link2 were modified to have a T-shaped extension on which
counterweights can be mounted. If the masses are known, the distances at which they
have to be placed can be determined from simple moment balance equations about
the joint axis as reported in [12].
The CAD model of the proposed mechanism with counterweights was subjected
to gravity test in Autodesk Inventor Dynamic Simulation module and it was verified
that the mechanism was gravity compensated.
After the verification of its principles using dynamic simulation, the first proto-
type of the mechanism was developed to propose and validate a method to determine
the positions of the counterweights in a physical prototype and has been submitted
elsewhere [13]. The second prototype, which is considered in this paper, is a 3D
printed version which has rotary encoders to measure the joint angles, and incor-
porates passive balancing. Section 14.2 of the paper gives a brief overview on the
kinematics of the mechanism followed by the procedure for obtaining the passive
balance in Sect. 14.4. Subsequently in Sect. 14.5, the applications of the mechanism
in controlling a 3-axis camera mount and controlling the position and orientation of
the camera in a CAD software are elaborated, followed by the conclusions.
14.2 Kinematic Analysis
The current prototype of the 3-ROSM has hollow bore rotary potentiometers as
encoders to measure the joint angles, but in order to provide smooth control of the
manipulator (ex. a camera mount) the joints of the prototype have to be mapped. As
mentioned above, the axes of the joints of the 3-ROSM intersect at the center point
in the end-effector, which being invariant enables the calculation of change in joint
angles with the help of forward kinematics. To aid with the study of the behavior
of each joint with respect to the other, the design architecture of the 3-ROSM is
represented using the Denavit Hartenberg (DH) representation which is a commonly
used convention for selecting the frames of reference in robotics. The nomenclature
and methodology described in [14] is followed in this paper. A single DH coordinate
frame consists of four different DH parameters which are joint angle (θ ), link offset
(b), link twist (α) and link length (A) which give the geometric relationship between
any two consecutive DH frames. Hence, they can collectively represent the position
and orientation of any link in the system with respect to the DH coordinate frame
attached on the previous link. Referring to Fig. 14.2a, the determination of the DH
frames located on the 3-ROSM begins with assuming the orientation of Link 0 along
the axis of Joint1 as the position of the EE remains invariant. As a result, the Z axis
of frame1 is taken along Joint axis 1 while the X axis is arbitrary. The Y axis is the
cross product of the Z and X axes. As per DH representation convention the Z axis
of the consequent frame (Frame2 ) must be assumed along the direction of the joint
axis of Joint1 which is again along Z 1 (Z-axis of Frame1 ) direction resulting in a 0
twist angle, and X 2 is chosen to be orthogonal to Z 2 . The difference in angle between
X 2 and X 1 is considered as the Joint angle θ 1 which is a variable in this case. Hence,
through the choice of the coordinate axes, the number of parameters needed reduces
making the further assignment of DH frames simpler. Frame3 with respect to Frame2
Fig. 14.2 DH frames attached to a prototype of the 3-ROSM

Table 14.1 DH parameters of the 3-R orientation sensing mechanism

Parameter joint Joint offset (bi ) Joint angle (θi ) Link length (ai ) Twist angle (αi )
Joint1 0 θ1 (variable) 0 0°
Joint2 b2 θ2 (variable) 0 −90°
Joint3 0 θ3 (variable) 0 −90°
is obtained on the central point of the EE along the axis of Joint2, with a joint offset
of b2 which is the distance between the central point and the fixed point on Link0. To
orient Z3 along the axis of Joint 2, a twist angle of 90° in the opposite direction must
be applied and X3 is taken as the common normal between Z2 and Z3 . To further
simplify the assignment of frames, the direction of X3 is taken as (Z3 × Z2 ) and not
(Z2 × Z3 ). This is also due to constraints placed on the prototype of the 3-ROSM in
setting its initial configuration. Similarly, the axis of Joint3 intersects with Joint2 and
the DH frame4 is attached, as shown in Fig. 14.2b. The values of the DH parameters
between all four DH frames are listed in Table 14.1.
A Homogeneous Transformation Matrix (HTM) can be used to represent the
configuration between two consecutive DH coordinate frames. Hence the HTM of
the final Frame4 with respect to the initial Frame1 is given by the product of the
HTMs on each individual joint denoted by T1 , T2 and T3 [12], as
⎡ ⎤
cos(θ1 + θ2 ) cos θ3 sin(θ1 + θ2 ) cos(θ1 + θ2 ) sin θ3 0
⎢ sin(θ1 + θ2 ) cos θ3 − cos(θ1 − θ2 ) sin(θ1 − θ2 ) sin θ3 0⎥
T = T1 T2 T3 = ⎢
⎣
⎥
sin θ3 0 − cos θ3 b2 ⎦
0 0 0 1
(14.1)
In the above HTM, the elements in the first three rows and columns represent the
direction-cosines corresponding to the orientation between Frame4 and Frame1 . The
elements in the fourth column and first three rows represent the position vector from
the origin of Frame1 –Frame4 , represented in Frame1 , which remains invariant with
respect to any change in joint angles, i.e., the position of the origin of Frame4 with
respect to Frame1 does not vary for any configuration/orientation of the 3-ROSM.
Equation (14.1) refers to the forward kinematic analysis which is the determination
of EE configuration for given joint angles. Based on the joint angles, the control of
the orientation of a camera can be controlled as explained in Sect. 14.4. The authors
acknowledge the presence of a singularity when the axis of Joint3 aligns with Joint1
resulting in the loss of a degree-of-freedom. This issue will be addressed in further
research on modification of the initial design of the 3-ROSM.
14.3 Passive Balancing of the 3-ROSM
Another feature of the 3-ROSM is that the mechanism is gravity compensated. This
is to stop the mechanism from reorienting itself to a stable configuration when the
user releases the EE, and also reduces the effort needed by the user to move the links.
Passive balancing, through the use of counterweights is performed about the joints
of the 3-ROSM. The theory applied in this mechanism is that the moment applied
by the links due to gravity is countered by an equal and opposite moment applied by
the counterweights. To accommodate the counterweights, the initial system of the
mechanism was modified by placing T-shaped extensions which enable the counter-
weights to be placed along and orthogonal to the joint axis and thereby reduce the
complexity of the balancing problem from 3D to two problems in 2D. Passive bal-
ancing using this methodology was first conducted on a CAD model as explained by
Antao et al. [12] and subsequently, an initial physical prototype was designed to test
and validate this method and also posed challenges in finding the practical centre of
mass which is necessary due to several inaccuracies arising in Centre of Gravity (CG)
calculations, which may be due to manufacturing irregularities, material defects etc.
The method applied in finding the CG of the various links of the 3-ROSM is referred
to as the Hanging Method [15]. This method involves suspending a link by different
points from a fixture and dropping a vertical plumb line from the same fixture. The
intersection of two plumb lines for different orientations of the link is taken as its CG.
Also, a third plumb line is also dropped for a different link orientation to validate the
intersection point. The plumb line and threads (considered mass less) by which the
link is hung, should be in line and parallax must be avoided to produce an accurate
result. In some cases, the CG does not lie on the link itself but off the link, on the same
plane. To map these types of points, the link is attached to a paper (considered mass
less) which acts as the plane upon which the CG can be plotted with the help of the
Hanging Method. The research involved in the development of this initial prototype
with the correlation of CG values found by theoretical means and using the hanging
method, is reported in [13].
Thereafter, the Hanging method was also used to find the CG locations in the
current prototype developed by 3D printing technology. A few images taken while
finding the CG location are shown in Fig. 14.3. The encoders were placed on the links
such that they are also symmetric about the same plane as the links of the prototype.
After locating the CG locations, the values were put into moment balancing equations
[12]. The mass of the counterweight were predefined and only the locations on the
T-shaped extensions had to be found. It should be noted that due to light weight of
the 3D printed model of the 3-ROSM, changes in the mass such as addition of wires
to the encoders, affect the overall CG locations. For this purpose, threads were made
on the T-shaped extension links to fine tune the location of the counterweights, which
is also a novel contribution of this mechanism.
The prototype with counterweights was then mounted onto a bearing and then
attached to a box. As illustrated in Fig. 14.4, the orientation of the mechanism does
Fig. 14.3 Hanging method performed on Link2 and Link3 of the 3-ROSM
Fig. 14.4 Passively balanced prototype of the 3-ROSM
not change if user releases the EE at any configuration, thus verifying the passive
balancing of the mechanism.
14.4 Applications of the 3-ROSM
Hollow bore encoders by Tyco Electronics were used to measure the joint angles
of the prototype. These have a voltage divider circuit in the form of a rotary poten-
tiometer and the angle varies linearly with the output voltage. The encoders on the
joint axis were connected to an Arduino microprocessor, which reads the value of
the encoder (0–1023). The value was then mapped to give the corresponding angle
of the corresponding joint. Each joint had a different mapping as the initial value
and the way the encoder was attached was different, as illustrated in Fig. 14.5. The
encoders had a range of motion of 300º and in the prototype, only 270° motion was
allowed through the usage of physical stoppers.
14.4.1 Motorized 3-Axis Camera Mount
The prototype of the 3-ROSM was initially set to control the orientation of a 3-axis
camera mount (Pan- Tilt-Roll) as a preliminary test of its controller capabilities. The
movement of the joints of the camera mount is controlled by servo motors whose axes
intersect with each other emulating the 3-ROSM i.e., this is a kinematically equivalent
application because there exists a 1:1 mapping of the change in joint angle between
the controller and the camera mount. The servo motors were actuated using the
input data from the encoders, which is programmed using the Arduino software, as
illustrated in Fig. 14.5. The goal of this application was to obtain smooth transition
of change in joint angles and to make it as intuitive as possible for the user. One
of the challenges faced in this application was the limited range of orientation of
the camera mount as the servo motors are constrained to a 180° rotation while the
available rotation is 270° (Fig. 14.6).
Fig. 14.5 Mapping of encoder value and joint angles
Fig. 14.6 Control of a

3-axis camera mount using
the 3-ROSM
Fig. 14.7 Circuit diagram and flowchart for controlling a 3-axis camera mount using the 3-ROSM
The camera mount and the orientation sensing mechanism prototype are shown
in Fig. 14.7. The camera was successfully controlled and once the user ceases to
hold the EE of the orientation sensor, the camera’s motion would also stop. Thus,
controlling a camera on a 3-axis mount can be one of the applications of the proposed
prototype.
14.4.2 Camera Control in CAD Software
CAD software with a 3D modelling capability has a provision to move the cam-
era view around the object of interest in the virtual environment. This is typically
achieved by using a mouse by the user. The camera location can be zoomed, panned
and rotated such that an object can be viewed from a different position. The posi-
tion and the orientation of the camera can also be controlled by using the proposed
mechanism. Most CAD software have their API (Application Programming Inter-
face) through which certain automation/customization can be done by developing
programs.
Autodesk Inventor is one such CAD software whose API can be called using
Visual C# programming language. Here, a C# Windows application was developed
which received the joint angle information from Arduino through a USB interface.
The angles received were used in Eq. (14.1) to determine the orientation of the EE
of the mechanism. The orientation of the EE was mapped to the camera in the CAD
software such that there was a 1:1 mapping between the motion of the EE and that of
Fig. 14.8 Controlling the configuration of the camera in a CAD software using the 3-ROSM
the camera in the CAD software. Using the API, the camera location and orientation
was modified according to the joint angles received.
Images of two different orientations of the input device and the corresponding
camera configuration in the CAD software are shown in Fig. 14.8. The control of the
camera location was found to be intuitive and thus, the proposed mechanism can be
used to intuitively control the camera motion in a CAD software.
14.5 Conclusions
The authors have developed a prototype of the 3-ROSM and presented its applications
in this paper. The device has encoders placed at its joints and by measuring the
joint angles, the orientation of an object, both physical and virtual, can be changed
accordingly. The device also has passive gravity balancing capabilities, thus requiring
lesser effort to use and is also apt for applications where the orientation of the object
has to be controlled and should not change once the user stops controlling it. The
developed prototype was used to successfully control a 3-axis camera mount and
also to move the camera around in a CAD software environment.
References
1. Brooks, T.L., Bejczy, A.K.: Hand Controllers for Teleoperation, A State-of-the-Art Technology
Survey and Evaluation, p. 84. JPL Publications, California (1985)
2. Long, G.L., Collins, C.L.: A pantograph linkage parallel platform master hand controller for
force-reflection. In: IEEE International Conference on Robotics and Automation, pp. 390–395
(1992)
3. Yoon, J., Ryu, J.: Design, fabrication, and evaluation of a new haptic device using a parallel
mechanism. Proceedings of IEEE/ASME Transactions on Mechatronics 6(3), 221–233 (2001)
4. Birglen, L., Gosselin, C., Pouliot, N., Monsarrat, B., Laliberté, T., SHaDe, A.: New 3-DOF
haptic device. IEEE Trans. Robot. Autom. 18(2), 166–175 (2002)
5. Lenarcic, J., Bajd, T., Stanišić, M.M.: Robot Mechanisms, p. 242 Springer Science & Business
Media (2012)
6. Gosselin, C.: Gravity compensation, static balancing and dynamic balancing of parallel mech-
anisms. In: Smart Devices and Machines for Advanced Manufacturing, pp. 27–48. Springer,
London (2008)
7. Azadi, S., Moradi, M., Esmaili, A.: Optimal balancing of PUMA-Like robot in predefined path.
J. Sci. Ind. Res. 74(04), 209–211 (2015)
8. Cheng, Z., Foong, S., Sun, D., Tan, U.X.: Towards a multi-DOF passive balancing mechanism
for upper limbs. In: Proceedings of 2015 IEEE International Conference on Rehabilitation
Robotics (ICORR), pp. 508–513, Singapore, 11 Aug 2015
9. Briot, S., Arakelian, V.A.: New energy-free gravity-compensation adaptive system for balanc-
ing of 4-DOF robot manipulators with variable payloads. In: Proceedings of Fourteenth Inter-
national Federation for the Promotion of Mechanism and Machine Science World Congress
(2015 IFToMM World Congress), Taipei, Taiwan, 25 Oct 2015
10. Lessard, S., Bonev, I.A., Bigras, P., Briot, S., Arakelian, V.: Optimum static balancing of the
parallel robot for medical 3D-ultrasound imaging. In: Proceedings of 12th World Congress in
Mechanism and Machine Science, Besançon, France, 18 Jun 2007
11. Whitney, J.P., Hodgins, J.K.: A passively safe and gravity-counterbalanced anthropomorphic
robot arm. In: Proceedings of 2014 IEEE International Conference on Robotics and Automation
(ICRA), pp. 6168–6173, Hong Kong, China, 31 May2014
12. Antao, S.A., Vishal, S., Rajan, S., Nair, V., Chittawadigi, R.G.: Passive balancing of a 3-R
orientation sensing mechanism. In: Proceedings of the 8th Asian Conference on Multibody
Dynamics ACMD, Kanazawa, Japan, 8 Aug 2016
13. Antao, S.A., Nair, V.S., Chittawadigi, R.G.: Methodology to determine counterweights for
passive balancing of 3-R orientation sensing mechanism using hanging method. In: Submitted
to the 3rd International and 18th National Conference on Machines and Mechanisms, Mumbai,
India, 13–15 Dec 2017
14. Saha, S.K.: Introduction to Robotics. Tata Mc-Graw Hill, Delhi, India (2014)
15. Koizumi, T., Tsujiuchi, N., Mori, K., Shibayama, T.: Identification of the center-of-gravity
and inertia terms of 3-dimensional body. In: Proceedings of the IMAC-XXV Conference &
Exposition on Structural Dynamics, Orlando, Florida, 19 Feb 2007
Chapter 15
Contrast Enhancement Using Region
Based Image Fusion in Wavelet Domain
S. Sharma, J. J. Zou and G. Fang
Abstract Contrast enhancement is an important research topic to improve visual

quality of the image. In this work a region-based image fusion method has been
applied in the wavelet domain. In addition, a technique for contrast enhancement
has been incorporated to make the final fused image look crisper, sharper and clearer
from the perspective of the human visual system. The proposed method has been able
to reduce the computational complexity, processing time, and it shows improvement
in both objective and subjective assessment when compared to our previous work.
Keywords Image fusion · Wavelet transform · Region · Processing time ·

Contrast enhancement
15.1 Introduction
The image fusion is a process of combining multiple images into an image in such a
that it not only removes redundant information, integrates complementary informa-
tion and increases the amount of image information, but also removes noise, improves
contrast and visual effect, enhances the scenic details, etc. The Human Visual Sys-
tem (HVS) is good at identifying salient features, such as colors, edges, contrast,
etc., present in an image. Devices nowadays have a limited set of colors which are
required to represent the color or grayscale image at a more vibrant level. This can
deteriorates the image quality, which leads to an image that can be unpleasant to
human eyes.
In general the image fusion techniques fall under two domains, one is the spatial
domain and the other is transform domain [1]. The spatial domain-based approach is
straightforward and includes methods such as taking average, maximum or minimum
of the source image’s pixel intensity. However, along with its simplicity, this tech-
nique generates side effects such as unwanted noise, artifacts, and reduced contrast
S. Sharma (B) · J. J. Zou · G. Fang

School of Computing, Engineering and Mathematics, Western Sydney University, Penrith,
NSW 2751, Australia

190 S. Sharma et al.
[1]. To eliminates the shortcomings of the spatial domain approach, the transform
domain approach was suggested in [2], where a pixel-wise comparison was made
between two source images and the ultimate result was gained after processing every
single pixel present in both images. The method in [2] worked well in improving
the quality of a fused image and enhancing the contrast. However, the method was
lengthy in terms of processing time. In order to reduce the computational complexity
and for further improvements in the contrast, a region-based image fusion technique
in transform domain has been proposed in this paper.
The rest of the paper is organized as follows. Section 15.2 gives a brief account of
the related work and our previous work. Section 15.3 discusses the proposed method.
Section 15.4 presents the experimental results which have been obtained using MAT-
LAB, and discussions of the results, followed by a conclusion in Sect. 15.5.
15.2 Related Work
In [2], Nirmala et al. have proposed a fusion method based on standard deviation.
Their method uses DWT (Discrete Wavelet Transforms) to achieve multi-level image
fusion and proposes that standard deviation can be performed on the approximate
coefficients before the final fused image is reconstructed using IDWT (Inverse Dis-
crete Wavelet Transforms). The method is novel, and that motivated our previous
work on contrast enhancement in image fusion.
In [3] a robust sparse representation (RSR) model to extract detailed information
in a set of input images has been proposed. It has been achieved by replacing the
conventional least squared reconstruction error with a so called sparse reconstruc-
tion error. In [3] the local information from each input image patch and its spatial
contextual information are collaboratively employed to determine the focused and
defocused regions in multi-focus input images. In [4] the author proposed a novel
boundary finding based multi-focus image fusion algorithm, in which the task of
detecting the focused regions is treated as finding the boundaries between the focused
and defocused regions from the source images. As these recent works have preferred
region-based processing over the pixel-based processing so they have served as a
motivation for the work proposed in this paper.
Two major shortcomings which have been identified in [1, 2] are:
1. Loss of contrast: A shift in the input signal can lead to a significant variation
in the energy distribution of the DWT coefficients at different scales which can
result in contrast loss [5].
2. Loss of processing time: As our previous work is based on a pixel-based process-
ing, so each and every pixel present in the source images needs to be processed
before the right selection is made. Hence this process is very lengthy and time-
consuming.
15 Contrast Enhancement Using Region Based Image Fusion … 191
15.3 Proposed Method
Wavelet based image fusion is achieved by performing the wavelet transform on

input images in addition to a fusion rule. A subsequent step includes performing an
inverse wavelet transform ω−1 to retrieve the final fused image. This process can be
expressed below in Eq. (15.1).
C = ω−1 (φ(ω(A), ω(B))) (15.1)
In the proposed method, we have dealt with the region of the image rather than
dealing with the actual pixels as was done in [2]. The word ‘region’ here stands for a
group of pixels combined together. There are various region-based image fusion algo-
rithms available but we have concentrated mostly on the region based image fusion
using energy estimation [6]. Two registered source images are shown in Fig. 15.1, out
of which one image is out of focus on the right-hand side and the other image is out
of focus on the left-hand side. They are considered as inputs and DWT is performed
on them to extract the wavelet coefficients by using Eq. (15.1), once we have the
coefficients they are clubbed under a 3 × 3 mask. Once we have created our desired
mask it is convolved with both of the input images simultaneously. And at every step
it computes the energy of the region that comes under the mask (i.e. squaring and
adding all the coefficient values that come under the mask) and once the energy has
been computed for a particular region in both the images (let’s say image A and B),
a comparison among these regions is made using a fusion rule.
15.3.1 Fusion Rule Used
The proposed method uses an absolute maximum selection rule in the transform
domain [7]. Let A(i, j) and B(i, j) be the two images to be fused and their approx-
imate wavelet coefficients corresponding to low-frequency subbands are C A A(i, j)
and C AB(i, j) respectively. So, a 3 × 3 mask for these coefficients is defined and the
energy associated with this mask can be calculated using Eq. (15.2), where M and N
stand for the total rows and columns number of the image and the size of an image is
M × N , i and j are the row and column location of the pixel, n represents the centre
point of the mask. Here α stands for the value of wavelet coefficients corresponding
to separate input images.
For i = 1 : M
For j = 1 : N

1
EC A A(i, j) = α A A (i + n, j + n)2
n=−1
(a)
(b) (c)
Fig. 15.1 Parent images (a) and (b) are fused together to give image (c) and image (d) Although
image (c) has less noise in comparison to both the parent images, it has lost sharpness and contrast.
Image (d) has been achieved by using a contrast enhancement technique while achieving image
fusion. In comparison to image (c) it has further suppressed the noise and has also maintained better
contrast characteristics (yellow oval) along with better detail preservation (red rectangle)

1
EC AB(i, j) = α AB (i + n, j + n)2 (15.2)
n=−1
Once the energy is calculated then both the regions can be compared by the
absolute maximum rule in the following manner.

EC A A(i, j) if |EC A A| ≥ |EC AB|
EC AF(i, j) = (15.3)
EC AB(i, j) if |EC AB| ≥ |EC A A|
In Eq. (15.3), EC AF(i, j) is the energy value of the region that will be present
in the final fused image. Same steps have been repeated until the mask has fully
mapped both input images and the final image C is filled up completely. Similarly,
the coefficients corresponding to the high frequency subbands or detail coefficients
are processed by the defined fusion rule and are made present in the final image.
To achieve the proposed region based fusion DWT is carried up to 3 levels as it was
done in our previous work in [2]. It can be observed that the fused image turned to be
better than the input images but still had a certain level of blurriness associated with
it. To eliminate this shortcoming, the approximate coefficient at level 1 are scaled
by multiplying it by a factor of 0.75, which is selected empirically. As a result, the
overall energy within the fused images has reduced. This makes the image to appear
darker although the features present within become more prominent and definite as
illustrated in Fig. 15.1. Once the features are captured improvement in contrast can
be addressed so that the final image appears to be sharper and crisper. This has been
achieved by using an image enhancement techniques, i.e., Gamma Correction [8]
given as:
S = P ∗ Rγ
where S is output pixel value, R is an input pixel value and P & γ are non-negative
real numbers. We have empirically defined γ = 1.1.
15.3.2 Flowchart for the Proposed Method
As indicated in Fig. 15.2, in the proposed method, firstly, we read two test images,
followed by performing DWT on three levels that give us approximate and detail
coefficients respectively. The next step is to generate masks and calculate energies
of the defined regions. The following step is the comparisons between the regions
comprising approximate coefficients i.e. ECAA and ECAB. This results in the gen-
eration of the region corresponding to ECAF using the maximum fusion rule in the
final image. In a similar fashion, we compare the rest of the regions comprising detail
coefficients of the images A and B. After analysing and generating the appropriate
regions, we can directly reconstruct the image to get the output fused image using
IDWT. Finally, we can use the power law to enhance the contrast to get the final
fused image.
15.4 Experimental Results
Test images which have been used to demonstrate the research work are shown
in Fig. 15.3. Yellow circles in Fig. 15.3 represent out of focus or blurry regions.
Figure 15.4 compares the proposed method with our previous method and an existing
method. The left column represents the results from our previous work in [2]. In the
left column both of the reconstructed images has less grayish and blurry appearance
in comparison to the images in the far right column. The far-right column in Fig. 15.4
demonstrates the results of method proposed in [1] and shows that the reconstructed
Fig. 15.2 Flowchart for the proposed region-based method
Fig. 15.3 Test images from left to right: plane and Lena
images possess a uniform gray texture. As a result, the reconstructed images lack
in edges and features as images look blurry and soft. The results of the proposed
method can be seen in the middle column of Fig. 15.4. An overall improvement
in contrast can be seen in comparison to the other methods. An improvement, in
contrast, makes the presence of feature more evident and improves the quality of the
edges present, which are considered as details of an image. Red markings are used
in the middle column to illustrate to the improved features and edges as a result of
the proposed method. In Fig. 15.4b, the hat linings are more prominent in contrast
as depicted by the red rectangle and the yellow oval illustrates the clarity in hair
follicles. In Fig. 15.4a, the red rectangles also project the visible improvements in
contrast, which makes the markings on the surface to look more crisp and clear.
The quality parameters taken into consideration are Peak, Signal to Noise Ratio
(PSNR) and entropy [9]. If I (i, j) represents the grey-level in input image at the ith
row and jth column and D(i, j) stands for the value in the output image, then an error
e(i, j) is defined as e(i, j) = I (i, j) − D(i, j). The mean- squared-error (MSE) is
defined as:
Fig. 15.4 Comparison of reconstructed images from left column to right column: method used in
[2], the proposed method, method used in [1]
1
MSE = [I (i, j) − D(i, j)]2 (15.4)
M N 0≤i<M
0≤ j<N
where M and N are the numbers of rows and columns of the image. Once MSE is
Max 2
calculated, PSNR can be calculated using, P S N R = 10 log10 M S Ei where Maxi =
255. PSNR should be as high as possible. Table 15.1, makes it evident that the
proposed method performs better in terms of PSNR than the other two methods.
The entropy [9] of an image is the measure of the information contained in the
fused image. Higher values of entropy indicatethat the fused image contains more
information. The entropy is given by E = − L−1 L=0 Pl log2 Pl , where L represents
Table 15.1 Comparisons among the existing methods and the proposed method
Test image Method Time (seconds) PSNR (dB) Entropy
PLANE Method in [1] 7.72 34.2217 7.7028
Proposed method 5.39 34.8293 7.7235
Method in [2] 6.43 33.7641 7.5623
LENA Method in [1] 7.54 35.5491 7.6423
Proposed method 5.32 35.8458 7.6649
Method in [2] 6.24 34.8361 7.4456
the number of gray levels, is the ratio between the number of pixels with gray level
Pl and the total number of pixels with gray level l and the total number of pixels.
Table 15.1 shows that the entropy is higher in the proposed method when compared
to the other methods. As the entropy value is higher in the proposed method, it means
that more information is contained in the fused image. In addition, the computational
time taken by the proposed method is less and hence it is more time efficient.
15.5 Conclusion
In this research, we have been able to enhance the contrast in image by fusion. The
proposed method produces a superior fused image with a reduction in processing
time in comparison to the other two existing methods.
The proposed method can be extended to enhancing coloured images. The pro-
posed method can also be extended to make further improvements in selecting regions
or masks for processing.
References
1. Paramanandham, N., Rajendran, K. (eds.): Proceedings of IEEE Wireless Communications,

Signal Processing and Networking, Chennai, India, 23–25 March 2016
2. Sharma, S., Zou, J.J., Fang, G. (eds.): Proceedings of the IEEE 2nd Contemporary Computing
and Informatics, Noida, India, 14–17 Dec 2016
3. Zhang, Q., Levine, M.: Robust multi-focus image fusion using multi-task sparse representation
and spatial context. IEEE Trans. Img. Process. 25(5), 2045–2058 (2016)
4. Zhang, Y., Bai, X., Wang, T.: Boundary finding based multi-focus image fusion through multi-
scale morphological focus-measure. Inf. Fusion 35, 81–101 (2017)
5. Selesnick, W., Baraniuk, R.G., Kingsbury, N.G.: The dual-tree complex wavelet transform. IEEE
Signal Process. Mag. 22(6), 123–151 (2005)
6. Sasikala, M., Kumaravel, N.: A comparative analysis of feature based image fu-sion methods.
Inf. Technol. J. 6(8), 1224–1230 (2007)
7. Prakash, O., Kumar, A., Khare, A. (eds.): Proceedings of IEEE 2014 International Conference
on Issues and Challenges in Intelligent Computing Techniques, Ghaziabad, India, 7–8 Feb 2014
8. Xiao, G., Jing, Z.L., Li, J.X. (eds.): Proceedings of IEEE International Conference on Intelligent
Transportation Systems, Shanghai, China, 12–15 Oct 2003
9. Wang, H.D.: Theoretical and Experimental Studies on Non-Fourier Heat Conduction Based on
Thermomass Theory, pp. 103–104. Springer Science & Business Media, Berlin (2008)
Chapter 16
Real-Time Monitoring of Fatigue Cracks
in Machine Parts Using Microwave
and Laser Imaging Techniques
P. Giri, S. Kharkovsky, B. Samali and R. Salama
Abstract This paper presents a multifunctional imaging system with a sensing unit
including integrated microwave and laser sensors. It focuses on the development of
software and algorithm for automated control of the movement of the sensing unit
attached to the 3-axis scanning system and optimization of the standoff distance
between the sensing unit and the material under test. The sensing unit consists of a
microwave antenna/sensor and two laser displacement sensors. Microwave sensor is
used for data acquisition and imaging of the surface and hidden flaws in the metal.
Laser displacement sensors are used for obtaining profile information of the structure
and an automatic adjustment of the sensing unit at optimized standoff distance. The
optimization of standoff distance is important mainly in non-plain machine parts as
their non-planarity might mask indications of minute flaws such as cracks. For this
purpose, a combined software program is developed for data acquisition, motion
control and synchronization in the LabVIEW platform. Further, the developed pro-
gram includes a signal processing module for post-processing of signals and image
smoothing. All these modules are synchronized using various functions with a trigger.
A suitable user-friendly graphic user interface is also developed. The applicability
of the system is demonstrated through a non-contact detection of hidden flaws in
metals used in machine parts.
Keywords Microwave imaging · Laser displacement sensors · Fatigue cracks ·

Nondestructive testing · Structural health monitoring
16.1 Introduction
Microwave imaging techniques have been successfully applied for a variety of appli-
cations such as non-destructive testing and evaluation and material characterization
[1, 2]. Near-field microwave non-destructive techniques use open-ended rectangular
P. Giri (B) · S. Kharkovsky · B. Samali · R. Salama

School of Computing, Engineering and Mathematics, Centre for Infrastructure Engineering,
Western Sydney University, Penrith, NSW 2751, Australia

200 P. Giri et al.
waveguide probes/antennas along with relatively simple microwave reflectometers

for inspection and damage detection of conducting, dielectric and composite mate-
rials such as metals, concrete, concrete-filled steel tubes and CFRP-strengthened
concrete structures [3–5]. It involves detection of flaws such as surface and hidden
cracks, debondings and delaminations.
Mechanical structures primarily composed of metals and alloys are used in aircraft
fuselage, railroad tracks and car wheels, turbine blades, generators, engines and
various machineries [6]. Mechanical structures and components are subjected to
varying loads during their lifetime which often leads to fatigue failure [7]. Meal
fatigue and subsequent failure usually originates from the surface as a crack and
propagates inwards. Hence, surface crack detection of metallic structures is of utmost
importance. Fatigue crack monitoring of metals and alloys are useful for the safe
running of these mechanical structures preventing injuries and financial loss [8]. It is
also critical to know the exact location of the crack for repair and reinforcement. Lamb
wave technique is one of the common methods for detecting fatigue cracks in metals.
Piezoelectric transducer arrays were bonded to the metal surface and damage indexes
were used to detect surface cracks in metal [9]. Similarly, a single piezoelectric
transducer was attached on the metal surface and laser doppler vibrometer was used
to sense the lamb wave generated by the attached transducer [10]. In both these
techniques, piezoelectric transducers need to be bonded to the surface of the metal
which might not be applicable in real life mechanical structures. Laser displacement
sensor was used for profiling and through-gap detection in metal by generating one-
dimensional profiles as well as two- and three-dimensional images of the surface of
the metal specimen using the displacement value from the sensor [11, 12]. The above
techniques are useful in detecting cracks in metal but are mainly focused on surface
crack detection.
However, several machinery parts are painted or lined with thin-layered elas-
tomeric materials such as rubber. The rubber linings are mainly used to reduce energy
dissipation and to improve the efficiency of the machine, to eliminate the lubrication
system, to protect the mechanical components from dust and to reduce vibration and
noise generation [13]. In these cases, the optical and lamb wave based crack detection
system is inapplicable as they are not able to detect hidden cracks under the layer.
The advantage of microwave techniques over these techniques is mainly due to the
ability of microwave signals to penetrate inside dielectric materials which is useful
for detecting and evaluating sub-surface cracks or cracks hidden under dielectric
coatings such as paints or rubbers [14].
Microwave imaging is performed by scanning a single antenna over a specimen
at some distance which is referred to as standoff distance. When a specimen with
a plane surface is being inspected, a simple two-dimensional raster scanning can
be performed and the reflected signal data can be used to generate the images [15].
However, the machine parts can have different geometric shapes, edges and joints.
Due to this reason, the standoff distance can change which masks the indication
of flaws such as cracks. To remove this effect, a few techniques have been used.
A 3-axis scanning mechanism was used for the optimization of standoff distance
and following the contour of the surface of the specimen using manual setting [16].
16 Real-Time Monitoring of Fatigue Cracks in Machine … 201
Then a laser displacement sensor (LDS) was incorporated in the system to provide
automated control of the movement of the antenna attached to the scanner [11]. This
paper mainly focuses on the algorithm that is developed to synchronize the integrated
sensing unit including a microwave antenna and the LDS with the scanner. The LDS
is used to determine the position of the antenna with respect to the specimen. The
workability of the developed algorithm was tested by imaging a layered dielectric
specimen with embedded rubber disks. The algorithm was then used to detect a crack
in the metal specimen hidden under the rubber layer.
16.2 Integrated Imaging System
The schematic of the integrated imaging system is shown in Fig. 16.1. The inte-
grated imaging system consists of hardware and software section. The hardware
section includes sensors, scanner, data acquisition unit and control unit. The soft-
ware section includes LabVIEW and MATLAB programs and algorithm to initialize
and synchronize different sensors with the scanning unit.
16.2.1 Hardware Setup
The integrated imaging system consists of the scanning unit, control unit, data acqui-
sition unit and integrated sensing unit as shown in Fig. 16.1. The scanning unit
includes the scanner and 3-axis motion controller card. This unit also includes linear
encoders along each axis to provide position feedback. The movement of the scanner
is controlled by the control unit. The control unit includes a computer, as well as
a microprocessor for control and data processing. This unit sends the signal to the
integrated sensing unit to acquire data after each movement of the scanner. The inte-
grated sensing unit includes the microwave antenna and two LDSs. The antenna and
LDS radiate microwave and optical signals, respectively, into the specimen, pick up
Fig. 16.1 Block diagram of

integrated imaging system
202 P. Giri et al.
the reflected signals, and send them to the data acquisition unit. The data acquisition
unit includes performance network analyzer (PNA) which generates and acquires
microwave data, and a microprocessor to acquire data from LDSs.
In this study, an Agilent N5225A PNA is used to generate and measure continuous
wave microwave signals with the output power of not greater than 10 mW. Similarly,
an LDS from Micro-epsilon is used which includes a laser head and a detector. The
measurement range of LDS is from 60 to 260 mm and the emitted laser operates at
a wavelength of 670 nm with a maximum laser output power of 1mW.
16.2.2 Development of Algorithm
An object-oriented LabVIEW program is developed to control the PNA via a GPIB

interface, to acquire data from the LDS and synchronize data acquisition and scanning
unit. The program drives the stepper motors using a 3-axis motion controller card
and individual current amplifiers. Similarly, the program reads displacement values
from two LDSs to obtain a profile of the specimen under test. According to this data,
the positioning platform provides the motion of the integrated sensing unit in such
way that it follows a surface profile of the specimen (i.e., standoff distance is kept
constant throughout the scan). The obtained microwave data is used to generate 2-D
(two-dimensional) magnitude and phase images.
Figure 16.2 shows the block diagram of an algorithm for the integrated imaging
system which includes three modules. An initialization module gives commands to
the scanner, microwave antenna and LDS for initialization. This initiates motion of
the scanner and the measurement of sensors using dimensions of scan area and step,
Fig. 16.2 Block diagram of algorithm for integrated imaging system

the sampling rate for the LDS and sampling rate and frequency for the microwave
antenna. A profiling and synchronization module acquires profile information from
the displacement reading of two LDSs and sends this information as a coordinate to
the scanner. The scanner positions itself based on the coordinate of the profile. Data
acquisition and imaging module then acquires microwave data at each scan point
and arranges the data as a matrix to generate 2-D magnitude and phase images. The
obtained images are then further smoothed using MATLAB algorithm.
A few specimens were used in this investigation to illustrate the applicability of

the system for the detection and evaluation of defects. The first specimen was a
six-layered foam structure with four embedded rubber disks. The rubber disks were
embedded at different layers of the foam specimen. This specimen was used to test
the applicability of the developed system in layered materials. The dimension of the
foam structure was 225 mm × 225 mm × 50 mm.
Figure 16.3a, b show two views of the schematic of the tilted foam specimen being
scanned by the integrated sensing unit along two scan planes, scan plane 1 and scan
plane 2. The scan plane 1 referred to the movement of scanner along X- and Y-axis
Fig. 16.3 a Schematic of the integrated sensor unit testing the layered foam specimen with embed-
ded rubber disks; b top view of the specimen (left) and cross-sectional side views of the specimen
showing the location of the rubber disks (right) inside the specimen
204 P. Giri et al.
Fig. 16.4 Microwave images of scanned area 1 of the layered foam specimen at: a scan plane 1
and b scan plane 2
without the Z-axis movement. This led to a variable standoff distance d between the
sensing unit and the specimen. The scan plane 2 referred to the movement of scanner
along all three axes using the feedback from two LDSs which led to the constant
standoff distance, d 0 , between the sensing unit and the specimen. Two different scan
areas were considered for the scan: first scan area was 90 mm × 90 mm which was
chosen to visualize single embedded rubber and was referred to as scan area 1 while
the second scan area was 210 mm × 210 mm which was chosen to visualize all four
embedded rubber disks and was referred to as scan area 2.
Figure 16.4 shows the 10.3-GHz raw images of the scan area 1 of the layered foam
specimen tilted at 8° using the magnitude of reflection coefficient. The step size was
set to be 3-mm along both X- and Y-axis at d 0 = 15 mm. Although the raw image
of foam at scan plane 1 showed indication of embedded rubber disk (cf. Fig. 16.4a),
the intensity change can be clearly seen which was highlighted by different color
gradients. This is mainly due to the standoff distance change over the scanned area
due to the tilt of the specimen. When scanning at scan plane 2, the gradual intensity
change significantly reduced and the indication of rubber became more prominent
(cf. Fig. 16.4b). This is highlighted by the smooth color of the image and a clear
demarcation between the area with and without the embedded rubber disk.
Figure 16.5 shows the 10.3-GHz raw magnitude images of scan area 2 of the
layered foam specimen with tilted at 8°. Similar to the previous result, the raw image
at scan plane 1 shows indication of rubber disks but different colored gradients
were present which blurred the image mainly around the boundary of the rubber
(cf. Fig. 16.5a). The gradient effect was reduced and the indication of rubber discs
became more prominent as the contrast between the area with rubber and without
rubber increased at scan area 2 (cf. Fig. 16.5b). The color of four different rubbers
was different in Fig. 16.5b because of the different depths of rubber discs in the
specimen. For example, the rubber discs with prominent indication were positioned
between the first and second layers of the specimen while the rubber discs with less
Fig. 16.5 Microwave images of scanned area 2 of the layered foam specimen at: a scan plane 1
and b scan plane 2
Fig. 16.6 Schematic of the integrated sensor unit testing the tilted metal slab with 1-mm width
through crack covered by the rubber sheet
prominent indication were positioned between the fifth and sixth layers of the foam
specimen.
The second specimen was a metal (steel) slab with the dimension of 300 mm ×
300 mm × 20 mm which possessed a 1-mm width through-crack. The slab was posi-
tioned on a tilted base. In addition, to demonstrate the effectiveness of the integrated
imaging system for crack detection under rubber lining, 3-mm thick rubber layer was
placed on the top of the metal slab. Figure 16.6 shows the side view of the schematic
of the tilted metal slab being scanned by an integrated sensing unit along scan plane
1 and scan plane 2. At scan plane 1, the standoff distance d between the sensing unit
and the specimen varied. At scan plane 2, the standoff distance was kept constant
(d 0 = 6 mm). The step size of the scanner was set to be 3-mm along both X- and
Y-axis. The scan area was chosen to be 45 mm × 90 mm.
The images of the metal slab covered with a rubber layer and tilted at 10° was
generated using magnitude of reflection coefficient at 10.3 GHz. The raw magnitude
images at scan plane 1 and 2 are shown in Fig. 16.7. The image at scan plane 1 shows
a gradual intensity change from top to bottom due to the change of standoff distance
over the scanned area (cf. Fig. 16.7a). The indication of the crack is completely
206 P. Giri et al.
Fig. 16.7 Microwave images of the scan area of the metal slab with 1-mm crack covered by a
3-mm thick rubber sheet at: a scan plane 1 and b scan plane 2
masked by this gradual intensity change. At scan plane 2, the effect of gradual inten-
sity change is removed and there is a prominent indication of crack (cf. Fig. 16.7b).
The results demonstrated the applicability of the integrated imaging system for the
detection of cracks in machine parts with different geometrical shapes.
16.4 Conclusion
The developed algorithm was applied for the integrated imaging system which was
tested using two specimens. The first specimen was a layered foam one with four
embedded rubber discs. The second specimen was a metal slab with a through crack
covered by a rubber layer.
The results showed that the integrated imaging system with the developed algo-
rithm was able to detect the rubbers discs located at different depth in the specimen
as well as the crack in tilted metal specimen covered by the rubber layer. The system
can generate microwave images of the specimens by providing an optimal standoff
distance automatically at the hardware level. The developed system can be an effec-
tive non-contact solution for real-time fatigue crack detection in a variety of machine
parts covered with dielectric coatings such as paints or layered machine components
lined with materials like rubber.
References
1. Kharkovsky, S., Zoughi, R.: Microwave and millimeter wave nondestructive testing and eval-
uation—overview and recent advances. IEEE Instrum. Meas. Mag. 10(2), 26–38 (2007)
2. Ramzi, M.R., Abou-Khousa, M., Prayudi, I.: Near-field microwave imaging using open-ended
circular waveguide probes. IEEE Sens. J. 17(8), 2359–2366 (2017)
3. Islam, M.A., Kharkovsky, S.: Detection and monitoring of gap in concrete-based composite
structures using microwave dual waveguide sensor. IEEE Sens. J. 17(4), 986–993 (2017)
4. Kharkovsky, S., Ryley, A.C., Stephen, V., Zoughi, R.: Dual-polarized near-field microwave
reflectometer for noninvasive inspection of carbon fiber reinforced polymer-strengthened struc-
tures. IEEE Trans. Instrum. Meas. 57(1), 168–175 (2008)
5. Zoughi, R., Kharkovsky, S.: Microwave and millimetre wave sensors for crack detection.
Fatigue Fract. Eng. Mater. Struct. 31(8), 695–713 (2008)
6. Yeh, C.-Y., Zoughi, R.: A novel microwave method for detection of long surface cracks in
metals. IEEE Trans. Instrum. Meas. 43(5), 719–725 (1994)
7. Forrest, P.G.: Fatigue of Metals. Elsevier, Amsterdam (2013)
8. Stephens, R.I., Fatemi, A., Stephens, R.R., Fuchs, H.O.: Metal Fatigue in Engineering. Wiley,
New York (2000)
9. Ihn, J.-B., Chang, F.-K.: Detection and monitoring of hidden fatigue crack growth using a
built-in piezoelectric sensor/actuator network: I. Diagnostics. Smart Mater. Struct. 13(3), 609
(2004)
10. Leong, W., Staszewski, W., Lee, B., Scarpa, F.: Structural health monitoring using scanning
laser vibrometry: III. Lamb waves for fatigue crack detection. Smart Mater. Struct. 14(6), 1387
(2005)
11. Kharkovsky, S., Giri, P., Samali, B.: Non-contact inspection of construction materials using
3-axis multifunctional imaging system with microwave and laser sensing techniques. IEEE
Instrum. Meas. Mag. 19(2), 6–12 (2016)
12. Giri, P., Kharkovsky, S., Samali, B.: Inspection of metal and concrete specimens using imaging
system with laser displacement sensor. Electronics 6(2), 36 (2017)
13. Rivin, E.: Properties and prospective applications of ultra thin layered rubber-metal laminates
for limited travel bearings. Tribol. Int. 16(1), 17–25 (1983)
14. Kharkovsky, S., Ghasr, M.T., Zoughi, R.: Near-field millimeter-wave imaging of exposed and
covered fatigue cracks. IEEE Trans. Instrum. Meas. 58(7), 2367–2370 (2009)
15. Case, J.T., Ghasr, M.T., Zoughi, R.: Optimum two-dimensional uniform spatial sampling for
microwave SAR-based NDE imaging systems. IEEE Trans. Instrum. Meas. 60(12), 3806–3815
(2011)
16. Kharkovsky, S., Giri, P.: Detection of crack in cement-based specimens using microwave imag-
ing with the 3-axis scanning system. In: 2016 IEEE International Instrumentation and Mea-
surement Technology Conference Proceedings. IEEE (2016)
Chapter 17
Modeling and Characterization
of the Double Helical Compliant Joint
Rasheed Kittinanthapanya, Yusuke Sugahara, Daisuke Matsuura

and Yukio Takeda
Abstract In this paper, the modeling of a double helical compliant joint and its
characterization are introduced. The Double Helical Compliant Joint (DHCJ) was
designed to increase the workspace in a compliant mechanism. The joint consists
of leaf springs that are symmetrically arranged between two helical frames. The
modified-type DHCJ was created to show several characteristics when some param-
eter is changed. Each leaf springs movement can be considered as a large deflection
of elastic beam problem, from which the mathematical model of the joint can be con-
structed based on Chained Beam-Constraint Model method. To support mechanical
design of the joint for a specific application, in which variability of characteristic
according to each parameter is utilized. One DOF characterization apparatus was
designed to characterize each joint specimen. Then load-displacement relationship
can be experimentally investigated and compared with the solution of a mathematical
model.
Keywords Compliant joint · Characterization · Large deflection · Robotics ·

Rotational stiffness
17.1 Introduction
Compliant joints have been used widely in the field of high-precision manufacturing
such as micro assembly robots, surgery manipulators and MEMS manufacturing
technology. The joint gives high-precision performance in addition to playing an
important role in a vacuum environment, which is sensitive to friction. Moreover,
compliant mechanisms have no backlash, need no lubrication, and are easier to wash.
Integration of compliant joints in other manipulators can expand their performance
and ability in the robotics field. However, some types of compliant joint can be
operated in just a small range of motion and most of the joint gives display a centre
R. Kittinanthapanya (B) · Y. Sugahara · D. Matsuura · Y. Takeda

Department of Mechanical Engineering, Tokyo Institute of Technology, 2-12-1,
Ookayama, Meguro-ku, 152-8552 Tokyo, Japan

210 R. Kittinanthapanya et al.
shift axis effect. Kozuka et al. [1] have presented a compliant parallel mechanism
that was designed by using a circular-shape compliant joint, which gives a range of
motion between −20.0 and +20.16°. Arnaud et al. [2] designed a large amplitude
compliant revolute joint, which uses multiple materials in the same part, and the
range of motion can be increased to around a 70–80° maximum when changing the
design parameters of the joint.
In this paper, we present the modelling and characterization of the Double Heli-
cal Compliant Joint (DHCJ) for various design parameters. This joint was origi-
nally designed by Yonemoto et al. [3] for a large-workspace compliant mechanism.
However, the design parameters have not yet been considered for matching to the
application requirements. Therefore, the purpose of this paper is to characterize the
DHCJ when some design parameters were changed, and all of the modified types
were tested on characteristics apparatus. The mathematical model of the DHCJ was
derived by using Chained Beam-Constraint Model (CBCM) to check the agree-
ment of the experimental and theoretical results. Moreover, this model can be used
to predict and simulate the effect from changing the design parameters. All of the
characteristics data were observed for use in a guideline for designing helical-shape
compliant joints for use in different application.
17.2 Design of the DHCJ
The Double Helical Compliant Joint (DHCJ) consists of two main parts; there are
two helical frames and sixteen leaf springs in the original design [3]. The helical
frames use a solid material for rigidity, and the leaf spring use a flexible material,
which can provide a large range of motion as shown in Fig. 17.1. The first modified
type is called pre-bending type, which was designed by Moriya et al. [4]. However,
there are many influences from other parameters such as the diameter and length of
the joint, thickness of the leaf spring, etc., which can affect its characteristics. Further
detail of the modified type will be discussed in this chapter.
17.2.1 The Original and Pre-bending DHCJ
The original designed of the DHCJ has a straight leaf spring. The 1st and 16th leaf
springs are placed in the same direction and orientation. The same trend applies to
the 2nd and 15th leaf springs case, 3rd and 14th leaf springs case and finally again
at the 8th and 9th leaf springs case as shown in Fig. 17.2. This kind of arrangement
provides rotational symmetry around the axis of rotation (z-axis). The 1st and 5th
leaf springs are placed in a reverse arrangement for both direction and orientation,
which is called a “pairing spring”. These pairing springs help provide symmetry of
rotational stiffness for the joint. Additionally, this mechanical constraint makes sure
that the end tip of each leaf spring will deflect in a circular path. In other words,
17 Modeling and Characterization of the Double Helical … 211
Fig. 17.1 Large range of motion of the DCHJ 0◦ initial state (upper), 45◦ rotation state (lower)
Fig. 17.2 Structure of the DHCJ and arrangement of leaf spring with one frame
Fig. 17.3 Original type of DHCJ (left) and pre-bending type of DHCJ (right)
the joint will rotate without center-shift around the z-axis with sufficient constraint
stiffness similar to a ball bearing motion.
As previously presented, the first modified type of the DHCJ is called the pre-
bending type. At the initial state of this type, the leaf springs have already been bent
by 20° for both fixed ends. Figure 17.3 shows the structure of the whole joint and
the front side of the joint as it differs between the original and pre-bending types.
This pre-bending type was designed for reducing rotational stiffness and using less
energy than original type [4]. However, the working range of the joint becomes
smaller due to the helical structure. A final drawback of the pre-bending type is that
the constraint stiffness and avoids buckling of the leaf springs by the external force
acting perpendicular to the z-axis.
17.2.2 The Modified Type of DHCJ
As pointed out by Bruyas et al. [2], the dimensions of the joint are changed, this
affect its characteristics of the joint. The design parameters of our DHCJ are shown
in Fig. 17.4; that is inner diameter d, length of the joint p, or even leaf spring thick-
ness h and leaf spring width b. As modified design, we selected four sets of design
parameters shown in Table 17.1.
For example, d13-h01 refers to the modified type of the DHCJ with a 13 mm
inner diameter and a 0.1 mm leaf spring thickness. In the case of ph8 and ph16, those
mean that the p parameter is half of the original size, and the number 8 and 16 specify
number of leaf spring. Figure 17.5a–e show the CAD drawing of each modified type
d18-h01, d13-h01, d9-h01, ph8 and ph16 respectively.
Fig. 17.4 Relevant parameters for the modified type of DCHJ
Table 17.1 Changing parameters from original type to modified type

Name d18 original d13 modified d9 modified ph8 ph16
Design parameters modified modified
Inner diameter, d (mm) 18 13 9 18 18

Joint length, p (mm) 100 100 100 50 50
Leaf spring width, b 5 5 5 5 2.5
(mm)
Leaf spring thickness, h 0.1 and 0.2 0.1 and 0.2 0.1 and 0.2 0.1 0.1
(mm)
No. of leaf spring (pcs.) 16 16 16 8 16
Leaf spring phase shift, 45 45 45 45 45
φ (deg.)
Fig. 17.5 The modified-type of the DHCJ

17.3 Modeling of the DHCJ
There are many methods for modeling a large deflection of a beam such as the
Bisshopp and Drucker method [5], which uses a complex elliptical integral form
and their solution is as a point load. Fertis [6] proposed a more simple integral
approach to determine the deflection of the beam but their solution is suitable for
shear load without the axial load problem. Recently Ma and Chen [7] proposed a
chained beam-constraint model (CBCM), which provides a simpler calculation and
an accurate solution for the large deflection problem. For modeling the DHCJ, we
applied CBCM to approximate the DHCJ mathematical model.
17.3.1 Chained Beam-Constraint Model (CBCM)
The CBCM is a numerical method that was developed from the Beam-Constraint
Model (BCM). In the CBCM method, the beam is divided into N elements and each
element is modeled by BCM as shown in Fig. 17.6. Then, the closed-form equations
can be expressed as Eqs. (17.1) and (17.2). Denoting that i = 0,1,2,…,N.
Fig. 17.6 Discretized

cantilever-beam model by
CBCM

fi 12 −6 pi 36 −3 yi
yi pi2 −9 4.5 yi
= + +
mi −6 4 αi
30 −3 4 αi 6300 4.5 −11 αi
(17.1)

t 2 pi 1 36 −3 yi pi −9 4.5 yi
xi = − yi αi − (17.2)
12 60 −3 4 αi 6300 4.5 −11 αi
where pi , f i , m i , xi , yi and θi are normalized load and deflection parameters

respectively of ith element. Those can be expressed as
Pi L 2 Fi L 2 Mi L N δxi N δ yi
pi = 2
, f i = 2
, mi = , xi = , yi = , αi = αi .
N EI N EI NEI L L
Similarly, for the whole beam, p0 , f 0 , m 0 , x0 , y0 and θ0 are normalized load
and deflection parameters respectively on the free end of the beam. Those can be
expressed as
P0 L 2 F0 L 2 M0 L N X0 N Y0
p0 = 2
, f 0 = 2
, m0 = , x0 = , y0 = , θ0 = θ0 .
N EI N EI NEI L L
The fixed end is node 0 and the free end is node N. There is a shear force F0 , axial
force P0 and tip moment M0 that act on the free end. The static equilibrium between
the beam tip and the 1st element can be derived by Eq. (17.3) and a static equilibrium
between the 1st element and the ith element can be expressed in Eq. (17.4),
p1 = p0 , f 1 = f 0 , m N = m 0 , (17.3)
⎡ ⎤⎡ ⎤ ⎡ ⎤
cos θi − sin θi 0 fi f1
⎣ sin θi cos θi 0 ⎦⎣ pi ⎦ = ⎣ p1 ⎦, (17.4)
(1 + xi ) − yi 1 mi m i−1
where θi is the rotation angle of the ith element’s coordinate frame with respect to
the global coordinate frame. θi can be expressed by the unknowns αk ,

i−1
θ1 = 0, θi = αk (i = 2, 3, . . . , N ). (17.5)
k=1
The geometric constraint equation of the entire beam can be expressed as

N
[(1 + xi ) cos θi − yi sin θi ] = x0 ,
i=1
N
i=1 [(1 +
xi ) sin θi + yi cos θi ] = y0 , (17.6)
N
i=1 αi = θ0 .
From Eqs. (17.1) to (17.6), there are in total 6 N + 3 equations. Among the six
parameters p0 , f 0 , m 0 , x0 , y0 and θ0 which are load and displacement parameters,
given any three parameters the other three can be obtained by numerical solving.
17.3.2 Simulation
Form Fig. 17.2, each leaf spring is modeled by a fixed support cantilever beam. Then,
we can make a kinematic diagram of one leaf spring as shown in Fig. 17.7. The leaf
spring length is L, deflection in x-axis and y-axis are X 0 and Y0 , and deflection angle
of the leaf spring tip is θ0 . As mentioned above, this joint is considered to rotate
without center-shift, so the right side tip of the leaf spring will move in a circular
path with L/2 radius and subjected to P0 , F0 and M0 loads. From geometry, the
kinematic equation of one leaf spring can be written as
L L
X0 = + cos θ0 ,
2 2
L
Y0 = sin θ0 . (17.7)
2
For determining how much torque we need for driving the leaf spring into desired
angle, kinetostatics equation of one leaf spring can be expressed as Eq. (17.8).
Accordingly, the DHCJ has 16 leaf springs, and each leaf spring will rotate by a
same angle. So the kinetostatics of whole joint (16 leaf springs) can be expressed
Fig. 17.7 Kinematic

diagram of one leaf spring in
the DHCJ
Fig. 17.8 Calculation flow
Fig. 17.9 Deflection of one leaf spring
by Eq. (17.9). The calculation flow is shown in Fig. 17.8. N is chosen as 23, and
the range of motion is from +50° to −50°. Then deflection of the leaf spring can be
plotted as in Fig. 17.9 and the required torque for driving the DHCJ into each angle
is shown in Fig. 17.10.
L L
τin = M0 + F0 cos θ0 + P0 sin θ0 (17.8)
2 2
τall = τin × 16 (17.9)
Fig. 17.10 Relationship between the input torque and rotation angle of DHCJ d18-h01
17.4 Experiments
Experiments were conducted for revealing the load-displacement relationship of the

DHCJ for variety of design parameters. A simple mechanical apparatus was designed
and manufactured to drive the joint around the rotational axis (z-axis). The joint itself
behaves like a spring mechanism; therefore to avoid vibration the joint was moved
slowly.
17.4.1 Experimental Set Up
The joint is placed vertically and takes only a z-axis moment load from the motor
to avoid undesired loads on the other axes. Angle data was collected by encoder in
servo motor, and moment data was collected by a force/torque sensor. The mechani-
cal apparatus is shown in Fig. 17.11, and all equipments are shown in Table 17.2. The
helical frame of the modified type was made by using a 3D printer, and is assembled
with leaf springs as shown in Fig. 17.12. There are two materials in the leaf spring:
Gum Metal (a very flexible alloy) [8] and SK5 (spring steel). Gum Metal was manu-
factured to be an alloy that is hard to be broken or destroyed. It provides high tensile
yield strength over 1200 MPa but with an extremely low young’s modulus.
Fig. 17.11 Characteristic

testing apparatus
Table 17.2 Equipment used in the characteristics testing apparatus

Equipment Specification
Maxon DC motor Max torque: 2.0 Nm
Encoder HEDL 5540 Counts per turn: 500, multiplication by four
BL autotec Load capacity: Fx, Fy = 78.4 N Resolution: Fx, Fy = 78.5 ×
10−3 N
Force/Torque sensor Fz = 156.8 N Fz = 235 × 10−3 N
MINI 8/40A Tx, Ty, Tz = 3.92 Nm Tx, Ty, Tz = 1.96 × 10−3 Nm
Misumi XY stage 60 mm × 60 mm, manual feeding screw
Controller OS: QNX
Control cycle: 1 ms
T104-C160 for counter board
T-ADA16S for AD-DA board
Fig. 17.12 DHCJ

prototypes
17.4.2 Experimental Procedure
All specimens were driven by a DC motor with a smooth slow trajectory as shown
in Fig. 17.13 to avoid dynamics effect; the maximum angle and time was set to
perform a 5th power polynomial curve. The joint move from the initial angle (0°) to
the maximum angle; we call this movement “phase A” or forward driving. Then, it
moves back to initial angle again; we call it “phase B” or backward driving. When
the joint goes forward to the opposite side and return backward to the initial angle
again, it is called “phase C” and “phase D” respectively. All phases have the same
setting time. All forces data and angle data are collected for every 1 ms, and noise is
filtered out by using a FIR filter at the end of the experiment.
To avoid dynamics effect, for example, d18-h01 was tested with a different of
setting time. From Fig. 17.14, the result is shown that 2 s gave some oscillating and
a small gap between phases A and B and also between phases C and D. More than
5 s provides a more stable result and less fluctuation.
Fig. 17.13 Motor trajectory
Fig. 17.14 Comparison of setting time

17.4.3 Results and Discussion
First, d18-h01, d13-h01 and d9-h01 were tested to move forward and backward and
compared with the theoretical result as shown in Fig. 17.15. There are small gaps
between forward and backward, but the theoretical result still lay on the middle of
both graphs. These prove that the experimental and theoretical results are reliable.
Then d18-h01, d18-h02 (which uses Gum Metal) and SK5 were tested and the result
is shown in Fig. 17.16. When comparing these three types, d18-h01 and SK5 seem
to fit the linear behaviour. But for d18-h02, the characteristics curve gave nonlinear
behaviour and is also asymmetric in phase A and phase B. This irregular curve is not
shown in a normal spring material, and is considered as a special characteristic to be
caused by Gum Metal.
Fig. 17.15 Comparison of h01 type
Fig. 17.16 Comparison of d18-h01, h02 and SK5

Fig. 17.17 Repeatability of d18-h02
Moreover, when thickness of Gum Metal was increased to 0.2 mm, rotational
stiffness increases significantly (~5times higher) as shown in Fig. 17.16. To observe
the repeatability of this loop, d18-h02 was tested ten times and the result is shown
in Fig. 17.17. These phenomena occur continuously inside the boundary value as
shown by the black dashed line and the average values on the red dashed line. The
polynomial function that is fit to this loop is expressed as a blue dashed line.
The other types such as ph16, ph8 and pre-bending were tested and compared with
the h01 type as shown in Fig. 17.18. The first modified type or pre-bending type shows
a drawback where it has higher stiffness than the original type. The ph8 type was
asymmetric stiffness between the plus and minus side, and a center-shift effect even
when it is the smallest stiffness. The reason is that there are no pairing springs, which
help for balancing stiffness of both sides. The characteristics of ph16 and the original
type are quite similar, so changing the overall length does not significantly affect the
rotational stiffness. However, there is an interesting point; the pre-bending type, ph8
type and ph16 type have a small gap between phases A and B when comparing to
the original type.
To summarize, when the thickness of leaf spring was increased from 0.1 to 0.2 mm
(h01 to h02), rotational stiffness is significantly increased. When the inner diameter of
the joint is decreased from d18 to d13 and d9, rotational stiffness is slightly increased.
The leaf spring arrangement must be symmetric by considering the number of leaf
springs and their phase shift. Reducing the joint length does not significantly affect
rotational stiffness.
Fig. 17.18 Comparison of other modified type
17.5 Conclusions
This paper proposes the modeling of a Double Helical Compliant Joint (DHCJ)
by using the CBCM method. The characteristics of the joint were investigated by
experiments and compared to the theoretical results. From theoretical calculation, the
required input torque and leaf spring deflection shape can be obtained and predicted.
From the experimental results, characteristics of each modified joint were shown with
regards to how changing design parameters would affect rigidity, rotational stiffness,
range of motion, balance of stiffness and linear behavior. All of these simulation
results and experimental results are useful data in the design process of DHCJ for a
variety of applications.
On the other hand, the Gum Metal material with high thickness has an irregular
characteristics curve, and the model proposed in this paper could not explain this
phenomena. So we plan to model this nonlinear characteristic by future experimen-
tal investigation. Especially, utilizing this phenomenon to achieve more desirable
characteristics for compliant joint is our future work.
References
1. Kozuka, H., Arata, J., Okuda, K., Onaga, A., Ohno, M., Sano, A., Fujimoto, H.: A bio-inspired
compliant parallel mechanism for high-precision robots. In: 2012 IEEE International Conference
on Robotics and Automation. 14–18 May 2012, pp. 3122–3127
2. Bruyas, A., Geiskopf, F., Renaud, P.: Design and modeling of a large amplitude compliant
revolute joint: the helical shape compliant joint. J. Mech. Des. (2015 by ASME) 137(085003),
1–8 (2015)
3. Yonemoto, K., Takeda, Y., Tong, Z., Higuchi, M.: A new flexure revolute joint with leaf springs
and its application to large workspace parallel robot. J. Adv. Mech. Des. Syst. Manuf. (2012 by
JSME) 6(1), 76–87 (2012)

4. Moriya, R., Matsuura, D., Takeda, Y.: Development of spatial 1-DOF torque transmission mech-
anism for characteristics evaluation of flexure revolute joints. In: Proceedings of the 6th Inter-
national Conference on Positioning Technology, 18 November 2014
5. Bisshopp, K.E., Drucker, D.C.: Large deflection of cantilever beams. Q. Appl. Math. 3(3),
pp. 272–275 (1945)
6. Fertis, D.G.: Basic theories and principles of nonlinear beam deformation. In: Nonlinear Struc-
tural Engineering With Unique Theories and Methods to Solve Effectively Complex Nonlinear
Problems, pp. 1–22. Springer, Berlin (2006)
7. Ma, F., Chen, G.: Modeling large planar deflections of flexible beams in compliant mechanisms
using chained beam-constraint-model. J. Mech. Robot. (2016 by ASME) 8(021018), 1–11 (2016)
8. Sato, T., Nishino, K., Furuta, T.: Introduction of GUMMETAL Soft Modulus & Flexible Titanium
Alloy. Toyota Central R&D Labs, Inc
Chapter 18
Optimization of the Geometry
of a Cable-Driven Storage and Retrieval
System
C. Reichert and T. Bruckmann
Abstract This paper addresses the improvement of the geometrical design of a

rack feeder that is based on cable-driven robot technology. Here, both the forces
needed to approach a storage position in the high rack as well as the stiffness need to
be optimized. An effective parametrization of the platform is introduced. A certain
sequence of optimization steps using a genetic algorithm is applied, combining both
optimization goals. This leads to a new cable robot geometry that provides superior
stiffness at slightly increased cable forces.
Keywords Cable robot · Wire robot · Rack feeder · Cable-driven storage and
retrieval system · Stiffness · Optimization · Genetic algorithm
18.1 Introduction
Storage and retrieval systems (also known as rack feeders) are an essential component
of any modern automated warehousing using high racks. Driven by the demand of
realizing a maximum turnover of goods, in the last decades multiple improvements
have been carried out. Mainly, these improvements were in the field of software
and IT, while the mechanical principle remained unchanged: A conventional rack
feeder consists of a rail on the ground, a skid and a lifter with a push-and-pull device,
which—in the sense of robotics—is the end effector. This allows to reach all storage
positions of a high rack.
However, this mechanical approach has a drawback: From a kinematical point
of view, this is a serial topology, i.e. following the structure from the ground to the
end effector, each component has to support the mass of the following parts. This
leads to large moving masses. Typically, even for small payloads of 20–50 kg, the
moving mass may be around one ton! As a result, the energy consumption increases
C. Reichert · T. Bruckmann (B)

Chair of Mechatronics, University of Duisburg-Essen, 47057 Duisburg, Germany
C. Reichert
226 C. Reichert and T. Bruckmann
Fig. 18.1 CAD drawing of current system including coordinate system. Based on: [15]
and—important as well—the performance of the system is limited, even with pow-

erful actuators.
Currently, a large research project that is driven by mechatronic manufacturers,
logistics end users and the University of Duisburg-Essen, is realizing a demonstrator
that replaces the conventional mechanical system by a cable robot (see Fig. 18.1).
Here, the only moving mass is the end effector, realized as a load handling device
on a moving platform. Approaches like this have been presented e.g. by Hassan [1].
Based on a feasibility study that was carried out in the previous project “Effizienz-
Cluster LogistikRuhr” [2], now the project “LEAN—Entwicklung von Leichtbau-
Regalbediengeräten auf Basis von Seilroboter-Technik aus NRW” aims at realizing
and testing a cable-driven rack feeder within the facilities of an international logistics
company. Within the project, the high rack uses a width of ±6 m on either side of
the origin of the inertial coordinate system (see Fig. 18.1) and a height of 9 m.
From the previous studies, potential for improvements was identified and analysed
in the first phases of the current project:
• Stiffness: The stiffness of the current design is comparatively low in the normal
direction of the high rack (y direction, see Fig. 18.1) which needs to be improved.
At the same time, low cable forces are desired in the workspace. Here, an improved
design is demanded.
• Energy consumption: As the cable-driven rack feeder uses flexible cables that
can only pull, but never push, portions of the torques provided by the motors are
needed to always maintain tension in the cables. Here, the integration of mechanical
energy storage devices providing tension (e.g. springs or counterweights) must be
investigated together with optimal trajectories.
18 Optimization of the Geometry of a Cable-Driven … 227
In this paper, the first aspect is addressed in detail while the second challenge will
be practically investigated in the current project in the next years. First analyses have
been published in the past [3–7], but intensive experimental studies are still missing
and part of future work.
In Sect. 18.2, the applied model for stiffness is introduced. Section 18.3 explains
the optimization process while in Sect. 18.4, the results are discussed. Section 18.5
provides a short summary and future steps.
18.2 System Modeling
Assume a pose x of the platform with m cables and the corresponding cable lengths
l. As the cable-driven robot has a parallel topology, all cables apply a wrench onto
the platform. Per cable, the applied tension can be decomposed in a unit vector in
the direction of the cable and a scalar force f i representing the tension in the ith
cable. Taking all cables plus external loads w into account, it follows the following
equilibrium
AT f + w = 0, (18.1)
where AT is the transpose of the Jacobian of the system, called structure matrix.
Its columns a T concatenate the unit vectors in the direction of the cables for force
equilibrium and the unit vectors times the position vector from the platform center
to the cable connection points of the platform for torque equilibrium. Details can be
found in [8, 9].
As a cable-driven system, a cable robot is subject to elasticity effects. This means
that under external wrenches δw (including forces and torques) onto the platform,
the platform will perform a displacement δx. Assuming a linear relation represented
by a stiffness matrix K (x), it follows
δw = K (x)δx. (18.2)
Now Eq. (18.1) can be rearranged
w = − AT f (18.3)
and the total derivative delivers (note that not only AT , but also f is pose dependent,
as a displacement causes elastic forces)

∂ AT T∂ f
δw = − f δx + A δx . (18.4)
∂x ∂x
Now, an infinitesimal platform displacement (keeping the set pose of the platform
constant) causes elastic changes in the cable lengths denoted as δl. As AT is the
transpose of the Jacobian of the system, these cable length changes can be written as
δl = Aδx (18.5)
As the set pose remains unchanged, displacement is associated with elasticity

forces δ f as follows
δ f = K l δl (18.6)

where K l = diag kl,1 , · · · , kl,m represents the stiffness of the actuated cable, influ-
enced by both the elasticity of the cable itself and the compliance of the position
controller of each winch. Combining Eqs. (18.5) and (18.6), it follows
δ f = K l Aδx. (18.7)
Now obviously,
∂f
δf = δx (18.8)
∂x
and therefore Eq. (18.7) becomes
∂f
= K l A. (18.9)
∂x
Equation (18.9) into Eq. (18.4) gives
m
∂ AT ∂ aT
δw = − f δx + A K l Aδx = −
T i
f + A K l A δ. (18.10)
T
∂x i=1
∂x
i
Now the elements of Eq. (18.10) can be compared with Eq. (18.2) which gives
⎛ ⎞
⎜ ∂ AT ⎟
K (x) = −⎜ K l A⎟
⎝ ∂ x f + A
T
(18.11)
⎠
Kc
Kg
This equation includes two physical effects: First, K c denotes the so-called passive
stiffness as known from a single spring. It includes the elasticity of the cables and
the compliance of the winch position controllers, respectively. Second, K g describes
the active stiffness [10]. This active stiffness describes that for the cable robot being
in equilibrium, any platform displacement by an external load leads to a reaction
force that is opposite to the disturbance. This force is due to the resulting changes
in the structure matrix and increases with both the changes of the structure matrix
and the forces level in the cables. Note, that the active stiffness is not related to
the material of the cables, but a property of the parallel topology of the system
that always tries to re-establish a state of lowest energy in force equilibrium. As an
example of active stiffness, one might think of a planar cable robot with three cables
in a triangular configuration attached to a point-shaped platform. Assume the cables
are inelastic(!) and not driven by motors but tensed by simple weights attached to
the loose cable ends. Depending on the weights, the system will always go to a
pose of force equilibrium at the platform. Any external load will lead to a platform
displacement where the forces exerted by the cables will pull the platform back to
the original pose. This effect resembles a stiffness, but does not rely on material
elasticity.
18.3 Optimization Procedure of Stiffness and Cable Forces
Since in Sect. 18.2 a model for elasticity was introduced, the stiffness of the cable
robot now can be optimized. As the stiffness properties are modelled in the stiffness
matrix K in Eq. (18.11), a compliance matrix C = K −1 can be defined that describes
how the platform pose x changes if an external disturbance force δw is applied:
δx = K −1 δw = Cδw (18.12)
Now for a given pose—or even a set of poses—Eq. (18.12) allows to numerically
calculate the displacement x for an applied wrench w. In order to finally be able to
derive a scalar measure for the compliance, these displacements must be computed
in homogenized form (see [8, 11, 12]) by the following matrix

J v = diag 1, 1, 1, jx , j y , jz (18.13)
where
1 P
m
jx = p xi (18.14)
m i=1
1 P
m
jy = p yi (18.15)
m i=1
1 P
m
jy = pzi (18.16)
m i=1
give the median of the distance of the cable attachment points to the origin of the
platform-fixed coordinate system per Cartesian direction. To homogenize the dis-
placement, the structure matrix applied in Eq. (18.11) (and implicitly Eq. (18.12))
has to be used in its homogenized form AhT = J −1 T
v A , resulting in an homogenized
displacement x h that can be used to compute a scalar measure for compliance
κ = x h 2 . (18.17)
Besides stiffness, most cable robot designs aim at requiring low cable forces in
its workspace. In addition to the potential for smaller drives, this helps to provide
a good wrench capability of the platform. For a given platform pose x and load w,
cable forces f can be calculated based on Eq. 18.1. Note, this is usually done using
numerical approaches as e.g. presented in [9]. From the resulting forces f ,
max( f i )
μ= (18.18)
min( f i )
can be computed as a measure for cable forces that are low, but still avoid slackness.
The optimization within this paper is performed using genetic algorithms (GA). As
both stiffness of the platform according to Eq. (18.17) and low cable forces according
to Eq. (18.18) are demanded, a carefully designed sequence of optimization steps
has been followed as presented by [13, 14]:
I. The intended workspace is discretized. On the resulting grid points, the cost
function—which is still to be designed—is evaluated later on.
II. The cable robot design is defined by the predefined arrangement of the eight
pulleys attached to the base frame (set as eight position vectors b1 to b8 ) and
the arrangement of the eight cable connection points on the platform (computed
with respect to the platform-fixed coordinate system as p1 to p8 ). As the frame
parameters were fixed by the application requirements, this still gives a total
number of 8 × 3 = 24 parameters to describe the platform geometry. Since this
is a large number of parameters to be optimized, an alternative parametriza-
tion is proposed using only nine symbols as shown in Fig. 18.2. The applied
parametrization bases on the assumption that the platform includes both bottom
and top panels which may be varied in their arrangement. Both panels are rect-
angular and carry the cable connection points. This approach both reduces the
number of parameters and supports the genetic inheritance steps of the GA well.
Note, that all parameters explicitly may also get negative(!) values assigned by
the optimizer. This allows e.g. to get designs with crossed cables (Figs. 18.3
and 18.4).
III. Note, that during the optimization done in the next steps, for each design a
collision check is performed for collisions between the cables and between the
cables and the platform. In case a collision is detected, a penalty term is applied.
IV. Using Eq. (18.17), a cost function cκ only minimizing the average platform
compliance on the grid points is used and a GA run is performed. The resulting
minimum cost value vκ is stored.
Fig. 18.2 Parametrization of platform for optimization. Based on: [14]
Fig. 18.3 Workspace of original design
V. Using Eq. (18.18), a cost function cμ only minimizing the average tension level
on the grid points is used as the forces needed to approach a storage position
should be low. Now a GA run is performed. Here, the resulting minimum
cost value is stored as vμ . Now, for each optimization criterion a measure for
the realizable cost value is at hand. These values can now be used for a first
weighting of both criteria by applying them as weights in the following steps
vμ
as λκ = 1 − vκ v+v
κ
μ
and λμ = 1 − vκ +v μ
.
Fig. 18.4 Workspace of improved design
VI. Using these weights that balance the criteria, a final GA optimization run com-
bining stiffness and cable forces is performed. The cost function cg now includes
the weighted sum of cκ and cμ as
cg = ηλκ cκ + (1 − η)λμ cμ (18.19)
where η is a second, user-defined weighing factor to influence the balancing between

both criteria. Note, that designs having generally high cable forces might improve
cκ due to the active stiffness term K g , but are not desirable for power consumption,
performance of the system and mechanical design.
18.4 Optimization Results and Discussion
The approach described in the last section was implemented in MATLAB® , using
the Global Optimization Toolbox™ and its genetic algorithm as implemented in the
ga() function.
As an external reference wrench w is needed, a wrench was composed that
addresses the major criteria:
• The stiffness is extremely critical in the normal direction of the high rack (y
direction). This is mainly for two reasons. First, from the geometry of the cable
robot, in this direction the stiffness is generally low and needs to be improved.
Table 18.1 Improved geometry. Unit is meters [m]

Axis Cable Cable Cable Cable Cable Cable Cable Cable
no. 1 no. 2 no. 3 no. 4 no. 5 no. 6 no. 7 no. 8
Platform geometry ( p1 to p8 )
x −0.3 −0.35 0.3 0.35 −0.35 −0.35 0.35 0.35
y −0.3 0.3 −0.3 0.3 −0.25 0.25 −0.25 0.25
z 0.35 −0.25 0.35 −0.25 0.3 −0.3 0.3 −0.3
Frame geometry (b1 to b8 )
x −7.5 −7.5 7.5 7.5 −7.5 −7.5 7.5 7.5
y 0.4 −0.4 0.4 −0.4 0.4 −0.4 0.4 −0.4
z 10.5 10.5 10.5 10.5 0.5 0.5 0.5 0.5
Second, if vibrations occur, a large amplitude in this direction might lead to a

collision between the high rack and the moving platform which is fatal and leads
to severe damage. Thus, in y direction, a force of 50 N is applied.
• Gravity has a major influence on the system, acting in z direction. Generally, the
stiffness in the z direction is no issue for the predominant area of the high rack,
but still may be optimized in upper regions. Thus, a force of –500 N is chosen in
z direction.
• Torques are typically limited except for the situation of loading and unloading
where large torques around the x axis might occur, depending on the type of
loading and unloading mechanism. To be prepared for these disturbances, a torque
of 25 Nm around the x axis is applied.
Concluding, w = [0, 50 N, −500 N, 25 Nm, 0, 0]T was chosen for the computa-
tions within the stiffness optimization procedure. Furthermore, within this investi-
gation, η = 2/3, as this is a reasonable balancing between the requirements for low
cable forces forces and high stiffness.
The optimization sequence was performed several times. As the number of gener-
ation in the GA was limited and due to mutation, different results were generated that
finally had to be judged by application engineers. In the following table, the geome-
try parameters of the most promising optimization result are given. Note, the frame
geometry is only given for completeness, but fixed by the application requirements
(Table 18.1).
Furthermore, both the original design as realized in 2014 and the optimized design
are shown in the following figures. First, the geometry is visualized in Figs. 18.5 and
18.6. The optimizer converged towards a design that utilizes crossed cables in several
projections.
Now Figs. 18.7 and 18.8 show the compliance of the original and improved
design. Note, that the figures were computed based on a limited wrench of w =
[0, 10 N, 0, 10 Nm, 0, 0]T . This is different from the wrench used for the optimization
procedure, as the latter one could not be supported by the original design in most
areas of the workspace. This already demonstrates the superior properties of the
Fig. 18.5 Original design
Fig. 18.6 Improved design
new design. As indicated by the applied scale, the improved design shows drastical
enhancements of the stiffness properties. Still, and as indicated by Figs. 18.3 and
18.4, the workspace of the new design is slightly smaller which correlates with
higher forces in the boundary regions of the workspace. As this drawback is limited
to some storage positions and the high rack is fully covered by the workspace, this
is acceptable (Figs. 18.5 and 18.6).
Fig. 18.7 Compliance of original design (lower is better)
Fig. 18.8 Compliance of improved design (lower is better)
18.5 Conclusions and Outlook
This paper presented the improvement on the design of a cable-driven rack feeder. By
formulating advanced stiffness criteria and a dedicated optimization procedure, the
stiffness could be dramatically improved. A slightly higher cable force level needs to
be accepted. Here, future investigations on passive elements like counterweights or
springs will help to lower the required winch drive power and thus contribute to a fast
and energy-efficient approach for intralogistics and warehousing. The realization of

an improved prototype is planned for 2018.
Acknowledgements The research leading to these results has received funding from the
EFRE.NRW (2014–2020) Joint Research Funding Programme of the European Union (EFRE)
and the Ministry of Economy, Energy, Industry, and Handicrafts of the German Federal State of
North Rhine-Westphalia (NRW) under grant agreement EFRE-0800365 (ML-1-1-019B, LEAN)]
References
1. Hassan, M., Khajepour, A.: Analysis of a large-workspace cable-actuated manipulator for

warehousing applications, vol. 49040, pp. 45–53 (2009). https://doi.org/10.1115/detc2009-
87424
2. Bruckmann, T., Lalo, W., Sturm, C., Schramm, D., Hiller, M. (eds.): Design and realization of
a high rack storage and retrieval machine based on wire robot technology. In: DINAME—In-
ternational Symposium on Dynamic Problems of Mechanics, 17 February 2013–22 February
2013, Committee of Dynamics of ABCM, the Brazilian Society of Engineering and Mechanical
Sciences, Buzios, Rio de Janeiro, Brasilien (2013)
3. Fehlberg, L., Reichert, C., Bruckmann, T., von Zitzewitz, J.: Ausnutzung energiespeichern-
der Elemente zur Effizienzsteigerung seilbasierter Regalbediengeräte. In: Bertram, T., Corves,
B., Janschek, K. (eds.) Fachtagung Mechatronik, 08 March 2013, Aachen: Fachtagung Mecha-
tronik; VDI/VDE-Gesellschaft Mess- und Automatisierungstechnik, Gesellschaft Produkt- und
Prozessgestaltung, pp. 171–176 (2013)
4. Reichert, C., Glogowski, P., Bruckmann, T. (eds.): Dynamische Rekonfiguration eines seil-
basierten Manipulators zur Verbesserung der mechanischen Steifigkeit. Tagung VDI Mecha-
tronik, 12–13 March 2015, Dortmund, Deutschland (2015)
5. von Zitzewitz, J., Fehlberg, L., Bruckmann, T., Vallery, H.: Use of passively guided deflec-
tion units and energy-storing elements to increase the application range of wire robots. In:
Bruckmann, T., Pott, A. (eds.) First International Conference on Cable-driven Parallel Robots
(CableCon 2012), 2–4 September 2012, pp. 167–184. Springer, Stuttgart (2013)
6. Abdolshah, S., Zanotto, D., Rosati, G., Agrawal, S.K.: Optimizing stiffness and dexterity of
planar adaptive cable-driven parallel robots. J. Mech. Robot. 9(3), 031004–031011 (2017).
https://doi.org/10.1115/1.4035681
7. Dagalakis, N.G., Albus, J.S., Wang, B.L., Unger, J., Lee, J.D.: Stiffness study of a parallel
link robot crane for shipbuilding applications. J. Offshore Mech. Arct. Eng. 111(3), 183–193
(1989). https://doi.org/10.1115/1.3257146
8. Reichert, C., Bruckmann, T. (eds.): Unified contact force control approach for cable-driven
parallel robots using an impedance/admittance control strategy. In: Proceedings of the 14th
World Congress in Mechanism and Machine Science, 25–30 October 2015, Taipei, Taiwan
(2015)
9. Gouttefarde, M., Lamaury, J., Reichert, C., Bruckmann, T.: A versatile tension distribution
algorithm for n-DOF parallel robots driven by n + 2 cables. Robotics 31(6), 1444–1457 (2015).
https://doi.org/10.1109/TRO.2015.2495005
10. Reichert, C., Unterberg, U., Bruckmann, T. (eds.): Energie-optimale Trajektorien für seil-
basierte Manipulatoren unter Verwendung von passiven Elementen. In: Proceedings of the 1st
IFToMM D-A-CH Conference, Dortmund, Deutschland (2015)
11. Nguyen, D.Q., Gouttefarde, M.: Stiffness matrix of 6-DOF cable-driven parallel robots and its
homogenization. In: Lenarčič, J., Khatib, O. (eds.) Advances in Robot Kinematics, pp. 181–191.
Springer International Publishing, Cham (2014)
12. Stocco, L., Salcudean, S.E., Sassani, F. (eds.): Matrix normalization for optimal robot design.
In: Proceedings 1998 IEEE International Conference on Robotics and Automation (Cat
No98CH36146), 16–20 May 1998 (1998)
13. Miettinen, K.: Nonlinear Multiobjective Optimization, vol. xvii, 298 p. Kluwer Academic
Publishers, Boston (1999)
14. Passens, H.: Synthese seilbasierter Manipulatoren unter Verwendung numerischer Optimierung
[Thesis]. University of Duisburg-Essen, Duisburg (2016)
15. Bruckmann, T., Sturm, C., Fehlberg, L., Reichert, C.: An energy-efficient wire-based stor-
age and retrieval system. In: IEEE/ASME International Conference on Advanced Intelligent
Mechatronics (AIM), 09 July 2013–12 July 2013, Wollongong, Australia, pp. 631–636. IEEE
(2013)
Chapter 19
Cascade ADRC Based Fault-Tolerant
Control for a Hovering PVTOL Aircraft
Xinli Xu and Chunwei Zhang
Abstract This paper presents a novel reconfigurable control method for the pla-
nar vertical take-off and landing (PVTOL) aircraft with potential actuator failures.
According to the position subsystem within the multivariable coupling, and the series
between subsystems of position and attitude, a cascade active disturbance rejection
controller (ADRC) is used to counteract the adverse effects when the actuator fail-
ure occurs. The coordinate transformation is used for model decoupling due to the
severe coupling between some variables. This approach does not require the accurate
mathematical model of the controlled system and ensures that the reference input
value can be tracked rapidly and accurately. In addition, the Taylor differentiator is
designed to improve the control precision based on the detailed research for tracking
differentiator. The stability and safety of the aircraft is much improved in the event
of actuator failures. Finally, the simulation results are given to show the effectiveness
and performance of the developed method.
Keywords Cascade ADRC · PVTOL aircraft · Coordinate Transformation ·

Fault-Tolerant Control · Actuator failure
19.1 Introduction
Over the past decades, due to the increasing demands for reliability and survivabil-
ity in modern aircraft, fault-tolerant control (FTC) has been widely developed and
used in the flight control system. Though many algorithms have been developed, the
research of FTC is still a challenging with the complexity and non-linearity of flight
control system on air and space vehicles such as civil aircraft, combat aircraft, satel-
lite, unmanned aerial vehicles (UAVs), vertical/short takeoff and landing (V/STOL)
aircraft, etc. [1]. In particular, the planar vertical take-off and landing (PVTOL) air-
craft is gaining more attention during the last few years. It is in fact a simplified
version of actual aircraft model and maintains the main characteristics [2]. Many
X. Xu · C. Zhang (B)
School of Civil Engineering, Qingdao University of Technology, Qingdao 266033, China

240 X. Xu and C. Zhang
control strategies have been studied in order to fulfill the increasing safety demand
and flight performance. It is extremely important to guarantee system security and
reliability in the event of various failures, namely uncertainties, parametric perturba-
tions, and external disturbances, due to unmolded dynamics, control surface damage,
actuator failures, and sensor failures, etc.
In [1], a control scheme based on output feedback is introduced for the regulation
of a PVTOL craft. This work uses a novel SMC and an energy control approach to
stabilize or balance the vertical variable and horizontal and angular variables, respec-
tively [3] presented a SMC approach for stabilization of a PVTOL aircraft combing
with super-twisting algorithms, and it focused on lateral motion of autonomous air-
craft. In order to stabilize a model of PVTOL aircraft, an output feedback method was
explored in [4]. In this work, the finite-time convergent control was used to construct
a stabilizer, and a fast convergent observer was designed to stabilize the system. In
[5], a SMC scheme was focused on combing with output feedback technology to reg-
ulate the PVTOL aircraft including actuators dynamics. The velocity of the aircraft
was estimated by a SMO (sliding mode observer) which was designed respectively,
and the feedback control was also investigated. An adaptive control for the PVTOL
craft could also be found in [6], which proposed minimum projection and dynamic
extension algorithm based on Lyapunov function, and an optimal control method
was also used.
In this paper, a novel reconfigurable control method based Cascade ADRC is
proposed for the PVTOL aircraft with potential actuator failures. The purpose is
mainly to use the proposed method to track the input instruction in the event of
actuator failures and achieve the stability of the system. Simulation is conducted to
show its performance and effectiveness.
The rest of the paper is organized as follows. Section 19.2 is focused on the
dynamic model of the PVTOL aircraft, the model decoupling and the statement of
faulty aircraft. Section 19.3 describes the fault tolerant control strategy. Simulation
results are presented and discussed in Sect. 19.4. Finally, a brief conclusion and
future work are given in Sect. 19.5.
19.2 Dynamic Models and Problem Statement
19.2.1 Dynamic Model of PVTOL Aircraft
The minimum number of state variable and input for hovering control of V/STOL
aircraft is retained by using the PVTOL aircraft model as a benchmark model.
Figure 19.1 shows the couple relationship between the roll moment and lateral thrust.
The roll torque U2 of a V/STOL aircraft is generated by jet engines coordination
control, which is under the both sides of wing tips. When the thrust that is generated
by engines under wingtips is not vertical with horizontal direction one, a lateral
19 Cascade ADRC Based Fault-Tolerant Control … 241
Fig. 19.1 Schematic

drawing of the PVTOL
aircraft
thrust ε0 U2 will be generated with the roll torque. This is why a V/STOL aircraft has
non-minimum phase characteristics.
The horizontal and vertical motion of aircraft mass center Y and Z are normally
treated as the output of the system, and the roll angle θ is used as internal state of
the system, which is relative to the horizontal direction. The main thrust U1 and roll
torque U2 on aircraft mass center are used as the control variable of the system.
In terms of aircraft modelling, the aircraft will be considered as a rigid system and
the bending effect of its fuselage and wings will be ignored. As shown in Fig. 19.1,
its motion equations can be established in the following form:
⎧
⎨ mŸ = −U1 sin θ + ε0 U2 cos θ + ω1
mZ̈ = U1 cos θ + ε0 U2 sin θ − mg + ω2 (19.1)
⎩
J θ̈ = U2

where Y and Z are the horizontal and vertical motion respectively; θ (|θ | << π 2)
is the roll angle which is relative to the horizontal direction; U1 and U2 are the main
thrust and roll torque; g is the acceleration due to gravity which can be normalized to
1; J is the mass moment of inertia; ε0 is the coupling coefficient between the lateral
thrust and roll torque; ω1 and ω2 are the bounded disturbance terms.
To get a simplified form, the new variables are introduced below:

x = −Y g, y = −Z g, u1 = U1 (mg), u2 = U2 J , ε = ε0 J (mg)
Then we have the following motion equations:

⎧
⎨ ẍ = −u1 sin θ + εu2 cos θ + ω1
ÿ = u1 cos θ + εu2 sin θ − 1 + ω2 (19.2)
⎩
θ̈ = u2
where x and y are the horizontal and vertical motion of aircraft mass center after
simplification; θ is the roll angle which is relative to the x axis; u1 and u2 are respec-
tively the main thrust and roll torque control; ε ∈ [0, 1] is the coupling coefficient
between the lateral thrust and roll torque and immeasurable.
We can see that the main thrust offset the gravity to maintain the stable system
when the aircraft is hovering, as shown in Fig. 19.1.
Remark 1 u1 and u2 should be not too big due to the limitation of aircraft jet engines
control and u1 > 0. y(t) should have a very small amount of change or y(t) = 0.
19.2.2 Model Decoupling and Problem Statement
The zero dynamic subsystem of a PVTOL aircraft is unstable due to coupling between
the roll torque and lateral thrust. Though the reference input value can be tracked
rapidly and accurately by the output x, the vertical position and the roll angle θ
which are the key internal state indicator cannot be remained stable. Therefore, the
decoupling between ε and y(t) is studied by the coordinate transformation for the
PVTOL aircraft model in this paper, and the Cascade ADRC is also designed in
terms of the model after coordinate transformation.
Consider the control matrix with coupling characteristics of the actual PVTOL
system (19.2), the controlled variable ũ1 and ũ2 are led into, as follows:

u1 = −(ũ1 − ω1 ) sin θ + (ũ2 − ω2 + 1) cos θ
(19.3)
u2 = 1ε (ũ1 − ω1 ) cos θ + (ũ2 − ω2 + 1) sin θ
Then the PVTOL model (19.2) can be converted into:

⎧
⎨ ẍ = ũ1
ÿ = ũ2 (19.4)
⎩
θ̈ = ũ2
We define:

ũ1 ū1 − εθ̇ 2 εū2 sin θ − cos θ ω1
= + (19.5)
ũ2 ū1 − εθ̇ 2 εū2 cos θ sin θ ω2 − 1

x x̃ ε sin θ
= + (19.6)
y ỹ −ε cos θ
Hence, the PVTOL aircraft model (19.4) can be transformed to:

⎧
⎨ x̃¨ = ū1 sin θ + ω1
ỹ¨ = −ū1 cos θ − 1 + ω2 (19.7)
⎩
θ̈ = ū2
We define
T T
u = ū1 ū2 , x = x̃ x̃˙ ỹ ỹ˙ θ θ̇
Then the Eq. (19.7) can be expressed, as:
ẋ = Ax(t) + Bu(t) + d (t) (19.8)
where
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 1 0 0 0 0 0 0 0
⎢0 0⎥ ⎢ sin θ 0⎥ ⎢ ω (t) ⎥
⎢ 0 0 0 0 ⎥ ⎢ ⎥ ⎢ 1 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢0 0 0 1 0 0⎥ ⎢ 0 0⎥ ⎢ 0 ⎥
A=⎢ ⎥, B = ⎢ ⎥, d (t) = ⎢ ⎥
⎢0 0 0 0 0 0⎥ ⎢ − cos θ 0⎥ ⎢ −1 + ω2 (t) ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣0 0 0 0 0 1 ⎦ ⎣ 0 0⎦ ⎣ 0 ⎦
0 0 0 0 0 0 0 1 0
Generally, actuator failures of the aircraft include loss of effectiveness (LOE),

hard over fault (HOF), float and lock in place (LIP), which can be described below.
ẋ = Ax(t) + Bσ u(t) + d (t) (19.9)
where
σ = diag{σ1 , σ2 , . . . , σm }
⎧
⎪
⎪ uci (t), σk = 1, normal
⎪
⎪
⎪
⎨ ki (t) × uci (t), 0 < σk < 1, LOE
ui (t) = uci (tf ), σk = 0, LIP
⎪
⎪
⎪ 0, σk = 0, float
⎪
⎪
⎩ u oru , σ = 0, HOF
cim ciM k
where ui (t) denotes the actual output signal of the control surface; uci (t) is the control
signal generated by the ith control surface; ki ∈ (0, 1) is the actuator effectiveness
coefficient; ucim and uciM are the minimum and maximum values, i = 1, 2, …, m.
Remark 2 For convenience, it is assumed that only one failure occurs on each control
surface in a certain period of time. It cannot be certain when the fault occurs or how
severe the fault is due to σk ∈ [0, 1] is unknown.
19.3 Control Strategies
In this paper, for the system (19.9), the Cascade ADRC is designed in the channel
where x(t) has a strong coupling with θ , to sustain the stability of the hovering aircraft
during lateral maneuver when the fault occurs. Figure 19.2 shows the system block
diagram of the Cascade ADRC method.
The Cascade ADRC has a strong ability for decoupling and is adaptable to distur-
bance as it is independent of the accurate mathematical model of the controlled plant.
It uses the external disturbance and the internal uncertainty of the control system as
a disturbance while compensating the total disturbance for tracking in real time.
When the actuator failure occurs, the circuit change can be estimated by the
Extended States Observer (ESO), and used to compensate for the system dynamically
since it includes the actuator fault and external disturbance. Then the dynamic and
steady-state performance of the circuit can be obtained by the nonlinear configuration
of the state feedback.
In this paper, the Taylor differentiator is designed as follows:

v̇1 = v2
(19.10)
v̇2 = ε22 [v1 − v] − 2ε v2
where v is given signal, v1 is tracking signal, v2 is tracking speed, 1ε is convergence

rate.
The formula (19.7) can be regarded as the two parts of the position subsystem and
the attitude subsystem, which constitute the outer ring and inner loop of the system
respectively. The two are series connection and can be expressed as:

x̃¨ = ū1 sin θ + ω1
(19.11)
ỹ¨ = −ū1 cos θ − 1 + ω2
Fig. 19.2 The schematic of the proposed FTC


θ̇ = θ̃
(19.12)
θ̃˙ = ū2
Defining:
¨ ẍ2 = ỹ¨
ẍ1 = x̃,
Then the formula (19.11) can be transformed to:

⎧
⎪
⎪ ẍ1 = ū1 sin θ + ω1
⎨
ẍ2 = −ū1 cos θ − 1 + ω2
(19.13)
⎪ y1
⎪ = x1
⎩
y2 = x2
We have that,

sin θ
B= (19.14)
− cos θ
∗ ∗ T
ū1 ū2 = Bū1 (19.15)
where ū1∗ and ū2∗ is the virtual controlled variable respectively.

For the position subsystem (19.11), which is a 2 × 2 coupled system, the ESO
can be established respectively, as follows:
⎧
⎪
⎪ e = z11 − x1
⎨
ż11 = z12 − β01 e
(19.16)
⎪ ż12 = z13 − β02 fal(e, 0.5, h) + ū1∗
⎪
⎩
ż13 = −β03 fal(e, 0.25, h)
⎧
⎪
⎪ e = z21 − x2
⎨
ż21 = z22 − β01 e
(19.17)
⎪
⎪ ż = z23 − β02 fal(e, 0.5, h) + ū2∗
⎩ 22
ż23 = −β03 fal(e, 0.25, h)
where β01 , β02 and β03 are the parameters of observer which need to determine. z11 ,
z12 , z13 , z21 , z22 and z23 are the input variables of ESO. fal(e, 0.5, h) and fal(e, 0.25, h)
are the nonlinear function. h is the integral step. The system will be tracked by z11 ,
z12 , z13 , z21 , z22 and z23 when the ESO in (19.17) is tuned properly.
Then the virtual controlled variable ū1∗ and ū2∗ can be obtained:
⎧
⎨ e1 = v11 − z11
e = v12 − z12 (19.18)
⎩ 2∗
ū1 = −fhan(e1 , e2 , r, h1 ) − z13
⎧
⎨ e1 = v21 − z21
e = v22 − z22 (19.19)
⎩ 2∗
ū2 = −fhan(e1 , e2 , r, h1 ) − z23
The actual control variable can be obtained by the formula (19.14), (19.15), (19.18)
and (19.19):
T
ū1 = B−1 ū1∗ ū2∗ (19.20)
From above, the attitude subsystem (19.12) and the position subsystem are series
connected, and then the ESO can be established as follows:
⎧
⎪
⎪ e = z31 − θ
⎨
ż31 = z32 − β01 e
(19.21)
⎪
⎪ ż = z33 − β02 fal(e, 0.5, h) + ū2
⎩ 32
ż33 = −β03 fal(e, 0.25, h)
After disturbance compensation, the control variable can be obtained by the error
feedback control law:
⎧
⎨ e1 = U − z31
e = −z32 (19.22)
⎩ 2
ū2 = fhan(e1 , e2 , r, h2 ) − z33
where U = ū1∗ + ū2∗ .

As mentioned above, β01 , β02 and β03 need to be tuned [7] uses the bandwidth
theory to determine the relevant parameters. For example, the characteristic equation
of ESO (19.16) can be expressed, as:
D(s) = s3 + β01 s2 + β02 s + β03 (19.23)
It is shown that the estimator is better when the characteristic Eq. (19.23) is stable
and then the equation coefficients meet the requirements.
Defining: D(s) = (s + ω0 )3
Then β01 , β02 and β03 can be converted to the function of the observer bandwidth
ω0 :
⎧
⎨ β01 = 3ω0
⎪
β02 = 3ω02 (19.24)
⎪
⎩
β03 = ω03
Hence, we will obtain a good performance for Cascade ADRC by tuning the
parameters.
19.4 Simulation Results
In this section, an example of a PVTOL aircraft model is presented to demonstrate

the feasibility of the proposed control method, as shown in Fig. 19.3, A Cascade
ADRC is used to counteract the adverse effects when the actuator failures occur. For
convenience, a simplified version of the actual aircraft model, namely, the PVTOL
is used to deal with the potential actuator failures.
The MATLAB software is used for a couple of numerical simulations. For com-
parison, the results under normal and fault are presented simultaneously. The initial
condition of the aircraft is x(0) = 0, and the control parameters are used as the
following:
b0 = 0.9, 0.9, −18 b1 = 0.8, 387, 0.17

b2 = 1.8, 39, 0.8 ω0 = 85, 1, 88
Two scenarios are considered in the numerical simulations: a loss of 30% con-
trol effectiveness and a loss of 60% control effectiveness in elevator. As shown in
Fig. 19.4, the fault is injected at the 20th second, which may lead to transient loss of
horizontal motion and roll angle, while the vertical motion is affected little. Using
the proposed method, the system stabilizes asymptotically quickly.
19.5 Conclusions and Future Work
A cascade active disturbance rejection controller is presented to accommodate partial

loss fault for the PVTOL aircraft. It does not require the accurate mathematical
model of the aircraft and fault detection module. For convenience, the coupling
between variables is decoupled by coordinate transformation. Two scenarios are
considered with respect to loss of 30% and 60% control effectiveness in an elevator
and the resulting control law is validated in the numerical simulations. The results
Fig. 19.3 The model of VTOL aircraft

3 3
no fault no fault
elevator 30% loss elevator 60% loss
2.5 2.5
Horizontal motion x
Horizontal motion x
2 2
1.5 1.5
1 1
0.5 0.5
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
time (sec) time (sec)
(a) x
-3 -3
x 10 x 10
2 2
no fault no fault
0 elevator 30% loss 0 elevator 60% loss
Vertical motion y
Vertical motion y
-2 -2
-4 -4
-6 -6
-8 -8
-10 -10
-12 -12
-14 -14
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(b) y
3 3
no fault no fault
elevator 30% loss elevator 60% loss
2 2
1 1
θ (rad)
θ (rad)
0 0
-1 -1
-2 -2
-3 -3
0 10 20 30 40 50 60 0 10 20 30 40 50 60
(c) θ
Fig. 19.4 a The horizontal output with respect to loss of 30 and 60% control effectiveness, b the
vertical output with respect to loss of 30 and 60% control effectiveness, c the roll angle with respect
to loss of 30 and 60% control effectiveness
demonstrate the satisfactory performance and robustness of the proposed method in

the event of actuator failures.
The future work is to apply the presented method in a VTOL aircraft which is no
longer a simplified version. The performance and effectiveness will be tested in an
actual aircraft model and then it could be used in real aircrafts for a variety of tasks.
References
1. Fikret, C., Chingiz, H.: Reconfigurable control of an UAV against sensor/actuator failures. IFAC
48(9), 7–12 (2015)
2. Carlos, A.-I., Miguel, S.S.-C., Nareli, C.-C.: Output feedback stabilization for a PVTOL aircraft
based on a sliding mode combined with an energy control strategy. In: The 12th International
Conference on Electrical Engineering, Computing Science & Automatic Control, pp. 1–6 (2015)
3. Alexander, P., Boris, A., Iuliia, Z.: Stabilization of PVTOL aircraft by supertwisting algorithms.
In: International Workshop on Recent Advances in Sliding Modes (RASM), pp. 1–6 (2015)
4. Frye, M.T., Ding, S., Qian, C., Li, S.: Fast convergent observer design for output feedback
stabilisation of a planar vertical takeoff and landing aircraft. IET Control Theory Appl. 4(4),
690–770 (2010)
5. Ricardo, C., Luis, T.A.: Output feedback sliding mode control of a PVTOL including actuators
dynamics. In: Proceedings of IEEE International Conference on Control Applications (CCA),
pp. 1482–1486 (2011)
6. Soki, K., Hisakazu, N., Ysuyuki, S.: Adaptive control for the PVTOL system via minimum
projection method. IFAC 48(11), 216–221 (2015)
7. Gao, Z., Huang, Y., Han, J.: An alternative paradigm for control system design. In: Proceedings
of IEEE Conference on Decision and Control, vol. 5, 4578–4585, 4–7 Dec 2001
Chapter 20
Power Analysis of Epicyclic Gear
Transmission for Wind Farm
N. Hu, S. Liu, D. Zhao and C. Chen
Abstract Epicyclic gear transmissions are intensively used in wind turbines for
driving the electric generators. To maximise the efficiency and minimise the rate
of failure, it is essential to gain a full and precise understanding of the power flow
inside the epicyclic gear systems. An algorithm based on constraint analysis and
virtual power ratio is applied to a one-degree-of-freedom, four-stage gearbox of a
wind turbine for prediction of efficiency. Power flow diagrams are constructed to
obtain the power flow patterns and validate the prediction, by means of balance in
normal and virtual power of the system. The analytical solution is derived for the
efficiency of the gearbox. The analytical solution reveals that all loss factors are
inversely proportional to the efficiency. Further, assuming equal loss factors on all
gear meshes, the maximum loss factor is 0.147 to yield a positive efficiency, i.e. no
self-locking of the gearbox.
Keywords Epicyclic gear transmission · Wind farm · Wind turbine · Constraint

analysis · Efficiency
20.1 Introduction
An epicyclic gear train (EGT) consists of at least one planet gear that rotates on a
planet carrier about a central axis [1]. The EGTs are widely used in robotic arms
[2], hybrid vehicle power transmissions and turbine generators [3], thanks to the
high-power density, compact structure and large payload capacity.
N. Hu · S. Liu · C. Chen (B)

Laboratory of Motion Generation and Analysis, Department of Mechanical and Aerospace
Engineering, Monash University, Clayton, VIC 3800, Australia
D. Zhao
China Energine International, Wan Chai, Hong Kong

252 N. Hu et al.
However, the mechanical efficiencies of EGTs are usually lower than those of
simple gear trains [4]. The principle power sinks are sliding friction between the
gear teeth surfaces, lubrication oil churning, and friction in shaft support bearings
[5]. Recent research [6] shows that the power loss in a gear transmission is caused
by gears, bearings, seals, and auxiliaries. Further, the gear and bearing losses include
no-load and load-dependent losses. In this work, the load-dependent loss of gear
mesh is considered in the analysis, as it is dominant in power transmissions.
The majority of horizontal axis wind turbines contain EGTs, which convert slow
rotation of the blades into a fast rotation that is more efficient to drive the generator.
Precise kinematic analysis and prediction of the efficiency play a vital role in the
design and optimisation of EGTs. A quick briefing on the important contributions
is provided below. Radzimovsky [5, 7] proposed the ratio of tooth mesh losses of
the same gear pair in epicyclic and simple gear trains to calculate the power losses.
Kahraman et al. [8] proposed a general formulation for kinematic analysis and power
flow analysis. Salgado and Del Castillo [9] developed power flow maps to analysed
efficiency of EGTs. Chen and Angeles [10] developed the concepts of virtual power
and virtual power ratio, by which detailed power flow patterns can be drawn. This
concept was extended to two-DOF epicyclic trains [11] and compound epicyclic
trains [12]. Fernandes, C. M. [13, 14] devised an average coefficient of friction
between meshing gears. Further, applied the lubricant factor of wind turbine gear
oils and the loss factor for helical gears to predict the power loss in a parallel axis and
a planetary gearbox. Fernandes, C. M. [15] predicted the power loss of a 2.5 MW
wind turbine gearbox lubricated with different wind turbine gear oils.
A general approach based on mechanical constraints is reported in [1]. It computes
the internal power flow through all links. The total power loss is obtained from
the genuine power through each gear mesh. It is assumed that the power losses
at individual gear meshes are independent of each other, i.e. no coupling effect is
considered. In this work, the constraint analysis is used to predict the efficiency of a
four-stage EGT in a wind turbine.
The rest of the manuscript is organised as follows: the framework of the constraint
analysis is briefed in Sect. 20.2. The analysis of the EGT is presented in Sect. 20.3.
Validation of the analysis using power flow diagram, as well as the analytical solution
of the efficiency is given in Sect. 20.4.
20.2 The Framework of the Constraint Analysis
The constraint analysis [1, 10] computes the internal and external power of the EGT
based on mechanical constraints. A summary is provided in this section for the
completeness of the manuscript.
The configurations of the bodies of an EGT can be described by an n-dimensional
column vector q of generalised coordinates. Two generalised coordinates are needed
for each body: the linear and angular velocities ν and ω, respectively, as shown in
Fig. 20.1.
20 Power Analysis of Epicyclic Gear Transmission … 253
Fig. 20.1 Graphical representation of bearing and gear mesh constraints
There are two kinds of mechanical constraints in an EGT: the bearing and gear
mesh. For a bearing constraint, the linear velocities of the bearing and the body it is
attached to (at the location of the bearing) are identical. Referring to Fig. 20.1, the
following constraint equation can be derived for the bearing located at the centre of
body i:

vi − v j + ω j d ji = 0 (20.1)
where d ji is the distance between the centres of bodies i and j.

For a gear mesh, the linear velocities of the gears at the contact are identical.
Referring to Fig. 20.1, the following constraint equation can be derived for the gear
mesh between gears u and v:

(vu + ωu duv ) − vv − ωv j dvu = 0 (20.2)
where d uv and d vu are the distances from the centres of the bodies to the contact,
respectively. Both Eqs. (20.1) and (20.2) relates two bodies (referred to as “body A”
and “body B” in Sect. 20.3), respectively.
For an EGT with m mechanical constraints, the corresponding constraint equations
are summarised in an m-dimensional column vector, c. The space spanned by c is
named the constraint space. Similarly, the space spanned by q is named the coordinate
space. The basis vectors in the two spaces are given by two identity matrices, I n and
I m , for the coordinate and constraint space, respectively
I n = [x1 x2 . . . xm ] (20.3)
I m = [ y1 y2 . . . ym ] (20.4)
254 N. Hu et al.
The generalised coordinates q can be further classified into independent and

dependent generalised coordinates, or qi and qd , respectively. The independent gener-
alised coordinates describe the inputs to the EGT. Individual generalised coordinates
can be expressed using qi and the elements xn from the basis vector I n as
q = Gq i (20.5)
where
−1
G = −Sd C q Sd C q Si + Si (20.6)
C q = ∂ c/∂q (20.7)
In Eq. (20.6), Si and Sd are the independent and dependent selection matrices,
respectively. Both matrices are constructed using xn . Examples of the S matrices are
given in Sect. 20.3.
The generalised constraint forces applied on individual generalised coordinates,
Qc , can be related to the generalised external forces, Qe , as
Qi = Qc + Qe (20.8)
where Qi represents the generalised inertia forces. In this work, the EGT is assumed
to be quasi-static, resulting in a zero Qi . Given such an assumption, Eq. (20.5) can
be further written as
Q e = − Q c = −C qT λ (20.9)
where λ represents the generalised constraint forces along the mechanical constraints.
Qe can be further classified into Qei and Qed , based on the individual generalised
coordinates they are applied onto. Qe can be expressed in terms of Qed and the
elements xn from the basis vector I n as
Q e = H Q ed (20.10)
where
T −T
H = −Si C q Si C q Sd + Sd (20.11)
The internal power of the EGT, P c , which describes the power transfer within the
EGT, is given by
⎡ ⎤ ⎡ ⎤
p1c −λT y1 y1T C q Sb1 (Sb1 )T q
⎢ p ⎥ ⎢ −λT y2 y T Cq Sb (Sb )T q
c ⎥
⎢ 2⎥ ⎢ 2 2 2 ⎥
Pc = ⎢ . ⎥ = ⎢ .. ⎥ (20.12)
⎣ .. ⎦ ⎣ . ⎦
pnc −λT ym ymT C q Sbm (Sbm )T q
where ym are the elements from the basis vector I m . Sbm are the selection matrices
to pick the “body B” corresponding to individual constraint equations, and are con-
structed using x n from the basis vector In. Examples of Sbm are given in the next
section.
The external power of the EGT, P eb , is given by
⎡ ⎤ ⎡ T ⎤
p1eb Q e x 1 x 1T q
⎢ p eb ⎥ ⎢ Q T x 2 x T q ⎥
⎢ 2 ⎥ ⎢ e 2 ⎥
P eb =⎢ . ⎥=⎢ .. ⎥ = Tq (20.13)
.
⎣ . ⎦ ⎣ . ⎦
pbeb Q eT x n x nT q
While T is a diagonal matrix with diagonal entries being those of external power,
P eb
⎡ ⎤
Q eT x 1 x 1T ⎡ eb ⎤
⎢ ⎥ P1 · · · 0
⎢ Q eT x 2 x 2T ⎥ ⎢ .. . . .. ⎥
T =⎢ .. ⎥=⎣ . . . ⎦ (20.14)
⎣ . ⎦
0 · · · P eb
Q eT x n x nT n
20.3 Analysis of the Gearbox
The schematic diagram of the targeted four-stage EGT is presented in Fig. 20.2.
There are ten bodies in this EGT. Body 1 is the ground. Bodies 2 and 10 are the
output and input shafts, respectively. The pitch radii of the gears are RP, RC, RS
and RR for the planet gear, carrier, sun gear and the ring gear respectively. The pitch
radii of the gears are: Rs4 = 1400 mm, Rp4 = 5600 mm, Rr4 = 127000 mm, Rc4 =
7000 mm, Rs3 = 1600 mm, Rp3 = 5600 mm, Rr3 = 12800 mm, Rc3 = 7200 mm, Rs2
= 3200 mm, Rp2 = 6000 mm, Rr2 = 15200 mm, Rc2 = 9200 mm, Rs1 = 5500 mm,
Rp1 = 7000 mm, Rr1 = 19500 mm, Rc1 = 12500 mm.
20.3.1 Constraint Equations
The EGT is fully described using 20 generalised coordinates, as defined by

256 N. Hu et al.
Fig. 20.2 Schematic diagram of the EGT
q = [q1 q2 q3 . . . q18 q19 q20 ]T = [vg ωg vs4 . . . ω p1 vc1 ωc1 ]T (20.15)
The constraint equations are derived following Eqs. (20.1) and (20.2), the sum-
marised in Table 20.1. There are 17 equations in total.
Using the information from Table 20.1, the Smb terms are constructed using the xn
for the “body B” in the constraint equations. An example is given with S1b .
S1b = [x3 ], S2b = [x5 ], S3b = [x5 x6 ], S4b = [x5 x6 ], S5b = [x7 ],
S6b = [x9 ], S7b = [x9 x10 ], S8b = [x9 x10 ], S9b = [x11 ], S10
b
= [x13 ], S11
b
= [x13 x14 ],
b
S12 = [x13 x14 ], S13
b
= [x15 ], S14
b
= [x17 ], S15
b
= [x19 ], S16
b
= [x17 x18 ], S17
b
= [x17 x18 ]
(20.16)
In Eq. (20.16), S1b is corresponding to the selection matrix for constraint equation
c1 . The “body B” of c1 is body 2, or the sun gear of stage IV. As such, the linear
velocity vs4 is picked, which is the third term in the generalised coordinates, i.e. x3 .
20.3.2 Normal Power Analysis
The normal power is computed by grounding body 1. The corresponding selection

matrices S are
Si1 = [x 1 x 2 x 20 ] (20.17)
Sd1 = [x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x14 x15 x16 x17 x18 x19 ] (20.18)
Table 20.1 Constraint equations for the EGT
Constraints Body A Body B Type Velocity A Velocity B Constraint equation
c1 1 2 Bearing vg vs4 vg − vs4 = 0
c2 4 3 Bearing vc4 + 7000ωc4 vp4 vc4 + 7000ωc4 − vp4 = 0
c3 2 3 Gearing vs4 + 1400ωs4 vp4 − 5600ωp4 vs4 + 1400ωs4 – (vp4 − 5600ωp4 ) = 0
c4 1 3 Gearing vg + 12700ωg vp4 + 5600ωp4 vg + 12700ωg − (vp4 + 5600ωp4 ) = 0
c5 1 4 Bearing vg vc4 vg − vc4 = 0
c7 4 5 Gearing vc4 + 1600ωc4 vp3 − 5600ωp3 vc4 + 1600ωc4 − (vp3 − 5600ωp3 ) = 0
20 Power Analysis of Epicyclic Gear Transmission …

c12 1 7 Gearing vg + 15200ωg vp2 + 6000ωp2 vg + 15200ωg − (V p2 + 6000ωp2 ) = 0
257
258 N. Hu et al.
Table 20.2 Internal and external power of the EGT, body 1 grounded
Constraints Body Body Type Internal power 1 Pc Body External power 1 Peb
A B
c1 1 2 Bearing 0 1 0
c2 4 3 Bearing −(25875*T s4 *ωc1 )/ 11 2 (25875*T s4 *ωc1 )/ 11
c3 2 3 Gearing (25875*T s4 *ωc1 )/ 11 3 0
c4 1 3 Gearing 0 4 0
c6 6 5 Bearing −(25875*T s4 *ωc1 )/ 11 6 0
c7 4 5 Gearing (25875*T s4 *ωc1 )/ 11 7 0
c8 1 5 Gearing 0 8 0
c10 8 7 Bearing −(25875*T s4 *ωc1 )/ 11 10 −(25875*T s4 *ωc1 )/ 11
c11 6 7 Gearing (25875*T s4 *ωc1 )/ 11
c12 1 7 Gearing 0
c13 1 8 Bearing 0
c14 10 9 Bearing −(25875*T s4 *ωc1 )/ 11
c15 1 10 Bearing 0
c16 8 9 Gearing (25875*T s4 *ωc1 )/ 11
c17 1 9 Gearing 0
The input velocities are 0, 0 and ωc1, respectively, for the terms in Eq. (20.17).
The generalised external loads applied (Qed ) onto the generalised coordinates is a
torque T s4 on the output shaft (body 2, the sun gear of stage IV), or
Q ed = [0 Ts4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]T (20.19)
The velocities of all bodies (q) can be expressed using ωc1 , from Eq. (20.5),
and C q can be computed from Eq. (20.7). With q and C q determined, the generalised
constraint forces along mechanical constraints (λ), and the generalised external forces
(Qe ) can be computed with Eqs. (20.9) and (20.10). Substituting λ and Qe into
Eqs. (20.12)–(20.14) yields the internal and external power of the EGT. The results
are listed in Table 20.2. The superscripts 1 for 1 Pc and 1 Peb indicate that body 1 is
grounded.
20.3.3 Virtual Power Analysis
The EGT has four cases in the virtual power analysis, obtained by grounding the
carriers of stages I to IV, respectively. The corresponding S i contains three elements:
Table 20.3 Internal and external power of the EGT, body 10 grounded
Constraints Body Body Type Internal power 10 Pc Body External power 10 Peb
A B
c1 1 2 Bearing 0 1 −(362107*T s4 *ωc1 )/ 154
c2 4 3 Bearing −(25765*T s4 *ωc1 )/ 11 2 (362107*T s4 *ωc1 )/ 154
c3 2 3 Gearing (362107*T s4 *ωc1 )/ 154 3 0
c4 1 3 Gearing −(127*T s4 *ωc1 )/ 14 4 0
c6 6 5 Bearing −(24885*T s4 *ωc1 )/ 11 6 0
c7 4 5 Gearing (25765*T s4 * ωc1 )/ 11 7 0
c8 1 5 Gearing −80*T s4 * ωc1 8 0
c10 8 7 Bearing −(40365*T s4 *ωc1 )/ 22 10 0
c11 6 7 Gearing (24885*T s4 *ωc1 )/ 11
c12 1 7 Gearing −(855*T s4 *ωc1 )/ 2
c13 1 8 Bearing 0
c14 10 9 Bearing 0
c15 1 10 Bearing 0
c16 8 9 Gearing (40365*T s4 *ωc1 )/ 22
c17 1 9 Gearing −(40365*T s4 *ωc1 )/ 22
two from x7/ 8 , x11/ 12 , x15/ 16 and x19/ 20 for stages I to IV grounded, respectively, plus
an x2 in all four cases. The input velocities are obtained by “rotating” the ground
with −ωc1 , −ωc2 , −ωc3 and −ωc4 , respectively.
The virtual internal and external power for each case is computed following the
same procedure as that for the normal power, but with different matrices S and input
velocities. An example is shown in Table 20.3 for the case in which stage I (body
10) is grounded.
20.4 Validation and Efficiency of EGT
The power flow of the EGT is illustrated in Fig. 20.3, where the directions of the
internal power flows are determined according to Table 20.1 (body A to body B for
each constraint equation). The constraint analysis detailed in the previous section is
validated by substituting the corresponding Pc and Peb terms into the figure. It can
be readily seen that a power-balanced system is achieved for both normal and virtual
power (Tables 20.2 and 20.3, respectively), and thereby validates the analysis.
260 N. Hu et al.
Fig. 20.3 Power flow of the

EGT
The genuine meshing power is the gearing power measured on the corresponding
carrier, which is obtained from the outcomes of normal and virtual power analysis
as below. The superscripts for Pc indicate the body to be grounded.
g
P1 = 10 P16 c
= (40365Ts4 ωc1 )/22
g
P2 = P17 = −(40365Ts4 ωc1 )/2
10 c
g
P3 = 8 P11 c
= (855Ts4 ωc2 )/2
g
P4 = P12 = −(855Ts4 ωc2 )/2
8 c
g (20.20)
P5 = 6 P7c = 80Ts4 ωc3
g
P6 = 6 P8c = −80Ts4 ωc3
g
P7 = 4 P3c = (127Ts4 ωc4 )/14
g
P8 = 4 P4c = −(127Ts4 ωc4 )/14
where all the angular velocities can be related to the output angular velocity. The
ratios of ωc /ωs4 are 308/729675, 56/29187, 14/1269 and 14/141 for ωc1 to ωc4 ,
respectively.
The power loss is related to the genuine meshing power. Assuming linear rela-
tionship, the total power loss is given by
10
g
L= μi Pi (20.21)
i=1
where μi are the loss factors at individual gear meshes. Note that the estimation
defined in Eq. (20.21) is larger than the actual loss, and is valid when all the loss
factors are small [1].
The efficiency of the EGT is defined as
η = 1 − L/Pi (20.22)
eb
where the total input power Pi to the EGT is 1 P10 , whose expression is listed in
Table 20.3. Substituting Eqs. (20.20) and (20.21) into (20.22) yields
η = 1 − 0.78(μ1 + μ2 ) − 0.8216(μ3 + μ4 ) − 0.8889(μ5 + μ6 ) − 0.9071(μ7 + μ8 )

(20.23)
For preliminary validation of the predicted efficiency all the loss factors in
Eq. (20.21) are assumed to be zero. The corresponding η is one, which agrees with
the zero loss assumption. Notice that EGT is self-locking when the efficiency is
negative. Assuming equal loss factors for all gear meshes, the upper threshold of the
loss factors corresponding to a positive η can be calculated, which is 0.147. As such,
it is essential to ensure that the gear loss is kept below this threshold in the design
process and during the operation of the EGT.
20.5 Conclusion
An analysis based on mechanical constraints is used to compute the normal and

virtual power of a four-stage EGT, as well as deriving the analytical solution for
prediction of efficiency. The outcomes of the power flow analysis are validated by
observing a power-balanced system in the power flow diagram. The efficiency of
the EGT is predicted by considering the load-dependent gear mesh loss factors.
The analytical solution is obtained by assuming all loss factors are linear. Further,
by assuming equal loss factors, their maximum threshold corresponding to positive
efficiency is calculated as 0.147. For future work, other loss factors will be taken into
account, along with the non-linear relation between the gear mesh loss factors and
the loads.
References
1. Chen, C.: Power analysis of epicyclic transmissions based on constraints. J. Mech. Robot. 4(4),
041004 (2012)
2. Kimura, Y., Oh, S., Hori, Y.: Realization of bi-articular driven robotic arm with planetary gear
based on disturbance observer. The University of Tokyo, Tokyo (2011)
262 N. Hu et al.
3. Tsai, S., Ye, S., Yu, Y., Tseng, J.: Design and analysis of the planetary gear drive with flexible
pins for wind turbines. In: Presented at the EWEA Conference, Vienna (2012)
4. Pennestrì, E., Freudenstein, F.: The mechanical efficiency of epicyclic gear trains. J. Mech.
Des. 115(3), 645 (1993)
5. Radzimovsky, E.I.: A simplified approach for determining power losses and efficiency of plan-
etary gear drives. Mach. Des. 28(3), 101–110 (1956)
6. Michaelis, K., Hohn, B.-R., Hinterstoier, M.: Influence factors on gearbox power loss. Ind.
Lubr. Tribol. 63(1), 46–55 (2011)
7. Radzimovsky, E.I.: How to FIND efficiency, speed, and power in planetary gear drives. Mach.
Des. 31(12), 144–153 (1959)
8. Kahraman, A., Ligata, H., Kienzle, K., Zini, D.M.: A kinematics and power flow analysis
methodology for automatic transmission planetary gear trains. J. Mech. Des. 126(6), 1071
(2004)
9. Salgado, D.R., Castillo, J.M.D.: Selection and design of planetary gear trains based on power
flow maps. J. Mech. Des. 127(1), 120 (2005)
10. Chen, C., Angeles, J.: Virtual-power flow and mechanical gear-mesh power losses of epicyclic
gear trains. J. Mech. Des. 129(1), 107–113 (2007)
11. Chen, C., Liang, T.T.: Theoretic study of efficiency of two-DOFs of epicyclic gear transmission
via virtual power. J. Mech. Des. 133(3), 031007 (2011)
12. Chen, C.: Power flow analysis of compound epicyclic gear transmission: Simpson gear train.
J. Mech. Des. 133(9), 094502 (2011)
13. Fernandes, C.M., Marques, P.M., Martins, R.C., Seabra, J.H.: Gearbox power loss. Part II:
Friction losses in gears. Tribol. Int. 88, 309–316 (2015)
14. Fernandes, C.M., Marques, P.M., Martins, R.C., Seabra, J.H.: Gearbox power loss. Tribol. Int.
88, 317–326 (2015)
15. Fernandes, C.M., Hammami, M., Martins, R.C., Seabra, J.H.: Power loss prediction: application
to a 2.5 MW wind turbine gearbox. Proc. Inst. Mech. Eng. Part J J. Eng. Tribol. 230(8), 983–995
(2015)
Chapter 21
Influences of Angular Position
of Unbalanced Force on Crack Breathing
Mechanism
H. M. Mobarak, Helen Wu and Chunhui Yang
Abstract In this study, a new mathematical model is developed to study crack

breathing behaviours considering effect of angular position of unbalanced force at
different crack locations. Crack breathing behaviours are determined by using effec-
tual bending angle and studying the transient change of the crack area. Different
crack breathing behaviours of the unbalanced shaft are observed for different com-
binations of angular position of unbalanced force with crack location, except for two
crack locations at 0.3 L and 0.8335 L, in which L is the total length of the shaft,
where unbalanced shaft behaves completely like a balanced shaft. Based on differ-
ent combinations of angular position of unbalanced force with crack location, the
stiffness of unbalanced shaft varies accordingly. An unbalanced shaft is stiffer than a
balanced shaft when angular position of unbalance force is between 90° to 270° and
crack located between 0.3 L and 0.8335 L and it is flexible when the crack is located
outside of this crack region. On the other hand, it is flexible when angular position
of unbalanced force is between 0° and 90° or 270° to 360° and the crack is located
in the middle region and it is stiffer when the crack is located outside of this crack
region.
Keywords Cracked shaft · Crack location · Shaft stiffness · Unbalanced force ·

Unbalanced force orientation
21.1 Introduction
Breathing of the fatigue cracks is one of the fundamental rotating machinery faults. It
attracted a lot of research attention in literature as one fundamental cause of dangerous
damages [1]. From the most recent decade, a wide variety of analytical and practical
H. M. Mobarak · H. Wu · C. Yang (B)

School of Computing, Engineering and Mathematics, Western Sydney University,
Penrith, NSW 2751, Australia
H. M. Mobarak

264 H. M. Mobarak et al.
models were developed and used for the analysis of fatigue cracks [2–4]. At early
stage of a cracked shaft analysis, gaping crack models was developed where the crack
is considered always completely open, which does not express to the real breathing
of a fatigue crack [5, 6]. Renovations on the non-linear nature of crack breathing
can be determined by using a switching crack model [7, 8]. This crack model was
established based on that the crack is considered either completely open or com-
pletely closed. Further, this crack model was linked with chaotic and quasiperiodic
vibrations, which actually were not found in experimental testing. Recently, more
realistic trigonometric functions were utilised to depict the crack breathing [9–11].
Almost all studies on cracked shaft were conducted only considering static load-
ing, which disallows demonstrating of crack breathing under unbalance force. Only
a few studies analysed the cracked shaft considering the effect of unbalanced force.
Bachschmid et al. [12] studied the non-linear crack breathing by significantly reduc-
ing the damping of the cracked rotor system as to amplify the impact of the breathing
of the crack. Cheng et al. [10] found that unbalance can re-establish the stability of a
rotor system. Rubio et al. [13] analysed impacts of unbalance force and angular posi-
tion of unbalanced force with respect to the crack direction on the crack breathing of
a Jeffcott rotor. In presence of the unbalance force, it was found the crack breathing
behaviours greatly differ from the static loading crack breathing.
However, to detect and diagnose fatigue cracks hinges on how well the crack
models emulate the actual cracked shaft system. The key aspect of any crack mod-
els is the reduction in stiffness introduced by the crack. The localised reduction in
stiffness is directly related to crack depth, whereas the global reduction in stiffness
is influenced by both crack depth and crack location. Unfortunately, all researchers
opted to either ignore crack location or mitigate its effects. Moreover, all existing
models are not applicable to analyse crack breathing at near the shaft critical speed
since equations of motion developed without considering effect of crack location and
unbalanced force.
There is a need to develop an accurate mathematical model for analysing the
actual crack breathing considering under unbalanced force at different crack loca-
tions. In this research, firstly a new mathematical model is developed to study the
effect of crack location on the breathing behaviours of the fatigue crack under unbal-
anced force with various angular positions. Secondly the newly-developed model
is employed to investigate the relative angle between the crack and shaft bending
directions to find the breathing behaviours of the crack at different crack locations.
Finally, the influences of angular position of unbalanced force on crack breathing
mechanism are analysed in terms of percentage of opening of a crack and examining
the duration of all three crack states including fully open, fully closed and partially
open/closed. Results are also compared with those obtained by using the balanced
model.
21 Influences of Angular Position of Unbalanced Force … 265
21.2 New Crack Breathing Model Under Dynamic Loading
To evaluate the crack breathing behaviours of the shaft in terms of crack status and
percentage of opening of crack at various angular positions under unbalanced force
at different crack locations, a shaft system is depicted in Fig. 21.1 which is a rigidly
supported two-disc rotor. This model represents a two-disc steel rotor supported
rigidly by two bearings in real world. To determine the crack breathing behaviours,
an effectual bending angle, ϕ, was introduced. The breathing of a fatigue crack is
governed by the effectual bending angle regardless of the type of loading (static or
dynamic). It is the proximity of the direction of the bending on the system relative to
the crack direction. Main parameters of this proposed model are given in Table 21.1.
The model was assumed to have a straight front oriented crack on a plane normal to
the axis of the shaft as shown in Fig. 21.2, in which R is the shaft radius and h is the
crack depth in the radial direction, respectively.
The shaft weight, ms g, was considered as a uniformly distributed static load,
where ms is the mass of the shaft. The discs’ weight, md g, was considered as a point
Fig. 21.1 Rigidly supported two-disc rotor
Table 21.1 Main parameters Shaft length, L 724 mm

of the analytical model for the
steel shaft Radius, R 6.35 mm
Density, ρ 7800 kg/m3
Poisson ratio, ν 0.3
Young’s modulus, E E = 210 GPa
Fig. 21.2 Relative position

of the unbalanced force with
respect to the crack
Fig. 21.3 Schematic

representation of
the effectual bending angle
load where md is the mass of each disk. The unbalance mass, mu , was treated as an
additional mass at a radial distance d from the center of the shaft and located on the
right disk. Due to the unbalance mass, the dynamic unbalance load F un = mu ω2 d was
considered, where ω is the shaft rotation speed (see Fig. 21.2). Also, the dynamic
load, F un , was considered act in the radial direction (θ + β), where θ is the shaft
rotation angle at time t and β is angular position of the unbalance force relative to
the crack direction.
All static loads were considered always acing downward along the negative Y-
axis. Therefore, the moments due to shaft weight load, M mg1 , and two disks weight
loads, M mg2 , were act along the negative X-axis, perpendicular to the loads direction.
And in the same way, the unbalance moment due to the dynamic load, M un , acted
perpendicular to the F un (see Fig. 21.3). In the present work, shaft rotation was
considered anticlockwise. All moments are described in the following equations for
different crack regions, where l0 is the crack location:
ms g
Mmg1 = (6Ll0 − L 2 − 6l02 ) when 0 ≤ l0 ≤ L (21.1)
12L
⎫
Mmg2 = m d gl0 − 3m16 d gL
⎬ 0 ≤ l0 ≤ L4
Mmg2 = m d gl1 − 3m16 d gL
when L4 < l0 < 3L (21.2)
3m d gL ⎭
4
Mmg2 = m d g(L − l0 ) − 16 3L
4
≤ l0 ≤ L

Mun = 5F32
un l0
− 3F64un L 0 ≤ l ≤ 3L
when 3L 0 4 (21.3)
Mun = 32 (L − l0 ) − 64
27Fun 9Fun L
4
≤ l0 ≤ L
The total moment for different crack regions is obtained according to the principle
of superposition theory:

In X-axis : M x = Mmg1 + Mmg2 + Mun cos(θ + β) (21.4)

In Y-axis : M y = Mun sin(θ + β) (21.5)
The dynamic unbalance load, F un , is not always located at the crack plane. In
Fig. 21.3 only exemplifies a projection of dynamic load on the crack plane. As we
know, the shaft will bend along the resultant load direction, perpendicular to the
resultant moment direction as shown in Fig. 21.3. The effectual bending angle, ϕ,
is described in Eq. (21.7), where δ is the angle between the shaft bending direction
with negative Y-axis. Modifications were made to ensure that δ and ϕ are within the
co-domain of full rotation of shaft values between 0 to 2π .

My
δ = tan−1 ( ) (21.6)
Mx
ϕ=θ−δ (21.7)
As shown in Fig. 21.4a, the crack starts to close when the upper end of the crack
edge reaches the compression stress field. It will happen when the effectual bending
angle, ϕ, will become same as ϕ 1 . The ϕ 1 is given in Eq. 21.8, where μ = h/R is
non-dimensional crack depth ratio and A1 is uncracked area and e is the location of
the centroid. Similarly, as shown in Fig. 21.4b, the crack becomes fully closed when
the crack is fully reached in the compression stress field. And it will happen when
the effectual bending angle, ϕ, will become same as ϕ 2 , which is given in Eq. 21.9.
The e and A1 can be calculated by Eqs. (21.10) and (21.11). These two crucial angles
were derived previously for the balanced shaft [14] to describe the crack breathing
in terms of shaft rotational angle.
e + R(1 − μ)
ϕ1 = tan−1 ( √ ) (21.8)
R μ(2 − μ)
π
ϕ2 = + cos−1 (1 − μ) (21.9)
2
2R 3 3
e= [μ(2 − μ)] 2 (21.10)
3A1

A1 = R 2 [π − cos−1 (1 − μ) − (1 − μ) μ(2 − μ)] (21.11)
The values of effectual bending angles at different crack locations are used to
identify the statuses of the crack during shaft rotation. The full statuses of the crack
breathing relative to the regions formed by ϕ 1 and ϕ 2 are shown in Fig. 21.5. The
transient changes in closed area of the crack are determined to study the percentages
V
of opening of cracks, , which is described in Eq. 21.12. As shown in Fig. 21.1,
Acrack is the area of the crack and Aclosed (t) is the closed portion of the crack segment
when time, t > 0 (see Fig. 21.6). Equation 21.13 was used to calculate the Acrack and
V
an iterative procedure seen in [14] was used to determine the Aclosed (t). is equal to
100 and 0 for a fully open crack or fully closed crack respectively.
Fig. 21.4 Crucial angles for

the unbalanced shaft: a ϕ 1
crack begins to close and
b ϕ 2 crack become fully
closed
Fig. 21.5 Statuses of the

crack breathing in the range
of the effectual bending
angle from 0° to 2π
Acrack − Aclosed (t)

(%) = × 100 (21.12)
Acrack

Acrack = R 2 [cos−1 (1 − μ) − (1 − μ) μ(2 − μ)] (21.13)
Fig. 21.6 Schematic

diagram of the closed portion
of a breathing crack
The crack breathing behaviours were analysed for first half of the unbalance force
orientation, such as β = 0°, 45°, 90°, 135° and 180° to know the effect of angular
position of unbalance force. Only first half of the unbalance force orientation was
considered due to first half symmetric about the second half. The analysis also con-
sidered different crack location factors, λ, the ratio of the crack position and shaft
total length, L. It was considered to determine the influences of crack location on
the crack breathing behaviour. MATLAB was used to analyse the crack breathing
behaviour. In this study, a crack with a depth ratio of μ = 0.5 was considered with
initial crack direction along the negative Y-axis. Different force ratios (the ratio of
the static force and the dynamic force) were considered to perform the analysis.
The evolution of the balanced and unbalanced shafts effectual bending angles are
shown in Fig. 21.7 along the shaft length. The effectual bending angle jumps from
0° to 360° at λ = 0.8053 in an unbalanced shaft is a result of the crack direction
changing from leading to following the bending direction. However, it is the same as
the balanced shaft at λ = 0.3 and 0.8335, it is remarkably different from the balanced
shaft along the shaft length. However, if the crack is located at λ = 0.3 and 0.8335, the
unbalanced shaft cracks will breathe as they would be in a balanced shaft. Thus these
two crack locations are independent of the unbalanced force.
Regardless of loading conditions, the effectual bending angles were found as a
function of crack location. Consequently, for different crack locations, breathing of a
fatigue crack would be different. Thus, to solve the equation of motions to investigate
the vibration behaviours of the cracked rotor, the effects of crack location cannot be
ignored since the stiffness of the shaft and the crack breathing are directly related
each other. The shaft will have a maximum stiffness when the crack will become
Fig. 21.7 Effectual bending

angles as a function of crack
location for balanced and
unbalanced cases where θ =
135° and β = 0°
Fig. 21.8 Effects of angular

position of unbalance force
on the crack breathing
behaviours at the crack
location 0.5 L
fully closed and it will virtually become as an uncracked shaft. Similarly, the shaft
will have a minimum stiffness when the crack will become fully opened. The shaft
will have its stiffness between the maximum and minimum stiffness when the crack
will become partially open or closed.
V
In Figs. 21.8 and 21.10, using percentage of opening of the crack, , crack breath-
ing behaviours are displayed quantitatively. The crack breathing behaviours at crack
location λ = 0.5 are shown in Fig. 21.8. The percentage of opening of the crack when
angular position of unbalanced force β = 135° and 180° is lower than that from the
balanced shaft, which also indicates that the unbalanced shaft is stiffer and achieves
the stiffest value when β = 180°. In addition, the percentage of opening of the crack
is larger than balanced shaft when β = 0° and 45°. Therefore, the unbalanced shaft
is more flexible when β = 0°. However, for the unbalanced shaft, the percentage
of opening of the crack is sometimes larger and sometimes is smaller than that of
a balanced shaft when β = 90°. Therefore, the overall stiffness of the unbalanced
shaft is similar to that of the balanced shaft.
The results demonstrate that, when the unbalance force is located between 90°
and 270°, the unbalanced shaft is overall stiffer than the balanced shaft. When β
values between 0° and 90° or 270° and 360°, the unbalanced shaft is overall more
flexible. It is noted that the angular position of the unbalanced force strongly affects
the crack breathing (see also Fig. 21.9).
For crack locations λ = 0.125 and 0.875, the crack breathing is shown in
Fig. 21.10. In both crack locations, the percentage of opening of the crack is lower
from the balanced shaft when β = 0° & 45°. Which indicates that the unbalanced
shaft is stiffer than the balanced shaft and stiffest when β = 0°. However, It is totally
opposite when β = 135° & 180°. Therefore, the stiffness of the unbalanced shaft
would be lower than the balanced shaft stiffness. However, the unbalanced shaft stiff-
ness would be sometimes larger and sometimes is smaller than of a balanced shaft
when β = 90°. Overall, when the crack located at λ = 0.125 and 0.875, compare to
the balanced shaft stiffness the unbalanced shaft is stiffer for 0° ≤ β < 90° and 270°
< β ≤ 360° and flexible for 90° < β < 270°, which is opposite to the crack location
with λ = 0.5 (see Figs. 21.7, 21.8 and 21.9).
Along the shaft length variation of shaft stiffness can be divided into three regions
at λ = 0.3 and 0.8335, respectively. When β is between 90° and 270° and the crack
is located in two side regions, the unbalanced shaft is flexible than the balanced shaft
and when the crack located in middle region, the unbalanced shaft is stiffer. On the
other hand, for two side regions and when β is between 0° and 90° or 270° 360°,
the unbalanced shaft is stiffer and for middle region is flexible (see Figs. 21.8 and
21.10). The angular position of the unbalance force plays an important role in the
peak amplitude of the vibration as observed by Cheng et al. [10], where the minimum
and maximum vibration amplitudes correspond to the eccentric mass being located
at and opposite the crack, respectively.
21.4 Conclusions
In order to analyse the impacts of angular position of the unbalanced force on crack
breathing at different crack locations, a series of analyses have been completed by
using MATLAB via considering five different angular positions of the unbalanced
force (β = 0°, 45°, 90°, 135° and 180°). The effectual bending angle has been
introduced to evaluate the breathing behaviour for different combination of angular
position of unbalanced force with crack location and it is further used to determine
the transient change in crack area to analyse the percentages of the opening of crack.
The effectual bending angle is found as a function of crack location. It is remarkably
different for unbalanced shaft compared to the balanced shaft for different crack
locations except for 0.3 L and 0.8335 L. Notably different effects of angular position
of the unbalanced force on crack breathing behaviours were identified for different
crack locations. At crack locations 0.3 L and 0.8335 L, the stiffness of an unbalanced
shaft is same as that of the balanced shaft. For other crack locations, unbalanced shaft
stiffness is between minimum and maximum values. Compared to balanced shaft
stiffness, unbalanced shaft stiffness variation with crack location can be identified at
three regions at crack locations 0.3 L and 0.8335 L, according to different angular
Effect of angular position of unbalanced force depend on

crack location
Balanced Balanced
Effect of angular position of unbalanced force with respect to the crack direction on the status of crack
β = 0° β = 0°
β = 45° β = 45°
β = 90° β = 90°
β = 135° β = 135°
β = 180° β = 180°
Fig. 21.9 Effects of the angular position of the unbalance force on the crack breathing behaviours
at crack locations 0.5 L (Left) and 0.125 L (Right), respectively (closed in red and open part in
yellow)
Fig. 21.10 Effects of

angular position of
unbalance force on the crack
breathing behaviours at
crack locations a 0.125 L
and b 0.875 L
positions of the unbalanced force. It is stiffer when the crack is located between 0.3 L
and 0.8335 L and angular position of unbalanced force β is between 90° < β < 270°
but it is flexible when the crack is located outside of this crack region. On the other
hand, it is also flexible in the middle crack region when β is between 0° ≤ β < 90°
or 270° < β ≤ 360° and stiffer for two side crack regions.
The newly-developed mathematical model in this study can be further developed
to investigate vibration behaviours of the cracked rotor system by obtaining the
time-varying stiffness matrix of the cracked shaft and then solving the equation of
motions.
Acknowledgements The authors would like to gratefully acknowledge the financial support given
by the School of Computing, Engineering and Mathematics, Western Sydney University, Australia
on this research.
References
1. Georgantzinos, S.K., Anifantis, N.K.: An insight into the breathing mechanism of a crack in a
rotating shaft. J. Sound Vib. 318, 279–295 (2008)
2. Ming, L., Fucai, L., Beibei, J., et al.: Multi-fault diagnosis of rotor system based on
differential-based empirical mode decomposition. J. Vib. Control. (2013). https://doi.org/10.
7754/6313502505
3. Kulesza, Z.: Dynamic behavior of cracked rotor subjected to multisine excitation. J. Sound
Vib. 333(5), 1369–1378 (2014)
4. Yan, G., De Stefano, A., Matta, E., et al.: A novel approach to detecting breathing-fatigue
cracks based on dynamic characteristics. J. Sound Vib. 332(2), 407–422 (2013)
5. Dimarogonas, A., Papadopoulos, C.: Vibration of cracked shafts in bending. J. Sound Vib.
91(4), 583–593 (1983)
6. Papadopoulos, C.A., Dimarogonas, A.D.: Coupled longitudinal and bending vibrations of a
rotating shaft with an open crack. J. Sound Vib. 117(1), 81–93 (1987)
7. Sekhar, A.S.: Crack detection through wavelet transform for a run-up rotor. J. Sound Vib.
259(2), 461–472 (2003)
8. Sinou, J.J.: Effects of a crack on the stability of a non-linear rotor system. Int. J. Non-Linear
Mech. 42(7), 959–972 (2007)
9. Sinou, J.J.: Detection of cracks in rotor based on the 2 × and 3 × super-harmonic frequency
components and the crack–unbalance interactions. Commun. Nonlinear Sci. Numer. Simul.
13(9), 2024–2040 (2008)
10. Cheng, L., Li, N., Chen, X.F., et al.: The influence of crack breathing and imbalance orientation
angle on the characteristics of the critical speed of a cracked rotor. J. Sound Vib. 330(9),
2031–2048 (2011)
11. Mobarak, H.M., Wu, H., Spagnol, J.P., et al.: New crack breathing mechanism under the
influence of unbalance force. Arch. Appl. Mech. (2017)
12. Bachschmid, N., Pennacchi, P., Tanzi, E.: Cracked Rotors: A Survey on Static and Dynamic
Behaviour Including Modelling and Diagnosis. Springer, Heidelberg (2010)
13. Rubio, L., Muñoz-Abella, B., Rubio, P., et al.: Quasi-static numerical study of the breath-ing
mechanism of an elliptical crack in an unbalanced rotating shaft. Lat. Am. J. Solids Struct 11,
2333–2350 (2014)
14. Al-Shudeifat, M.A., Butcher, E.A.: New breathing functions for the transverse breathing crack
of the cracked rotor system: approach for critical and subcritical harmonic analysis. J. Sound
Vib. 330(3), 526–544 (2011)
Chapter 22
Effects of Elliptical Crack Shape Ratio
on Transverse Trajectory of a Cracked
Shaft
Joseph Spagnol, Helen Wu and Chunhui Yang
Abstract Fatigue striations and beach marks seen on the cross-section of shafts
subject to fatigue-induced failure reveal that the crack-front transitions from being
highly rounded (elliptical/circular) to almost straight as the crack propagates. This
study examines the transverse displacement of cracked rotors with varying ellipti-
cal crack shape ratios and depths, however, in particular, when the crack depth is
increased, the elliptical crack shape ratio is also decreased. Furthermore, it is shown
that straight-front cracks and elliptical-cracks models result in significantly different
transverse trajectories at one-third and one-half of the critical speed. The findings
of this study suggest that failure to account for the highly rounded nature of shal-
low cracks may result in large inaccuracies when predicting vibration behaviours of
cracked rotors.
Keywords Cracked rotor · Elliptical crack · Fatigue · Jeffcott rotor · Machine

health monitoring · Vibration
22.1 Introduction
When a shaft is subject to a repeated bending load for tens of thousands to millions of
cycles transverse fatigue cracks will form and propagate. Insufficient or ineffective
maintenance of rotors may result in fatigue cracks going undetected and may lead
to failure of the machine. Fortunately, the flexure of a cracked shaft results in a phe-
nomenon known as crack breathing occurs, which is the gradual opening and closing
of the crack. Because crack breathing results in the periodic change of stiffness of
a cracked rotor, defining characteristics may appear in the vibration response of the
J. Spagnol · H. Wu · C. Yang (B)

2751, Australia
J. Spagnol

276 J. Spagnol et al.
cracked rotor. When these characteristics appear, the shaft is repaired or replaced to
prevent destructive vibration from ever occurring.
Shaft cracks can be identified by conspicuous changes to the normal vibration of a
machine. Bachschimd et al. [1] stated that monitoring of the 2X vibration component
is the most effective candidate for crack detection in rotating machinery. It is also
stated that the 1X component can be masked by permanent shaft bowing and the
3X component, which may be indicative of a breathing crack, is typically small and
can be masked by irregularities. Also, fatigue cracks in shafts may also be identified
through the sub-resonance peaks at one-third and one-half of the critical speed of the
cracked rotor during a run-down of the rotor [2–4].
When a shaft is ruptured by fatigue-induced failure, the ruptured cross-section
contains microscopic markings that propagate from the origin of fracture, known
as fatigue striations. These markings are approximately semi-elliptical in shape and
so literatures on cracked shafts sometimes produce models with elliptical cracks,
however many more studies simplified models by using a straight-front crack. Rubio
et al. [5] examined breathing behaviours of elliptical cracks under the influence of
mass unbalance using FEA. The breathing behaviours were seen to differ from the
patterns seen in weight dominant breathing models. In [6, 7], the stress intensity at the
crack front was calculated to study the breathing behaviour of elliptical cracks with
varying depths and shapes. In particular, Han and Chu [7] examined the stability and
steady-state responses of a weight-dominant shaft with an elliptical crack and found
that the steady-state response and instability regions are reduced with increasing
crack shapes.
Typically, studies disassociate crack depth with elliptical shape ratio despite visual
evidence for the two properties being linked, as seen by beach marks and fatigue stri-
ations. Therefore, the goal of this study is to model cracked rotor vibration with semi-
elliptical cracks by decreasing the elliptical shape ratio as the crack depth is increased.
22.2 Crack Breathing Model
Al-Shudeifat and Butcher [8] presented a novel method for developing time-varying
area moments of inertia of the crack based on the geometry of a straight-front crack.
A similar mathematical approach is utilised in [9] to model the time-varying area
moments of inertia of a semi-elliptical crack. In this study, we utilised the breathing
model presented by Wei et al. [9] to determine the transverse displacements of a
cracked Jeffcott rotor.
Consider the transverse cross-section at the location of a fatigue crack in a shaft,
shown in Fig. 22.1. The shaded region is representative of a semi-elliptical crack
that is mathematically represented as the intersection area of an ellipse and circle
(circumference of the shaft). The elliptical crack shape ratio β is the ratio of the
lengths of the semi-minor axis to the semi-major axis of the ellipse, i.e. β = b/a. The
normalised depth of the semi-elliptical can be described as μ = b/R.
22 Effects of Elliptical Crack Shape Ratio on Transverse … 277
Fig. 22.1 Geometry of a semi-elliptical crack
Fig. 22.2 Visual representation of elliptical crack shape ratios at varying depths
In this study, we examined a number of elliptical crack shape ratios, some of

which are visualised in Fig. 22.2. A β value of zero corresponds to a crack with a
straight-front and an increase in the value of β causes the crack front to become more
elliptical/circular, where β = 1 is representing a circular-front.
When the shaft rotates and whirls, the fatigue crack will gradually open and close
based on whether the stresses acting on the crack faces are compressive or tensile.
Figure 22.3 shows the behaviours of the elliptical crack in a weight dominant system
(dynamic loading is considered being negligible relative to the static loading) in the
first half of the shaft rotation. The crack begins the rotation completely open until
some angle θ 1 , and then the crack gradually closes until angle θ 2 , where the crack
is then completely closed. In the second half of the shaft rotation, the crack remains
closed until 2π − θ 2 and then gradually opens until it is fully open at 2π − θ 1 .
It should be noted that the θ = 0° when the crack direction is coincident with the
negative Y axis, as in the first diagram of Fig. 22.3.
Mayes and Davies [10] demonstrated that change in stiffness of a rotating shaft
due to a transverse crack can be modelled by a periodic change in the area moment
of inertia of the shaft cross-section. It should be noted that the area moment of inertia
is taken about a horizontal and vertical axis originating at the centroid of the total
uncracked region, where the total uncracked region is defined as the uncracked area
plus the area of the partially open crack.
Fig. 22.3 Elliptical crack stages showing crack behaviour and major milestones
The area moment of inertia of the transverse cross-section is considered to vary

based on the status of the crack, i.e. the crack is fully open, partially open or fully
closed. When the crack is fully closed, the area moment of inertia is equal is maxi-
mum, which is equal to that of a non-cracked shaft cross-section, I = π r 4 /4. Further-
more, the area moment of inertia will vary depending on the openness of the crack
and its relative position to the horizontal and vertical centroid axes.
To calculate the time-varying area moment of inertia of a semi-elliptical crack,
the procedure presented in [9] was used. However, the method described in [9]
will not be included in this study due to the length of the rigorous mathematical
approach. Instead, we provide the periodic area moment of inertia and product of
inertia obtained using the method, as this is the ultimate objective for using the
approach (Figs. 22.4, 22.5 and 22.6). The area moment of inertia in the horizontal
and vertical centroid directions are denoted as I X and IY , and the product of inertia
is denoted as I X Y . We implore readers to read the study for more information on
calculate the area moments of inertia.
Figures 22.4, 22.5 and 22.6 show the differences in the periodic area moment of
inertia and product of inertia for the selected elliptical crack shape ratios and crack
depths. For any given crack depth μ, it is clear that the average area moment of inertia
of the total non-cracked area about the horizontal and vertical centroid directions
increases with an increase in β. This is strongly suggested by visual inspection of
Fig. 22.2, where an increase in β results in a decreased cracked area and therefore
higher average area moment of inertia of the total uncracked area.
Fig. 22.4 Periodic change in the area moment of inertia about the horizontal centroid axis: a μ =
0.25; b μ = 0.5 and c μ = 0.75
Fig. 22.5 Periodic change in the area moment of inertia about the vertical centroid axis: a μ =
0.25; b μ = 0.5 and c μ = 0.75
Fig. 22.6 Periodic change in the product of inertia about the vertical and horizontal centroid axes:
a μ = 0.25; b μ = 0.5 and c μ = 0.75
22.3 Rotor Model
A four degree-of-freedom Jeffcott rotor was used in this study, where the degrees-
of-freedom relate to the transverse translations (x and y) and rotations (θ x and θ y )
about the X and Y axes, as seen in Fig. 22.7. The rotor was considered to be simply
supported at each side and the shaft is considered massless.
The equations of motion associated with the cracked rotor system can be described
in state-space form as a system of first order differential equations
Fig. 22.7 Jeffcott rotor with

crack at mid-span

Q̇ = [A]{Q} + {F} (22.1)
where the state variable, Q, is the vector of transverse translation and rotations {x, y,
θ x and θ y }. The matrix [A] is equal to
⎡ ⎤
0 0 m 0
1⎢ 0 0 0 m ⎥
[A] = ⎢ ⎥ (22.2)
m ⎣ −k x x (t) −k x y (t) −c 0 ⎦
−k yx (t) −k yy (t) 0 −c
where k xx (t) and k yy (t) is the instantaneous time-varying stiffness’s of the shaft in the x
and y directions, respectively. The parameters k xy (t) and k yx (t) are the cross-coupling
stiffness’s of the shaft, m is the mass of the disk and c is the external damping. The
stiffness values are related to area moment of inertia values using Eq. (22.3). The
parameters E and L are the modulus of elasticity and length of the shaft, respectively.
The values for IX and IY for the chosen elliptical crack shape ratio and crack depths
are shown in Sect. 22.2 of this study.

k x x (t) k x y (t) 48E I X (t) I X Y (t)
= 3 (22.3)
k yx (t) k yy (t) L IY X (t) I X (t)
In Eq. (22.1), the parameter F is the vector of loads comprising of unbalance of

the disk and the disk weight, which is written as
⎧ ⎫
⎪
⎪ 0 ⎪
⎪
1⎨ 0
⎬
{F} = (22.4)
m⎪
⎪ m ed 2 sin(t + δ) ⎪⎪
⎩ ⎭
m ed 2 cos(t + δ) − mg
where med is the mass unbalance, Ω is the rotor spin speed, δ is the mass unbalance
angle and g is the gravitational acceleration constant.
Table 22.1 Physical Parameter Value

parameters of the cracked
rotor model Length of rotor, L 0.7239 m
Radius of rotor, R 7.94 mm
Mass unbalance, med 1.125 × 10−5 kg•m
Modulus of elasticity, E 200 GPa
Mass of disk 1.126 kg
External damping 6.11 N · s/m
The physical parameters of the rotor model are given in Table 22.1. These param-
eters are based on the Spectra Quest MFS-RDS rotor dynamic simulator.
The dynamic response of the rotor was obtained using the MATLAB ode45 function
which requires the equations of motion to be written in state-space form, as shown
in Eq. (22.1). The period of numerical integration was chosen to be 20 s as it was
beyond sufficient for reaching the steady-state solution. Because ode45 is a variable
step-solver, the time-steps automatically change throughout the solution depending
on the whether the state is changing slowly or rapidly. The generated time increments
are typically within the range of 10−5 to 10−9 s.
Figure 22.8 shows the radial displacement of the disk during a run-down of the
cracked rotor for a straight-front crack with a normalised depth of 0.2. The radial
displacement is calculated by r = x 2 + y 2 . These particular parameters were
chosen to avoid instability of the numerical solution and to additionally gauge the
vicinity of the critical speed of the cracked rotor. Increasing the crack depth results
in instabilities, therefore a shallow crack was chosen.
From Fig. 22.8, the critical speed occurs at 2575 RPM which is denoted by the
peak amplitude. Sub-resonance peaks are seen in the vicinity of one-third and one-
half of the critical speed. As previously mentioned, these sub-resonance peaks may
be indicative of fatigue cracks in real rotors.
Fig. 22.8 Radial displacement of disk during run-down of the cracked rotor for μ = 0.2 and β = 0
Fig. 22.9 Transverse trajectory of disk in the vicinity of one-third critical speed (850 RPM) and
unbalance angle of δ = 0°. Comparison of straight-front and semi-elliptical front cracks with a μ
= 0.25, β = 0.8, b μ = 0.5, β = 0.5 and c μ = 0.75, β = 0.3
Fig. 22.10 Transverse trajectory of disk in the vicinity of one-third critical speed (850 RPM) and
= 0.25, β = 0.8, b μ = 0.5, β = 0.5 and c μ = 0.75, β = 0.3
Examination of the transverse trajectory (x-displacement versus y-displacement)

in the vicinity of the one-half and one-third of the critical speed is seen in Figs. 22.9,
22.10, 22.11, 22.12. As seen in [11, 12] trajectories around one-third of the critical
speed result in a characteristic double inner loops related to a significant 3X harmonic
component. Additionally, trajectories around one-half of the critical speed result in
a characteristic single inner loop related to a significant 2X harmonic component. In
this study, these characteristics are present, with double inner loops in Figs. 22.9 and
22.10 and single inner loops in Figs. 22.11 and 22.12; these findings suggest that the
results are plausible.
For each figure the same set of elliptical crack shape ratios, crack depths and
unbalance angles are chosen. For the semi-elliptical cracks, the combinations of β
and μ shown in the Figs were chosen such that the deeper the crack the smaller the
elliptical crack shape ratio. In particular, for crack depth of μ = 0.25, μ = 0.5 and
μ = 0.75 corresponded to elliptical crack shape ratios of β = 0.8, β = 0.5 and β =
0.3, respectively. Additionally, the values for δ were chosen arbitrarily to examine
the effect of unbalance angle on the transverse trajectory.
Fig. 22.11 Transverse trajectory of disk in the vicinity of one-half critical speed (1280 RPM) and
= 0.25, β = 0.8, b μ = 0.5, β = 0.5 and c μ = 0.75, β = 0.3
Fig. 22.12 Transverse trajectory of disk in the vicinity of one-half critical speed (1280 RPM) and
= 0.25, β = 0.8, b μ = 0.5, β = 0.5 and c μ = 0.75, β = 0.3
There is a clear difference in the predicted transverse trajectory between the mod-
els when using semi-elliptical-front and straight-front cracks. The unbalance angle
does not have any effects on making the trajectory of the two models more similar.
Due to this significant difference, it becomes clear that researchers should take care
when modelling vibration of cracked rotors as the crack shape ratio has a notable
effect on the predicted trajectory.
22.5 Conclusion
Ultimately, there is a clear, significant difference in the predicted transverse trajectory

of models using a straight-front crack and a semi-elliptical front crack for speeds in
the vicinity of one-half and one-third of the critical speed, regardless of the unbalance
angle. Typically, studies use straight-front cracks for even moderately shallow cracks
(μ < 0.5), however beach marks show that the crack front is highly elliptical for
small cracks. Therefore, not accounting for the elliptical nature of cracks may result
in inaccurate predictions of the transverse vibration of a cracked rotor.
Acknowledgements The funding provided by Western Sydney University for this research is
greatly appreciated.
References
1. Bachschmid, N., Pennacchi, P., Tanzi, E.: Cracked Rotors: A Survey on Static and Dynamic
Behaviour Including Modelling and Diagnosis. Springer, Heidelberg (2010)
2. Guo, C., Al-Shudeifat, M.A., Yan, J., et al.: Application of empirical mode decomposition to
a Jeffcott rotor with a breathing crack. J. Sound Vib. 332(16), 3881–3892 (2013)
3. Darpe, A.K., Gupta, K., Chawla, A.: Transient response and breathing behaviour of a cracked
Jeffcott rotor. J. Sound Vib. 272(1–2), 207–243 (2004)
4. Sinou, J.J., Lees, A.W.: A non-linear study of a cracked rotor. Eur. J. Mech. Solids 26(1),
152–170 (2007)
5. Rubio, L., Muñoz-Abella, B., Rubio, P., et al.: Quasi-static numerical study of the breathing
mechanism of an elliptical crack in an unbalanced rotating shaft. Lat. Am. J. Solids Struct.
11(13), 2333–2350 (2014)
6. Rubio, P., Rubio, L., Muñoz-Abella, B., et al.: Determination of the stress Intensity Factor of
an elliptical breathing crack in a rotating shaft. Int. J. Fatigue 77, 216–231 (2015)
7. Han, Q., Chu, F.: Dynamic instability and steady-state response of an elliptical cracked shaft.
Arch. Appl. Mech. 82(5), 709–722 (2012)
8. Al-Shudeifat, M.A., Butcher, E.A.: New breathing functions for the transverse breathing crack
of the cracked rotor system: Approach for critical and subcritical harmonic analysis. J. Sound
Vib. 330(3), 526–544 (2011)
9. Wei, X., Liu, D., Zhao, L., et al.: Time-varying stiffness analysis on rotating shaft with elliptical-
front crack. Int. J. Ind. Syst. Eng. 17(3), 302–314 (2014)
10. Mayes, I.W., Davies, W.G.R.: Analysis of the response of a multi-rotor-bearing system con-
taining a transverse crack in a rotor. J. Sound Vib. 106(1), 139–145 (1984)
11. Al-Shudeifat, M.A.: On the finite element modeling of the asymmetric cracked rotor. J. Sound
Vib. 332(11), 2795–2807 (2013)
12. Patel, T.H., Darpe, A.K.: Influence of crack breathing model on nonlinear dynamics of a cracked
rotor. J. Sound Vib. 311(3–5), 953–972 (2008)
Chapter 23
Seismic Performance Comparison
of Base-Isolated and Conventional RC
Frames Compliant with ASCE 7–10
Using FAHP Method
Longjun Xu, Yabin Chen and Qingyang Liu
Abstract This paper presents comprehensive comparison of base-isolated and fixed

reinforcement concrete frames (RCF) designed minimally compliant with ASCE
7–10. The investigation employs fuzzy analytic hierarchy process (FAHP) method
which estimates the performance grades based on accounting for the weighting
factors of each performance variable. Equivalent strength modification factors are
developed to design the isolated structures. Structures that consider degradations
and P-Delta effect are rationally modeled for nonlinear response history analysis
(NRHA) under multiple-intensity ground motions. The results show that the seismic
performance of code-designed isolated structures under three scenario earthquakes
is superior to that of the fixed-base structures. Moreover, the lower isolated structures
perform better than the higher ones with the same period. Finally, the collapse mode
of isolated structures may be of great difference from the conventional structures
due to the cumulative speed of energy dissipation besides the property of the ground
motion using incremental dynamic analysis (IDA).
Keywords Base-isolated · FAHP · Scenario earthquakes · Incremental dynamic

analysis
23.1 Introduction
The design philosophy of seismic isolation, to insulate the superstructure by installing

flexible devices at the base floor, is reasonable to offer superior performance under
design level earthquake by lengthening the vibration period to lower acceleration
spectrum. In that case, the isolated structure performs elastically to reduce or elimi-
nate the damage of the structural units relative to a fixed-base structure that suffers
plastic deformations. In addition, non-structural components (NSC), no less than
L. Xu · Y. Chen (B)
Department of Civil Engineering, Harbin Institute of Technology at Weihai, Weihai 264209, China
Q. Liu
Shandong Electric Power Engineering Consulting Institute Corp., LTD, Jinan 250013, China

286 L. Xu et al.
80% of the total building investment, would be better protected if excessive inter-
story drifts or floor accelerations were suppressed [1].
The implicit performance targets, embedded in the United States seismic codes
according to the study are higher for base-isolated structures than those of fixed-
base structures [2]. The maximum compliant reduction factor R, up to eight, which
limits the damage of newly-built conventional structures through significant inelastic
behavior and preserves a “Life safety” performance target under design earthquake
is permitted. In contrast, the isolated structures are limited to remain undamaged
within elastic response and, the corresponding performance target can be interpreted
as “Fully operational” [3]. Consequently, the minimally compliant design forces of
isolated structures are usually larger than the counterparts of conventional structures
[4]. While most of isolated structures are deemed as an expensive strategy for not
only their additional isolation system and testing, but also their over-strength beyond
minimum code compliance due to jurisdiction review or personal objectives.
According to provisions of SEAOC and UBC (1997), failure of isolation system
and excessive plasticity of isolated structure are prohibited for the coded-design
isolated structures under maximum credible earthquake [3, 5]. Intuitive reasoning
that superiority of isolated structures in design events also appears in other earthquake
scenarios should be critically examined. The detrimental effects of yielding isolated
structures have been presented by several researchers. Palazzo and Petti firstly derived
the minimum strength spectra for isolated structure survival through collapse spectra
[6]. Ceccoli et al. looked into the cause of structural damage through its hysteric
energy dissipation [7]. Benefits of isolation would be discounted if further reduction
of isolated structure through the seismic demands of yielding isolated structures.
Yielding acceleration demands of isolated structures are lower than those of the
corresponding fixed base structures. And it is easier to damage for the yielding
isolated structures than the fixed base structure [8]. Premature yielding and excessive
stiffness of isolated structure will lead to exorbitantly large displacement ductility
and may result in sidesway collapse compared with the corresponding fixed-base
structure [9].
Seismic comparison of base-isolated and fixed-base structures has been studied
by many researchers [4, 10–14]. Some of them have evaluated the performance
of both structures referenced with code provisions. Shenton and Lin compared the
fixed-base and base-isolated RCF structures according to SEAOC (1995) [3, 11], and
the isolated structures were designed in varying levels of base shear. However, the
comparison of seismic performance was performed without consideration of multiple
hazard intensities in addition to the evolvement of the building codes these years.
The structures can achieve the performance objectives through accommodating
the strength modification factor “R”. To address these objectives, the code ASCE
7–10 prescribes quantitative functions of the minimal strength for the new isolated
and conventional structures [15]. Although the maximum drift ratio of isolated struc-
tures has been provided, more detail limitations for different heights of isolated struc-
tures has not been given. Namely, isolated structures of different heights designed
with minimal strength may have different margins of safety against failure, especially
under multiple hazard intensities.
23 Seismic Performance Comparison of Base-Isolated … 287
Quantitative comparison of the seismic performance of isolated and fixed-base

RCF structures minimally compliant with ASCE 7–10 not only can provide a direct
evidence for the final decision-making but also can identify the best system to achieve
a specified performance target with multiple seismic response variables. However,
to our best knowledge, the quantitative method to compare the seismic performance
of isolated and fixed-base RCF structures has not been reported. The aim of this
paper is to introduce a quantitative comparison method using fuzzy analytic hierar-
chy process (FAHP) based on fuzzy synthetic evaluation (FSE) model firstly. The
FAHP method can identify the optimum system with multiple variables which are
weighted objectively by the FSE model using prior seismic response results consid-
ering multiple hazard intensities. Besides, an equivalent strength method which can
facilitate the design of isolated structure is proposed. Three suites of ground motions
are selected representing 2, 10 and 50% probabilities of exceedance within 50 years
and each suite of ground motions are rationally scaled to specified seismic hazards.
Finally, the IDA curves of minimally designed isolated and fixed-base structures and
the yielding process were compared.
23.2 Fuzzy Synthetic Evaluation Model (FSE)
The traditionally statistical mean or median of fixed-base and base-isolated structures

is difficult to identify an optimum system with consideration of multiple variables,
such as acceleration, drift, performance level and cost. The ingenious RI index,
proposed by Sayani and Ryan [13], can be used to comprehensively compare the
performance of base-isolated and fixed-base systems. However, the weight factors
in the RI function have not been explicitly defined. Given this, the fuzzy synthetic
evaluation (FSE) is firstly proposed in this study to evaluate the weight factors of
selected performance variables on which the fuzzy analytic hierarchy process (FAHP)
method is based to compare the systems.
One of the major challenges to compare the seismic response of different systems
is to determine the weight factors of selected performance variables. Fuzzy syn-
thetic evaluation (FSE) theory developed by Zadeh [16], has been widely adopted in
engineering domain, such as seismic risk assessment by Carreño et al. [17], inelas-
tic displacement ratio predicting from Ozkul et al. [18], and structural retrofitting
decision-making [19]. FSE, dependent on the prior knowledge of data, is an objective
method to evaluate the weights or scores of selected objects. It mainly depends on
the entropy method that is used to process the fuzzy matrix generated by membership
function.
288 L. Xu et al.
23.2.1 Generation of Fuzzy Matrix
According to the fuzzy set theory, a fuzzy set would be defined in terms of member-
ship function which normalizes the variable domain of interest onto the interval [0,
1]. The membership functions represent the weighting factors of the specified vari-
ables belonging to the set. Performance evaluation matrix of systems with respect to
selected performance variables should be derived in advance to establish the fuzzy
matrix. Suppose X i j denote the value of ith structure relative to jth performance vari-
able, and X j,max denote the maximum value of all structures relative to corresponding
variable. The membership function is defined as
Xi j
μ Xi j = (23.1)
X j,max
No doubt if a system performs best it derives a value of μ = 1 from Eq. (23.1).

Subsequently, the fuzzy matrix X f is derived as
⎡ ⎤
μ1 (X 11 ) · · · μ1 X 1q
⎢ .. .. ⎥
Xf =⎣ ⎦ (23.2)
. μ j Xi j .
μn X n1i j · · · μn X nq
where q is the number of selected performance variables, and n denotes the number
of systems.
23.2.2 Evaluation of the Weights by the Maximum Entropy

Method
According to the maximum entropy principle, the distribution with maximum entropy
would be the proper one. It implicates the importance of given performance variable
can be reflected by the entropy values. Four steps are needed to solve the weights of
each performance variable.
(1) The characteristic weight matrix P would be calculated from the fuzzy matrix
X f firstly to determine the weight for each performance variable.
X f (i, j)
Pi j =
n (23.3)
i=1 X f (i, j)
Namely, performance of ith structure with respect to jth performance variable,

X i j , is normalized by the summation of jth column of fuzzy matrix.
(2) Then the entropy of jth vector of performance evaluation matrix X with respect
to the jth characteristic weight matrix P is obtained as
−1
n
Ej = pi j ln pi j (23.4)
ln(n) i=1
It is easy to find that the range of E j is from zero to one. And it indicates that
larger entropy represents a less diversity factor.
(3) The diversity factor for jth aspect grade is defined as
Gj = 1 − Ej (23.5)
It is difficult to distinguish between the good from bad if the diversity factor
approximates zero.
(4) Finally, the normalized weight for each evaluation aspect grade is derived as
Gj
w j =
n (23.6)
i=1 Gj
where W is the right weight vector that can be used in FAHP in the following
section.
23.2.3 Fuzzy Analytic Hierarchy Process (FAHP)
Analytic hierarchy process (AHP), based on pairwise comparison by one-nine

method, proposed by Wind [20], is used to determine the relative performance of mul-
tiple objectives. Chang extended the AHP method improved the conventional AHP
method to closed intervals by using triangular fuzzy membership function which was
named as FAHP [21].
Note that there are two important steps to derive the ranking of systems, including
the assessment of weight of each variable (i.e., weight matrix W) and evaluation of
each system with respect to each variable (i.e., weight matrix A). For the first step,
FAHP method from Chang is used to determine the fuzzy number which is assumed
as triangular fuzzy function for simplicity [21]. The triangular membership function
equals to
⎧
⎪
⎨ m−l − m−l , l ≤ x ≤ m
x l
μ Mi j (x) = m−u − m−l , m ≤ x ≤ u

x l
(23.7)
⎪
⎩ 0, other s
where μ is the membership function, l and u are the lower and upper values of the
interval, respectively; the fuzzy degree depends on the interval size. A non-fuzzy
number is derived when l = m = u. Let Mi = (li , m i , u i ) and M j = l j , m j , u j
denote the ith and jth triangular fuzzy number of ith and jth systems, respectively.
By dividing m j by m i , a pairwise comparison element, μ Mi j , is derived. The fuzzy
290 L. Xu et al.
number Mi j is (1, 1, 1) when i equals to j. Thus, a pairwise fuzzy evaluation matrix

E relative to each objective is presented in Eq. (23.8).
⎡ ⎤
(1, 1, 1) e12 . . . e1n
⎢ e21 (1, 1, 1) . . . e2n ⎥
E =⎢
⎣ ...
⎥ (23.8)
··· ... ... ⎦
em1 em2 . . . emn
where the element ei j denotes the fuzzy number ofselected performance variable of
ith system with respect to jth system, i.e., ei j = li j , m i j , u i j , which are satisfied
with
1 1 1
li j = , mi j = , ui j = (23.9)
l ji m ji u ji
Let Mi j denote the values of ith system relative to jth system of a performance
variable. Then the fuzzy value of ith system with respect to the selected performance
variables is defined as
⎡ ⎤−1

m
n
m
Mi ⎣ Mi ⎦
j j
Si = (23.10)
j=1 j=1 j=1
where is a fuzzy operator; n equals to m as the number of systems. Then m fuzzy

numbers for all systems relative to all selected performance variables are derived.
The next step is to evaluate the degree of possibility of Mi ≥ M j according to Chang
[21],

V Mi ≥ M j = 1 i f f mi ≥ m j
(23.11)
V M j ≥ Mi = hgt M j ∩ Mi = μ Mi (d) other s
where d denotes the highest intersection value between fuzzy membership functions
μ Mi and μ M j , which equals to Eq. (23.12)
li − u j
V M j ≥ Mi = hgt M j ∩ Mi = (23.12)
m j − u j (m i − li )
The weight factor, Ci , of ith system of each performance variables is defined as
Ci = min V (Si ≥ Sk ), k = 1, 2, 3, . . . , n, and i = k (23.13)
The final weight vectors of systems relative to selected performance variables are
derived via normalization, which is derived as
C
C i =
m i (23.14)
i=1 Ci
Finally, the final scores of systems are in terms of the performance comparison
matrix C and the weight matrix W, and shown as
Y = CW (23.15)
23.3 Review of Base-Isolated and Fixed-Base Structure

Design According to ASCE 7–10
The major challenge for the isolated structure is to perform the design of super-
structures the same way as fixed-base structures. The equivalent lateral force (ELF)
methodology, in spite of being suitable for structural height no more than 65 ft or
four stories, is useful to check against the nonlinear time history response [4]. Hence,
the design shear of isolated structures according to code-defined equation, instead
of being based on the period of isolation system, is prescribed as a function of the
stiffness and design displacement of the isolation system
kI DD
VI = (23.16)
RI
where k I and R I are the lateral stiffness and the strength modification factor of isola-
tion system; D D , the design displacement
ofisolators in the case of design earthquake,
is determined by D D = gS D1 TI / 4π 2 B D , where S D1 is the 5% damped spectral
acceleration at 1 s period in units of g-s; TI is the effective period of isolation system
at the design displacement using the minimum effective stiffness of the isolation sys-
tem. And B D , a coefficient related to the effective damping, β D , which is correlated
with the design displacement and dissipated energy of isolated system is set forth in
code ASCE 7–10 [15]. Substituting D D and equation of k I = (W/g)(2π/TI )2 into
Eq. (23.16), where m is the total mass of isolated structure including isolation base
floor.
S D1
VI = W (23.17)
B R I TI
where W is the total structural weight, including isolation floor, of base-isolation

system. Thus, the shear for isolated structure is specified, which, however, cannot be
directly used for the design of isolated structure within structural software. In contrast,
the design base shear of fixed-base structure is similar to that of base-isolation, which
is shown as
292 L. Xu et al.

S DS S D1
VF = min WF , WF (23.18)
R/Ie TF (R/Ie )
where S DS is the design spectral acceleration parameter in the short period range; R
is the strength modification factor, which is no more than 8 for fixed-base structures;
Ie is the importance factor according to risk category of building, and is assumed
to be one here; W F is the weight of superstructure and, TF is the structure period,
TF = 0.016H 0.9 for concrete moment-resisting frame, where H is the height of
structure, in ft. Suppose the design shear force, VI , of an isolated structure equals
to VF of an equivalent fixed-base structure, the equivalent modification factor Re is
derived

S DS
TI B NN+1 R I , TF < SSDS
D1
Re = S1D1 (23.19)
TF I
T B N +1 R I , TF ≥ SDS
N S D1
Assuming the weight of isolated floor equals to other typical story weight, the total
weight of base-isolated structure, W, is W F × (N + 1)/N , where N is the number
of storey excluding the isolated level. Figure 23.1 shows the maximally allowable
equivalent strength modification factors, Re , of isolated structures.
In Fig. 23.1, the strength reduction, R I , of isolated structure is set to the maximum,
2.0, and the damping coefficient, B, is equal to 1.2, i.e. the damping ratio is 10%
for the isolators; the spectral acceleration ratio of S DS /S D1 is equal to 0.794, where
S DS = 1.13 g and S D1 = 0.9 g; and the story heights are assumed to be 10 ft.
Figure 23.1 shows that the equivalent strength modification factors (ESMF) of the
isolated structures are less than 8 when the isolation period is no more than 3 s.
Namely, the minimum design shears of isolated structures are larger than those needed
for fixed-base structures. It also can be observed that the longer the isolation period
is the less strength demand of isolated structure needs for a superstructure.
Fig. 23.1 Relation between

ESMF and vibration periods
of isolated structures
23.4 Modeling RCF Structure and Base-Isolation System
23.4.1 Design and Modeling of Superstructure
The superstructures, including 1-storey, 2-storey, 4-storey, and 8-storey are designed
according to the US seismic codes [15, 22, 23]. The structural models used for
isolated structures are modified according to Haselton’s models [24, 25] where the
P-Delta effects are taken into consideration using leaning columns with gravity loads
rigidly connected with frame as shown in Fig. 23.2. The horizontal displacement of
base leaning column is coupled with the base floor of the main RC frame.
Structural models are completed in Open Systems for Earthquake Engineering
Simulation (OpenSees) [26], which is effectively to simulate the collapse of RC
frame considering deterioration. The isolated structures are designed according to
ESMFs derived from Eq. (23.19). Note that the actual vibration periods of isolated
structures, TF , are adopted according to eigenvalue in OpenSees analysis. In contrast,
the strength modification factor, R, of fixed-base structures are chosen as eight, which
correspond with special moment frame. Space frames are considered in this study
which the ratio of tributary area for gravity and lateral loads is set to one.
Joint2D element, well capturing cycle and in-cycle deterioration hysteric modes
is used herein for simulating sideway collapse. There includes zero-length lumped
plastic hinges which can model the cyclic deterioration placed at the external nodes
of Joint2D element [27]. The tri-linear monotonic backbone curve and some hysteric
rules are shown in Fig. 23.3b, while model without consideration of degradation is
compared in Fig. 23.3a. Yield strength and other parameters of the backbone curve
are derived in terms of the results calibrated by FEMA P695 [28]. In addition, the
shear panel of Joint2D element whose stiffness is only considered according to its
cracked stiffness for simplicity [29]. For these models, the damping is only applied
to the linear elements that the factor applied to linear elements initial stiffness is 1.1
[30]. Besides, the damping ratio is equal to 5% at the first mode period.
Note also that the isolated structures have additional beam-slab system at the base
floor not only to resist the gravity and horizontal seismic loads but also to ensure
integrity of movement of base floor. For simplicity, an identical design of the second
floor is assumed in the base floor of isolated structures.
Fig. 23.2 Schematic

diagram of base-isolated
structure
294 L. Xu et al.
Fig. 23.3 Force-displacement relation of beam-column joint a without deterioration; b considering

deterioration
Fig. 23.4 Force-

displacement relation of
isolation system
23.4.2 Isolation System
In order to derive the value of ESMF of isolated structures, parameters of isolation

system should be determined initially. The zero-length elastomericBearingBoucWen
element, no contribution to the Rayleigh damping, is modeled at the base of column
to simulate the isolators. Bilinear hysteresis model is used to character the force-
deformation behavior of isolation systems shown in Fig. 23.4. Even though there
are several outstanding isolation devices, their hysteric behavior can be described
with a similar bilinear model. The characteristic strength, Q, and yield strength, Fy ,
of isolator should be designed strong enough to resist the horizontal loads without
yielding in exclusion of earthquake load. The post-yield stiffness, k p , is the key
parameter that controls the fundamental period of isolation system.
As the isolator yields, the post-yield stiffness, k p is defined by the period of
isolation system, T p , and the gravity, W, shown as
4π 2
kp = W (23.20)
gT p2
where g is the gravity constant. From Fig. 23.4, the effective stiffness of isolation
system can be computed as
Q
ke f f = + kp (23.21)
DD
where Q is the characteristic strength and D D is the displacement of isolator under

design earthquake. As such, the effective period of isolation system is achieved
similar with Eq. (23.20) as

w
TI = 2π (23.22)
ke f f g
The effective period of isolation system cannot be specified until the design dis-
placement is given. Take in mind the effective damping of isolation system has not
been defined and, is expressed as

ED
βD = (23.23)
2π ke f f D 2D
where the parameter of E D representing the energy dissipation

for each cyclic motion
of the bearing which is denoted as E D = 4Q D D − D y , where D y is the yield
displacement of isolator. Combining the Eq. (23.20) through (23.23) in addition to
D D defined as previously, the isolation system parameters, i.e., displacement D D ,
effective damping ratio β D , and effective period TI for the design earthquake (DE)
are derived when D y assumed as 10 mm. Similarly, other parameters of isolation
system, such as D M , TM , and β M for MCER earthquake are also calculated as T p
and Q/W are given, and results are shown in Table 23.1. In this study, B M will be
preferred to derive a larger design force for the superstructure for its corresponding
ESMF would be smaller than that derived from B D .
The parameters of isolator iteratively computed given in Table 23.2 will be used
for the structural modes illustrated following. Note also that the ESMF is updated
thereafter to redesign the reinforcement in ETABS based on Table 23.2. Nonlinear
response history analyses (NRHA) of the base-isolated and fixed-base structures are
carried out in OpenSees. Adaptive algorithm is used to solve the large deformation.
23.4.3 Verification of Proposed Model
Nonlinear static analysis, namely pushover analysis, is carried out based on the load
pattern recommended by ASCE 7–10. An 8-storey model has been used to verify
296 L. Xu et al.
Table 23.1 Design parameters of isolation system according to ASCE (D y = 10 mm)

Design earthquake(DE) MCER
No. T p (s) Q/W D D (m) TI (s) βD D M (m) TM (s) βM
1 2.5 0.05 0.256 2.19 0.142 0.466 2.315 0.089
2 2.5 0.1 0.166 1.797 0.289 0.316 2.048 0.203
3 2.5 0.15 0.121 1.463 0.384 0.246 1.792 0.297
4 3 0.05 0.274 2.528 0.178 0.512 2.718 0.112
5 3 0.1 0.173 1.981 0.338 0.341 2.332 0.245
6 3 0.15 0.126 1.569 0.426 0.256 1.974 0.347
7 3.5 0.05 0.288 2.832 0.212 0.553 3.1 0.135
8 3.5 0.1 0.177 2.123 0.38 0.359 2.575 0.284
9 3.5 0.15 0.131 1.65 0.457 0.261 2.112 0.39
10 4 0.05 0.305 3.111 0.243 0.585 3.455 0.159
11 4 0.1 0.181 2.239 0.413 0.372 2.781 0.32
12 4 0.15 0.134 1.714 0.481 0.27 2.234 0.422
Table 23.2 Design parameters of isolated structure and conventional structure

Design No. of Bay RI R(R I ) Period Design Peak Over-
number story width of super- base base strength
(ft) structure shear shear
(s) (kips) (kips)
Base 1 20 2 7.6 0.42 49.3 112.8 2.29
isolated 2 20 2 10.1 0.62 51.5 125.2 2.43
4 20 2 8.3 0.89 89.5 167.4 1.87
8 20 2 4.7 1.75 157.5 173.3 1.1
Fixed 1 20 – 8 0.42 46.9 102 2.17
base 2 20 – 8 0.62 65 127.6 1.96
4 20 – 8 0.89 92.9 167.5 1.8
8 20 – 8 1.76 92 129.5 1.41
the model built in this study with the same design reinforcement and the results are
shown in Fig. 23.5a. The structural base shear capacities of both models are almost
the same, even though the initial stiffness of both models are of little difference due to
the different cracked stiffness functions used. The ductile capacity of SMF structure
is stronger than that of IMF structure shown in Fig. 23.5b. While the latter structure
gains much higher maximum base shear.
The minimal design strength coefficient compliant with respect to ASCE 7–10 of
base-isolated and fixed-base structures are both computed as
S D1
Cs = (23.24)
TF R
Fig. 23.5 Base shear capacity evaluation from PUSH-OVER analysis a verification with Haselton’s
Model; b comparison from structures with different strength reduction factors
where R is the strength modification factor which is replaced with Re for the isolated
structure. Subsequently, the design base shear force is obtained using Eq. (23.24)
times the structural weight, and the over-strength factor equals to peak base shear from
pushover curve divided by design base shear. The isolated structures are designed
with the isolation period equal to 4 s and Q/W equal to 0.05. All of the dynamic
properties of both types of structure prototype are shown in Table 23.2.
23.5 Ground Motion Selection and Modification
Sixty ground motions, divided into three bins, are selected to represent three types
of earthquake scenario: 50, 10 and 2% probability of exceedance (PE) in 50 years
with respect to group 1, 2 and 3. Most of the ground motions are selected from Los
Angeles SAC suite [31], while twenty ground motions are selected from the PEER
NGA database. And some ground motions in each bin are reused in other bins, which
are listed in Table 23.3. The earthquakes are amplitude scaled based on the method
proposed by Constantinou et al. [32] and Pant et al. [33], Scaling factors are also
listed in Table 23.3.
The target spectra, namely the uniform hazard response spectrum (UHRS) with
respect to three levels of hazard are derived from US Geological Survey (USGS)
(https://geohazards.usgs.gov/hazardtool/application.php). The comparison of the
response spectra with 5% damping ratio of scaled ground motions, the mean spectra
and the target spectra is shown in Fig. 23.6.
298 L. Xu et al.
Table 23.3 Information of the selected ground motions

Group Earthquake and station M D (km) SF
Group 1 for 50% PE Chi-Chi, Taiwan, 1999 HWA049 7.62 50.76 1.499
(049N/049S)
Northridge, USA, 1994 W Pico 6.69 5.48 0.491
Canyon Rd (WPI046/WPI316)
St. Elias, USA, 1979 Yakutat 7.54 80 1.403
(v2009/v2079)
Landers, USA, 1992 Palm 7.28 36.15 1.871
Springs Airport
(PSA000/PSA090)
Northridge, USA, 1994 Sylmar 6.69 5.35 0.255
Converter Sta.
(SCS052/SCS142)
Northridge, USA, 1994 Pardee 6.69 7.46 0.388
SCE (PAR-L/PAR-T)
Northridge, USA, 1994 Rinaldi 6.69 6.5 0.232
Rec. Sta. (RRS228/RRS318)
Imperial Valley, USA, 1979 El 6.53 6.17 0.785
Centro Array 10
(E10050/E10320)
Kobe, Japan, 1995 Shin-Osaka 6.9 19.15 0.643
(SHI000/SHI090)
Northwest China, 1997 J411 6.11 19.11 0.588
(N000/N270)
Superstition Hills, USA, 1987 6.54 13.03 0.865
Westmorland Fire Sta.
(WSM090/WSM180)
Imperial Valley, USA, 1979 6.53 22.03 0.583
6605 Delta (DLT262/DLT352)
Northridge, USA, 1994 Newhall 6.69 5.92 0.202
Fire Sta. (NWH090/NWH360)
Imperial Valley, USA, 1987 El 6.54 18.2 0.753
Centro Imp. Co. Ctr.
(ICC000/ICC090)
Corinth, Greece, 1981 Corinth 6.6 10.27 0.709
(COR-L/COR-T)
Northridge, USA, 1994 Sun 6.69 10.05 0.538
Valley - Roscoe Blvd
(RO3000/RO3090)
Coalinga, USA, 1983 Pleasant 6.36 8.41 0.464
Valley PP Bldg
(PVB045/PVB135)
(continued)
Table 23.3 (continued)

Coalinga, USA, 1983 Cantua 6.36 24.02 0.593
Creek School
(CAK270/CAK360)
Coalinga, USA, 1983 Pleasant 6.36 8.41 0.257
Valley PP Yard
(PVY045/PVY135)
Chalfant Valley, USA, 1986 6.19 7.58 0.342
Zack Brothers Ranch
(ZAK270/ZAK360)
Group 2 for 10% PE Imperial Valley, USA, 1940 El 6.9 10 0.604
Centro (SAC la01/la02)
Array 05 (SAC la03/la04)
Array 06 (SAC la05/la06)
Landers, USA, 1992 Barstow 7.3 36 1.017
(SAC la07/la08)
Landers, USA, 1992 Yermo 7.3 25 0.959
(SAC la09/la10)
Loma Prieta, USA, 1989 Gilroy 7 12 0.511
(SAC la11/la12)
(SAC la13/la14)
Rec. Sta. (SAC la15/la16)
Northridge, USA, 1994 6.7 6.4 0.541
Sylmar-Olive View Hospital
(SAC la17/la18)
North Palm Springs, USA, 1986 6 6.7 0.402
North Palm Springs (SAC
la19/la20)
Converter Sta.
(SCS052/SCS142)
Kobe, Japan, 1995 JMA 6.9 0.96 0.517
(KJM000/KJM090)
Superstition Hills, USA, 1987 6.54 0.95 0.892
Parachute Test Site
(B-PTS225/B-PTS315)
Northridge, USA, 1994 Pardee 6.7 7.46 0.822
SCE (Par-L/PAR-T)
San Salvador, El Salvador, 1986 5.8 9.54 0.851
National Geographical Inst.
(NGI180/NGI270)
(continued)
300 L. Xu et al.

Superstition Hills, USA, 1987 El 6.54 18.2 1.595
Centro Superst/B
(ICC000/ICC090)
Kobe, Japan, 1995 Takatori 6.9 1.47 0.475
(TAK000/TAK090)
Erzican, Turkey, 1992 Erzican 6.69 4.38 0.744
(ERZ-NS/ERZ-EW)
Dinar, Turkey, 1995 Dinar 6.4 3.36 0.852
(DIN090/DIN180)
Kobe, Japan, 1995 Shin-Osaka 6.9 19.15 1.361
(SHI000/SHI090)
Group 3 for 2% PE Kobe, Japan, 1995 Kobe St. 6.9 3.4 0.63
(SAC la21/la22)
Loma Prieta, USA, 1989 Los 7 3.5 0.964
Gatos Pres. Ctr. (SAC la23/la24)
Rec. Sta. (SAC la25/la26)
Northridge, USA, 1994 6.7 6.4 0.546
Sylmar-Olive View Hosp. (SAC
la27/la28)
Tabas, Iran, 1974 Tabas (SAC 7.4 1.2 0.741
la29/la30)
Elysian Park Fault Simulated 7.1 17.5 0.372
(SAC la31/la32)
(SAC la33/la34)
(SAC la35/la36)
Palo Verdes Fault Simulated 7.1 1.5 0.892
(SAC la37/la38)
Palo Verdes Fault Simulated 7.1 1.5 1.164
(SAC la39/la40)
Conv. Sta. (SCS052/SCS142)
N. Palm Springs, USA, 1986 6.06 12.07 2.549
Morongo Valley
(MVH045/MVH135)
Kobe, Japan, 1995 JMA 6.9 0.96 0.851
(KJM000/KJM090)
Northridge, USA, 1994 6.7 7.46 1.36
Pardee-SCE (PAR-L/PAR-T)
(continued)

Fire Sta. (NWH090/NWH360)
Kobe, Japan, 1995 Takatori 6.9 1.47 0.783
(TAK000/TAK090)
Erzican, Turkey, 1992 Erzican 6.69 4.38 1.225
(ERZ-NS/ERZ-EW)
Duzce, Turkey, 1999 Duzce 7.14 6.58 1.251
(DZC180/DZC270)
Dinar, Turkey, 1995 Dinar 6.4 3.36 1.39
(DIN090/DIN180)
Parkfield, USA, 2004 Parkfield 6 4.8 1.197
Fault Zone 1 (360/90)
Fig. 23.6 Ground motion processing a scaling results of 50% PE ground motions; b scaling results
of 10% PE ground motions; c scaling results of 2% PE ground motions; d UHRSs and design
spectra
302 L. Xu et al.
23.6 Application of FAHP in Performance Comparison
23.6.1 Weights Factors of Performance Variables Using FSE

Method
In this study, 40 prototype structures have been modeled in OpenSees, including

4 fixed-base structures and 36 base-isolated structures. The corresponding model
information is listed in Table 23.4. Isolated structures of systems including M12,
M21, M30 and M39 referenced as prototypes for other isolated structures are mini-
mally compliant with ASCE 7–10. It implies that the strength of isolated structures
except the prototypes are less than the code-defined strength. Six types of perfor-
mance variables have been selected to evaluate the performance of 40 systems. Each
performance variable, such as peak drift ratio, peak floor acceleration, peak drift
differential from average, peak acceleration differential from average, drift ratio per-
formance and acceleration performance are divided into three sub-variables shown
in Table 23.5.
Table 23.4 Structural information of fixed-base and base-isolated structures

Type Storey # Model # Tp Q Storey # Model # Tp Q
Fixed base 1 M1 – – 4 M3 – –
2 M2 – – 8 M4 – –
Base 1 M5 2.5 20 4 M23 2.5 20
Isolated M6 2.5 1.76 M24 2.5 6.49
M7 2.5 2.22 M25 2.5 8.32
M8 3.5 20 M26 3.5 20
M9 3.5 1 M27 3.5 3.29
M10 3.5 1.46 M28 3.5 4.97
M11 4 20 M29 4 20
M12 4 1 M30 4 2.68
M13 4 1.23 M31 4 4.05
2 M14 2.5 20 8 M32 2.5 20
M15 2.5 3.14 M33 2.5 20.53
M16 2.5 4.05 M34 2.5 40.06
M17 3.5 20 M35 3.5 20
M18 3.5 1.76 M36 3.5 8.32
M19 3.5 2.53 M37 3.5 14.43
M20 4 20 M38 4 20
M21 4 1.38 M39 4 6.19
M22 4 2.22 M40 4 10.77
Table 23.5 Definition of Variables Description Variables Description

performance level of
structures A1 Peak accel under A10 Drift differential
2% PE from average
under 2% PE
A2 Peak accel under A11 Drift differential
10% PE from average
under 10% PE
A3 Peak accel under A12 Drift differential
50% PE from average
under 50% PE
A4 Peak drift under A13 Accel
2% PE performance
under 2% PE
10% PE performance
under 10% PE
50% PE performance
under 50% PE
A7 Accel differential A16 Drift ratio
from average performance
under 2% PE under 2% PE
For the candidate systems, weights of performance variables should be evaluated

by metrics that rationally reflect the relative performance differential from each other.
Note that three metrics are used to assess the seismic performance of systems with
respect to the 18 performance variables.
(1) The performance of one system undoubtedly surpasses the other under the same
ground motion if its seismic demand is smaller. Grade point is used to record the
wins or losses of a system with respect to others within the same performance
variable under a same ground motion. Then the total wins are summed up within
each bin of ground motions and calculated as

m
X i1 = wik (23.25)
k=1
where wik denotes wins of ith structure from the other structures under m ground
motions within each bin of ground motion events, m equals to 20 and i is from
304 L. Xu et al.
1 to 32 in this study. This evaluation function is only limited to peak drift ratio
and peak acceleration. The best system derives 32 points, whereas, the worst
system derives 1 point. Note that a collapsed system will be precluded in order
to not to generate a non-positive definite matrix in the following fuzzy matrix.
(2) It is in great need to see how much a system wins or loses under each bin of
ground motions from the overall average of all systems from the quantitative
evaluation perspective. Hence, the average differential of a system from the
average of all systems is proposed to evaluate how much better or worse a
system performs with respect to other systems. Let X i2 denotes the average
differential which it can be calculated as
1 v̄k − v̄ki + v̄k,max − v̄k,min

m
X i2 = (23.26)
m k=1 v̄k,max − v̄k,min
where v̄k is the average response value of all systems under kth bin of ground
motions, and v̄ki denotes the average response of ith system under kth bin of
ground motions; v̄k,max and v̄k,min denote the maximum and minimum response
values under kth bin of ground motions, respectively. By introducing the differ-
ential term between the maximum and minimum, Eq. (23.26) avoids negative
values in the fuzzy matrix.
(3) Performance levels of the base-isolated and fixed-base systems would be dif-
ferent even though their seismic responses are the same. The damage state of
fixed-base and base-isolated structures are defined according to HAZUS in terms
of the drift and acceleration values [34], which are listed in Table 23.6. Keep in
mind that a system is in collapsed damage state when the singularity takes place
in this study. A system derives grade points according to its damage states. Take
for example a system in slight damage, moderate damage, extensive damage
and complete damage states are recoded as 24 , 23 , 22 , 21 points, respectively.
The sensitivity of other evaluation functions that are available for the evalua-
tion weight of performance variables will be further analyzed in the following
section. The grade point a system derives is given as

m
X i3 = vik (23.27)
k=1
where vik denotes the point of ith structure achieved under kth ground motion.
Hence, the fuzzy matrix of these systems with respect to the 18 performance
variables is derived and, the final weight vectors of performance variables are derived
following the steps from Eq. (23.3) to Eq. (23.6), and shown in Fig. 23.7.
The corresponding weights of A13 and A14 are much smaller than those of the
others because the normalized grade points each system derived in the fuzzy matrix
are the same. In contrast, performance variables of peak drift ratios, i.e., A4–A6 derive
the highest values than others. And the performance variables of peak accelerations,
Table 23.6 Definition of performance level of structures

Type Damage Floor Drift ratio Type Damage Floor Drift ratio
state accel (g) state accel (g)
BI Slight 0.25 0.4 C1L Slight 0.3 0.5
Moderate 0.5 0.69 Moderate 0.6 1
Extensive 1 1.57 Extensive 1.2 3
Complete 2 2 Complete 2.4 8
C1M Slight 0.3 0.33 C1H Slight 0.3 0.25
Moderate 0.6 0.67 Moderate 0.6 0.5
Extensive 1.2 2 Extensive 1.2 2
Complete 2.4 5.33 Complete 2.4 4
Fig. 23.7 Normalized

weights of the selected
performance variables using
FSE method
i.e., A1–A3 are not far behind from Fig. 23.7. Besides, other performance variables,
such as A16 also plays an important role to the systems.
23.6.2 Comparison Results Based on FAHP
Without consideration of the failure systems, performance comparison matrix C has

been shrunk to a 32 × 32 matrix. For the FAHP method, pairwise comparison of
each system relative to selected performance variables will be evaluated firstly. The
pairwise comparison of each system relative to performance variables of A1–A3 is
performed using the evaluation matrix proposed by Chang [21], Taking the peak
accelerations of systems under 2% PE ground motions for example, the mean accel-
eration response, μ E D , is chosen as the maximum value of triangular fuzzy number
and μ E D ∓ σ E D are selected as the lower and upper values of interval. The pairwise
comparison fuzzy matrix E of peak acceleration and pairwise comparison of other
performance variables is carried out with conventional AHP method.
306 L. Xu et al.
Fig. 23.8 Final normalized scores of systems based on FAHP method
Then the fuzzy values of ith system with respect to the performance variables are
derived with Eq. (23.10). The final weights of systems with respect to each perfor-
mance variable are obtained with Eq. (23.11) through Eq. (23.14), subsequently. The
final comparison result of systems can be simply derived by Eq. (23.15) and shown
in Fig. 23.8.
The M12 and M9 systems rank the first and second places, while the fixed-base
structures rank at the bottom with consideration of larger intensity earthquakes. The
performance of isolated structures is superior to that of the fixed-base structures
without consideration of collapsed structures from Fig. 23.8. The isolated structures
perform worse overall with the increase of superstructure story number. The per-
formance levels of isolated 8-storey systems are less distinct from Fig. 23.8. That
implies the seismic performance of isolation system may decrease as the periods of
superstructure become larger.
As shown in Fig. 23.8, more than half of the isolated 8-storey systems fail, which
means the further strength reduction of isolated 8-storey structures should be cau-
tious. And the maximum drift ratios of all un-collapsed isolated structures are less
than 2%, which implies that the isolated structures would fail if extensive plastic-
ity takes place. The corresponding maximum drift ratio of isolated structures and
fixed-base structures are shown in Fig. 23.9. All the median maximum drift ratios
of isolated structures are nearly the same. An inference is a conclusion that the seis-
mic response of isolated structure would cause less damage than that of fixed-base
structure without consideration of collapse. Moreover, the higher-rise isolated struc-
tures would be more easily failed if their strengths are reduced compared with the
lower-rise ones.
Fig. 23.9 Statistical results

of maximum drift ratio of
systems
23.6.3 Sensitivity Analysis
23.6.3.1 The Sensitivity of Performance Evaluation Function

on the Final Ranking
The performance evaluation function of Eq. (23.27), based on power of two, can
be used in other forms to enlarge the performance differential from slight damage
to complete damage. In this section, the evaluation function with power of three is
introduced to illustrate its sensitivity to the final comparison result of systems.
Firstly, the weights of performance variables from two types of functions are
shown in Fig. 23.10a. The performance variables of performance levels become more
important when their performance differentials get larger compared to the result from
original evaluation function. However, the final ranking of systems original keeps
unchanged according to Fig. 23.10b. It is because the evaluation function can only
change the performance differential among systems but not the ranking results of
systems with respect to the performance variables.
23.6.3.2 The Sensitivity of Membership Function on the Final Ranking
The comparison result of each system under the same ground motion is relatively
fixed based on previously pairwise comparison model, which implies that the final
ranking of systems is mainly dependent on the weight of each performance variable.
In the FSE analysis, the membership function defined in Eq. (23.2) is linear previ-
ously, which however, can be defined as other formula. In order to illustrating the
effect of membership functions on the final weighted result, the sensitivity of mem-
bership functions on the weight of each performance variable is performed using
equation as
k
Xi j
μ j Xi j = (23.28)
X jmax
308 L. Xu et al.
Fig. 23.10 Sensitivity analysis of the evaluation function on a the weight factors of performance
variables; b the final ranking of systems
The weight distribution of each performance variable is analyzed according to

Eq. (23.28), and the results are shown in Fig. 23.11a. As the value of k increases,
performance variables from M1 to M6 tend to be more important, whereas the other
aspects tend to be less important and, the aspect of M16 decreases the most. However,
the final ranking of all systems has not changed even the value of k increases to three,
shown in Fig. 23.11b. It implies that linear membership function of Eq. (23.28) is
effective for evaluating the weights of performance variables.
23.7 Conclusions
This study focuses on the comparing the seismic performance between the fixed-base
and base-isolated structures using multiple combinations of isolation and isolated
structure characteristics. Equivalent strength modification factor (ESMF) is utilized
to establish four verified prototype isolated structures minimally compliant with
ASCE 7–10 in addition to all fixed-base structures. Then, the FSE method is used to
evaluate the weights of performance variables and the FAHP method is subsequently
used to identify the best system from results under multi-intensities earth-quakes in
Fig. 23.11 Sensitivity analysis of the membership function on a the weight factors of performance
variables; b the final ranking of systems
quantitative way. Evaluation function and membership function are certified to be

effective by sensitivity analysis.
The minimal strength an isolated structure needs is estimated by ESMF with
which the isolated structure can be conveniently designed in structural software. It
is found that the superstructures with longer periods of isolation systems may need
much more strength than those of shorter periods. In contrast, isolated structures
with more damping ratios in the isolation system need less strength than those with
less damping ratios. However, the minimal strength the isolated structure needs is
overall more than that of corresponding fixed-base structure.
The comparison method of seismic performance with respect to the isolated fixed-
base structures is explained in detail. Results show that the peak drift ratio and peak
acceleration are the two most important aspects. Performance variables evaluation
function and membership function used in this study are effective via sensitivity
analysis. Note that all the isolated structures minimally compliant with code survive
under all the excitations to which they are subjected, while fixed-base systems of
M1 and M3 both suffer sidesway collapse once. The performance of base-isolated
structures are comprehensively superior to the fixed-base structures, meanwhile their
performance becomes less distinctly excellent as the isolated structures get higher.
Acknowledgements This work is supported by the Shandong Province Science and Technol-
ogy Development Plan Item (2014GSF122001), National Natural Science Foundation of China
310 L. Xu et al.
(51678208, 51238012), and the Co-operative Innovation Center of Engineering Construction and
Safety in Shandong Peninsula Blue Economic Zone. We also thank the scientific research innova-
tion fund (key cultivation project) of Harbin Institute of Technology (HIT.NSRIF.201709). These
are gratefully acknowledged.
References
1. Taghavi, S., Miranda, E.: Response Assessment of Non-structural Building Elements, Pacific
Earthquake Engineering Research Centre, Richmond, CA (2003)
2. Ramallo, J.C., Johnson, E.A., Spencer Jr., B.F.: “Smart” base isolation systems. Earthq. Eng.
Struct. Dyn. 128, 1088–1099 (2002)
3. SEAOC: Vision 2000—Performance based seismic engineering of buildings, vols. I, II. Vision
2000 Committee, Sacramento, Calif (1995)
4. Chimamphant, S., Kasai, K.: Comparative response and performance of base-isolated and
fixed-base structure. Earthq. Eng. Struct. Dyn. 45, 5–27 (2016)
5. UBC: International Conference of Building Officials. UBC, Whittier, CA (1997)
6. Palazzo, B., Petti, L.: Reduction factors for base isolated structures. Struct. Saf. 60, 945–956
(1996)
7. Ceccoli, C., Mazzotti, C., Savoia, M.: A refined seismic analysis and design of buried pipeline
for fault movement. Earthq. Eng. Struct. Dyn. 28, 633–653 (1999)
8. Kikuchi, M., Black, C.J.: On the response of yielding seismically isolated structures. Earthq.
Eng. Struct. Dyn. 37, 659–679 (2008)
9. Vassiliou, M.F., Tsiavos, A.: Dynamics of inelastic base-isolated structures subjected to ana-
lytical pulse ground motions. Earthq. Eng. Struct. Dyn. 42, 2043–2060 (2013)
10. Lin, A.N., Shenton III, H.W.: Seismic performance of fixed-base and base-isolated steel frames.
J. Eng. Mech. ASCE 118, 921–941 (1992)
11. Shenton III, H.W., Lin, A.N.: Relative performance of fixed-base and base-isolated concrete
frames. J. Struct. Eng.-ASCE 119, 2952–2968 (1993)
12. Ryan, K.L., Morgan, T.A., Sayani, P.: Consistent performance comparison of seismic-isolated
and fixed-base buildings. In: Proceedings of 8th US National Conference of Earthquake Engi-
neering, San Francisco, Oakland, CA (2006)
13. Sayani, P.J., Ryan, K.L.: Comparative evaluation of base-isolated and fixed-base buildings
using a comprehensive response index. J. Struct. Eng.-ASCE 135, 698–707 (2009)
14. Cutfield, M., Ryan, K., Ma, Q.: Comparative life cycle analysis of conventional and base-
isolated buildings. Earthq. Spectra. 32, 323–343 (2016)
15. American Society of Civil Engineers.: Minimum Design Loads for Buildings and Other Struc-
tures, Reston, VA (2010)
16. Zadeh, L.A.: Fuzzy algorithms. Inf. Control 12, 94–102 (1968)
17. Carreño, M.L., Cardona, O.D., Barbat, A.H.: New methodology for urban seismic risk assess-
ment from a holistic perspective. Bull. Earthq. Eng. 10, 547–565 (2012)
18. Ozkul, S., Ayoub, A., Altunkaynak, A.: Fuzzy-logic based inelastic displacement ratios of
degrading RC structures. Eng. Struct. 75, 590–603 (2014)
19. Caterino, N., Iervolino, I., Manfredi, G.: Comparative analysis of multi-criteria decision-
making methods for seismic structural retrofitting. Comput.-Aided Civ. Inf. 24, 432–445 (2009)
20. Wind, Y., Saaty, T.L.: Marketing applications of the analytic hierarchy process. Eur. J. Oper.
Res. 26, 641–658 (1980)
21. Chang, D.Y.: Applications of the extent analysis method on fuzzy AHP. Eur. J. Oper. Res. 95,
649–655 (1996)
22. American Concrete Institue., ACI 318-14: Building Code Requirements for Structural Concrete
and Commentary, Farmington Hills, MI (2014)
23. International Code Council.: 2012 International Building Code, USA (2012)
24. Haselton, C.B.: Assessing Seismic Collapse Safety of Modern Reinforced Concrete Moment
Frame Buildings, Stanford University (2006)
25. Chen, Y., Xu, L., Zhu, X., et al.: A multi-objective ground motion selection approach matching
the acceleration and displacement response spectra. Sustainability 10(12), 4659 (2018)
26. Mazzoni, S., McKenna, F., Scott, M.H., Fenves, G.L.: OpenSees command language manual.
Pacific Earthquake Engineering Research (PEER) Center (2006)
27. Federal Emergency Management Agency., Quantification of Building Seismic Performance
Factors, Washington, D.C. (2009)
28. Panagiotakos, T.B., Fardis, M.N.: Deformations of reinforced concrete members at yielding
and ultimate. Struct. J. 98, 135–148 (2001)
29. Sugano, S., Koreishi, I.: An empirical evaluation of inelastic behavior of structural elements in
reinforced concrete frames subjected to lateral forces. V WCEE 841–844 (1974)
30. Ibarra, L.F., Medina, R.A., Krawinkler, H.: Hysteretic models that incorporate strength and
stiffness deterioration. Earthq. Eng. Struct. Dyn. 34, 1489–1511 (2005)
31. Somerville, P.G., Smith, N.F., Graves, R.W.: Modification of empirical strong ground motion
attenuation relations to include the amplitude and duration effects of rupture directivity. Seis-
mol. Res. Lett. 68, 199–222 (1997)
32. Constantinou, M.C., Kalpakidis, I.V., Filiatrault, A.: LRFD-based analysis and design proce-
dures for bridge bearings and seismic isolators. MCEER, USA (2011)
33. Pant, D.R., Constantinou, M.C., Wijeyewickrema, A.C.: Re-evaluation of equivalent lateral
force procedure for prediction of displacement demand in seismically isolated structures. Eng.
Struct. 52, 455–465 (2013)
34. Federal Emergency Management Agency., Flood Model: Technical Manual, FEMA (2003)
Chapter 24
A Pitman Style Fiber Bragg Grating
Displacement Sensor Based on Wedge
Cavity Structure
Sun Li, Su Zhongxin, Zhang Chunwei and Song Yansheng
Abstract In this paper, a new design of fiber Bragg grating (FBG) displacement
sensor is presented based on wedge cavity structure. A new type of pitman FBG
displacement sensor with a simple structure, small size and anti-electromagnetic
interference is invented. Calibration of the FBG displacement sensor is carried out.
The calibrating data shows that the sensitivity is 5.58 pm/mm, adjusted R square is
up to 0.99 and the static error is 5.168%. Moreover, the FBG displacement sensor
is applied in the hysteresis test of steel frame-reinforced concrete infill wall. The
monitoring results from the FBG displacement sensor in this test match well with
the resistance strain displacement meter, to prove the FBG displacement sensor with
a good accuracy. It is also showed that this new FBG displacement sensor produces
smaller influence on the structure, and it is suitable for long-term displacement mon-
itoring of engineering structures.
Keywords Structural health monitoring · Fibre bragg grating (FBG) ·

Displacement sensor · Comparative testing
24.1 Introduction
Displacement is the most basic physical quantity in structural health monitoring to

reflect the structural deformation and failure. The excessive displacement not only
affects the normal use of the structure, but also causes additional stress, with which the
S. Li (B) · S. Zhongxin · S. Yansheng

School of Civil Engineering, Shenyang Jianzhu University Shenyang, Shenyang
110168, Liaoning, China
S. Zhongxin
Z. Chunwei
School of Civil Engineering, Qingdao University of Technology, Qingdao
266033, Shandong, China

314 S. Li et al.
structure capacity would be reduced. Therefore, an accurate measure of the structural

displacement monitoring makes a great significance for the safety of the engineering
structures and reducing the maintenance costs.
The traditional methods of displacement monitoring include electrical measure-
ment method, magnetic sensitive measurement method and ultrasonic measurement
method [1–3]. Electrical measurement method includes potentiometric measurement
method, resistance strain measurement method, capacitive measurement method and
inductive measurement method. First, the major approach of magnetic sensitive mea-
surement is the Hall measurement method, which is the most widely applied method
at present. Second, the electrical and magnetic sensitive monitoring methods are sus-
ceptible to electromagnetic radiated interference but the stabilities are relatively poor.
Third, the ultrasonic measurement method has a large error because of the blind zone.
However, these methods have their own disadvantages and especially have limitations
in the long term displacement monitoring of engineering structures. Therefore, it is
urgent to develop a displacement sensor with high sensitivity, anti-electromagnetic
interference and capable to be applied in long-term health monitoring of engineering
structures.
Fiber Bragg grating is a photosensitive element with good performance, according
to the coupled-mode theory [4]:
λ B = 2n e f f (24.1)
where λB is the Bragg central wavelength, neff is the effective index of refraction,
and is the grating period. Fiber Bragg grating can limit the influence of external
factors by changing of the central wavelength. Because of its advantages such as
small volume, high sensitivity, anti-electromagnetic interference, easy to layout, and
convenient for quasi-distributed measurement, it has been successfully applied in the
health monitoring for some engineering structures [5–8].
Fiber Bragg grating, as the core sensing element in this paper, is developed to be
a pitman style fiber Bragg grating displacement sensor based on wedge cavity struc-
ture. Moreover, the sensor is applied in the hysteresis test of steel frame-reinforced
concrete infill wall, to prove the efficiency of a new monitoring method for displace-
ment monitoring in structural health monitoring.
24.2 Sensor Working Principle
The developed pitman style fiber Bragg grating displacement sensor is based on
wedge cavity structure, which is composed of displacement probe, limit block, spring,
pulley, strut, wedge block, metal case, cantilever beam and fiber Bragg grating pack-
aged on its surface. The pitman style fiber Bragg grating displacement sensor based
on wedge cavity structure is shown in Fig. 24.1 and the working principle is shown
in Fig. 24.2. In Fig. 24.1, Strut 1 ensures the movement of the displacement probe in
24 A Pitman Style Fiber Bragg Grating Displacement Sensor … 315
Fig. 24.1 Cutaway view of the displacement sensor
Fig. 24.2 The schematic

diagram of displacement
sensor the theoretical
derivation is as follows
the horizontal direction, and when the range is reached, the movement of the limit
block is prevented, so the sensor could be protected. When the displacement probe
moves by displacement x, the cantilever beam also moves the same displacement on
the surface of the wedge block. Therefore, when the change of the strain occurs, the
center wavelength of the fiber Bragg grating tagged on the cantilever beam is changed
so the change of the displacement is found by demodulating the central wavelength
variation. The spring provides restoring force for the displacement probe, that is,
when the sensor stops working, the displacement probe is restored to the state before
operation.
The inclination angle of wedge block is α, so the algebraic relation between
displacement x of probe and end deflection δ of cantilever beam is:
x = tan α · δ (24.2)
For the cantilever beam fixed at one end, σ max is the maximum stress at the joint of
the cantilever beam and the displacement probe, F is the resultant force perpendicular
to the cantilever beam at the contact of the wedge block and the end of the cantilever
beam, ε is the strain of fiber Bragg grating packed on the surface of cantilever beam.
Through basic principles of material mechanics:
Mx 6F I
σmax = = (24.3)
Wz bh 2
FI3
δ= (24.4)
3EI
σmax = E · ε (24.5)
where l is the effective working length of cantilever beam, b is the section breadth
of cantilever beam, h is the thickness of cantilever beam, E is the elastic modulus of
cantilever beam, and I is the sectional inertia moment of cantilever beam.
Using (24.2) to (24.5) can be obtained:
316 S. Li et al.
2 · tan α · I 2 · ε
x= (24.6)
3h
From the working principle of fiber Bragg grating, if the temperature is neglected,
the working principle expression of the fiber Bragg grating is as follows:
λB
= Kε · ε (24.7)
λB
with K ε = 1 − Pe , Pe is the valid elastic-optic constant of fiber Bragg grating, λB is

the Bragg central wavelength, λB is the central wavelength variation of fiber Bragg
grating.
Using (24.6) and (24.7) can be obtained:
λ8 3h · λ B · (1 − Pe )
= (24.8)
x 2 · tan α · I 2
Formula (24.8) is the optical wavelength-displacement equation for pitman style
fiber Bragg grating displacement sensor based on wedge cavity structure. In the
ideal state, sensitivity coefficient of the sensor is related with the valid elastic-optic
constant of fiber Bragg grating, the inclination angle of wedge block, the effective
working length of cantilever beam and the thickness of cantilever beam. When the
inclination angle of wedge block is changed, the ratio of the wavelength of the sensor
to the displacement of the probe is changed. Therefore, the sensitivity of the sensor
should be adjusted. When the temperature is constant, the change of displacement is
calculated by the change of central wavelength.
24.3 Sensor Calibration Test and Results
The test device adopts micrometer and multi-type synchronous demodulator. Multi-
type synchronous demodulator is developed by Dalian University of Technology,
which is suitable for simultaneous acquisition of fiber Bragg grating sensors and
electricity sensors. The demodulator has high-power wavelength scan laser in it and
each optical channel has an 80 nm wavelength range (1510–1590 nm). With the
demodulation accuracy of wavelength as 1 pm, the scan laser can be connected
to 240 fiber Bragg grating sensors. When the output function of the acquisition
module is built, the demodulator can realize the signal acquisition of velocity, force,
displacement, inclination, strain, temperature, pressure, etc. The central wavelength
of fiber Bragg grating in the test is 1552 nm, and the peak reflectivity is more than
90%.
Before the test, the strain gauge on the corresponding position of the cantilever
beam is used to compare the performance of the displacement sensor shown in
Fig. 24.3. In the elastic range, the strains of the cantilever beam, the fiber Bragg
grating and the resistance strain gauge are regarded as the same.
Fig. 24.3 Calibration test of FBG displacement sensor
According to the sensor performance test standards, three groups of tests are
completed. In the tests, the fiber Bragg grating for temperature compensation is
maintained at about 1552.233 nm. The maximum deviation is 2 pm. Therefore, it
can be assumed that the temperature does not affect the sensor during the whole
calibration test. According to the variation of the strain and the central wavelength
of the first calibration test, the relation between the central wavelength and the strain
is obtained under the forward and backward stroke, as shown in Figs. 24.4 and 24.5.
In the fitting curve of Figs. 24.4 and 24.5, it can be obtained the strain sensitivity
coefficient of pitman style displacement sensor is 1.1600 pm/με. From the literature
[9], it can be obtained the strain sensitivity coefficient is 1.2090 pm/με. When the
central wavelength of fiber Bragg grating is 1550 nm, the test value is similar to the
theoretical value.
24.3.1 Sensitivity Coefficient
The test data of 3 groups of forward and backward stroke are fitted, and the fitting
images are shown in Figs. 24.6 and 24.7. The fitting equations of forward stroke
test data of 3 groups are: y = 0.00560x + 1552.1222, y = 0.00564x + 1552.1216,
y = 0.00559x + 1552.1234, all the adjust R square are up to 0.99. The sensitivity
coefficients are 5.60 pm/mm, 5.64 pm/mm and 5.59 pm/mm. The fitting equations of
backward stroke test data of 3 groups are: y = 0.00551x + 1552.1238, y = 0.00556x
+ 1552.1233, y = 0.00557x + 1552.1233, whose adjust R square is up to 0.99.
The sensitivity coefficients are 5.51 pm/mm, 5.56 pm/mm and 5.57 pm/mm, respec-
tively. Therefore, the displacement sensitivity coefficient of displacement sensor is
5.58 pm/mm.
The theoretical displacement sensitivity coefficient of fiber Bragg grating dis-
placement sensor is 8.105 pm/mm. However, according to the test data fitting, the
318 S. Li et al.
Fig. 24.4 Strain-wavelength curve fitting
Fig. 24.5 Strain-wavelength curve fitting in forward stroke in backward stroke
displacement sensitivity coefficient of the displacement sensor is 5.58 pm/mm. Result

shows that the measured value of pitman style fiber Bragg grating displacement sen-
sor is 68.85% of the theoretical value.
Fig. 24.6 Displacement-wavelength curves
Fig. 24.7 Displacement-wavelength curves in forward stroke in backward stroke
24.3.2 Linearity
Linearity is also called “nonlinear error”, that is, the maximum deviation between
calibration curve and fitting line as a percentage of the full scale output. The smaller
the value, the better the linearity:
320 S. Li et al.
λmax
eL = ± × 100% (24.9)
yF S
where is λmax the maximum difference between arithmetic mean value and ref-
erence value on fitting line of central wavelength; FS y is the variation of central
wavelength in the full scale output, which is the mean value of central wavelength,
obtained in the 3 calibration tests.
After calculating the arithmetic mean of central wavelength and the displacement,
the fitting line is obtained as y = 0.00558x + 1552.1229. The fitting value of the
sensor wavelength is obtained, and the absolute value of the maximum deviation is
0.0037 nm. Then put the data into the Eq. (24.9) to get the linearity for the pitman
style fiber Bragg grating displacement sensor based on wedge cavity structure, which
is 2.434%.
24.3.3 Hysteresis
Hysteresis is an important factor to evaluate the static index of the sensor, to cause
the sensor accuracy unstable. In the calibration process, the same value of input is
used and the direction of travel is not the same. Although the input range is the same,
the output signal value is not the same. That is, the hysteresis is used to indicate the
non-coincidence degree between the input and output characteristic curves of the
sensor in the forward and backward stroke [10]. Hysteresis is also called “hysteresis
error”, mainly caused by the elastic hysteresis of elastic sensitive units, the friction
between components, the gap of transmission parts and the looseness of the fastening
pieces. It can be expressed as:
Hmax
eH = ± × 100% (24.10)
yF S
where Hmax is the deviation value of central wavelength in the course of 3 forward
and backward stroke tests; FS y is the variation of central wavelength in the full scale
output; max H is 0.0030 nm. The calculated hysteresis error of fiber Bragg grating
displacement sensor in this test is 1.974%.
24.3.4 Repeatability
Repeatability reflects the different levels of characteristic curves when the sensor
is inputted with the same value in different tests. When the input conditions are
consistent, with the higher similarity of the output characteristic curves after several
tests, the repeatability is better, and the repeatability error is relatively lower. The
repeatability error can get the discrete degree of calibration data, which represents a
random error. Therefore, it can be obtained the repeatability through calculating the
standard deviation. The standard deviation is calculated by using the range analysis
method or the Bessel method. Because of the number of the data collected in this
test, the Bessel method is better than the range analysis method.
From the Bessel formula:

n
(yi − ȳ)2
σ = i=1
(24.11)
n−1
The standard deviation of 3 forward and 3 backward stroke data can be obtained.
According to the equation of repeatability error:
ασmax
ez = ± × 100% (24.12)
yF S
24.3.5 Static Error
It can be concluded that the static error of the pitman style fiber Bragg grating
displacement sensor based on wedge cavity structure:

e S = e2L + e2H + ez2 = 5.168% (24.13)
24.4 Application of Fiber Bragg Grating Sensor
In the hysteresis test of steel frame-reinforced concrete infill wall, the sensor is
used to measure the one story of the shear wall’s lateral displacement and the data is
compared with the resistance strain displacement meter. In the steel frame-reinforced
concrete infill wall, the steel frames are made of HPB235 steel. Steel frame beams
are made of section steel H100 × 100 × 6 × 8, and steel frame columns are made
of section steel H125 × 125 × 6 × 8. The steel frame beams and the steel frame
columns are connected with four-angle steels by the bolts. The dimension of the one
story of the shear wall is 780 mm × 500 mm × 80 mm, and the concrete strength
grade is C25. The reinforcement in the concrete is made of HPB235 steel bars. The
concrete shear wall and the steel frame columns are connected by common bolts,
while the concrete shear wall and the steel frame beams are connected by SMA bolts.
The second story of the shear wall is the same with the one story of the shear wall.
The steel frame-reinforced concrete infill wall is connected with the loading
device-lifting jack. Fix the pitman style fiber Bragg grating displacement sensor
on the surface of the resistance strain displacement meter, paralleling to the resis-
tance strain displacement meter. A pull rope is fixed at the end of the displacement
322 S. Li et al.
Fig. 24.8 The installation diagram of FBG displacement sensor
probe of the displacement sensor. Pull the rope when the half range of the sensor is
reached, in this state, and the other end of the pull rope is fixed on the side of the
steel frame-reinforced concrete infill wall. The fiber Bragg grating sensor and the
resistance strain displacement meter are perpendicular to the steel frame-reinforced
concrete infill wall. The displacement probe of the resistance strain displacement
meter is attached to the side of the steel frame-reinforced concrete infill wall, as
shown in Fig. 24.8.
24.5 The Results and Analysis
In the hysteresis test of steel frame-reinforced concrete infill wall, the sustained
loading is applied manually by the lifting jack. Each 20 kN is used as a loading cycle,
pushing the steel frame-reinforced concrete infill wall in 20 kN, and then pulling the
steel frame-reinforced concrete infill wall in 20 kN until it attains the ultimate bearing
capacity. With controlling the lifting jack in displacement the hysteretic curve is
obtained. As the load is controlled by force, the central wavelength variation of the
multi-type synchronous demodulator is acquired in every two loading cycles. The
acquisition is finished until the load is controlled by displacement. The wavelength
of temperature-compensated fiber Bragg grating is stable in 1552.2785 nm, so the
effect of temperature on the test is ignored.
Figure 24.9 shows that the displacement comparison of the resistance strain meter
and the fiber Bragg grating displacement sensor under the same loadings, which
means the fiber Bragg grating displacement sensor and the resistance strain meter
are in good conformity and the change trend is accordable.
The test shows that the fiber Bragg grating displacement sensor can be used to
measure displacement accurately. Compared with the resistance strain meter, the
fiber Bragg grating displacement sensor has the advantages of simple structure and
anti-electromagnetic interference.
Fig. 24.9 Comparison of measured displacement between resistance strain displacement meter and
FBG displacement sensor
24.6 Conclusion
This paper developed a pitman style fiber Bragg grating displacement sensor based
on wedge cavity structure. The sensor has the advantages of simple structure and
anti-electromagnetic interference, meeting the needs of long term displacement mon-
itoring of engineering structures. Through the calibration test, it can be obtained the
sensitivity coefficient of pitman style fiber Bragg grating displacement sensor is
5.58 pm/mm; the adjust R square is up to 0.99 and the static error is 5.168%. The
calibration test shows that the design is reasonable. The developed fiber Bragg grat-
ing sensor is applied in the hysteresis test of steel frame-reinforced concrete infill
wall, successful monitoring the one story of the shear wall’s lateral displacement.
The test shows that the sensor is of high reliability, flexibly layout and is suitable for
measuring the displacement of engineering structures.
After the metal case of the fiber Bragg grating displacement sensor, it can be used
with the different installation methods such as sticking, welding, threaded connection,
etc. It is more convenient to arrange the displacement sensor on the construction site
and to be applied to the long term displacement monitoring of the harsh and complex
engineering structures.
Acknowledgements This work was supported by the National Natural Science Foundation
of China (Grant No. 51578347), Liaoning Natural Science Foundation of China (Grant No.
2015020578), and the Program supported by Characteristic Discipline and Talent Team of Taishan
Scholar.
324 S. Li et al.
References
1. Wang, X.L., Tang, D.Y., Zhu, S.S.: A new design of capacitive displacement sensor measuring
circuit. Comput. Meas. Control 18(7), 1701–1720 (2010)
2. Yu, J.P., Wang, W., Lu, K.Q., Mei, D.Q., Chen, Z.C.: A planar capacitive sensor for 2D long-
range displacement measurement. Front. Inf. Technol. Electron. Eng. 14(4), 252–275 (2013)
3. Matsuya, I., Matsumoto, F., Ihara, I.: Ultrasonic lateral displacement sensor for health moni-
toring in seismically isolated buildings. Sensors 15(7), 17000–17012 (2015)
4. Zazwani, N.R., Ibrahim, R.K.R., Musa, S.M.A., Hosseinian, S.R., Azmi, A.I., Ahmad, N.:
Reactor temperature profiles of non-thermal plasma reactor using fiber Bragg grating sensor.
Sens. Actuators A: Phys. 244, 206–212 (2016)
5. Sun, L., Chen, C., Sun, Q.Q.: Experimental and finite element analyses on the corrosion of
underground pipelines. Sci. China 58(6), 1015–1020 (2015)
6. Sun, L., Hao, H., Zhang, B., Ren, X.L., Li, J.: Strain transfer analysis of embedded fiber Bragg
grating strain sensor. J. Test. Eval. 44(6), 2312–2320 (2016)
7. Mao, J., Xu, F., Gao, Q., Liu, S., Jin, W.: A monitoring method based on FBG for concrete
corrosion cracking. Sensors 16(7), 1093–1098 (2016)
8. Mieloszyk, M., Ostachowicz, W.: An application of structural health monitoring system based
on FBG sensors to offshore wind turbine support structure model. Marine Struct. 51, 65–86
(2017)
9. Raw, Y.J., Jackson, D.A.: A prototype multiplexing system for use with a large number of fiber-
optic-based extrinsic Fabry-Perot sensors exploiting low coherence interrogation. Proc. SPIE
2507, 90–98 (1995)
10. Wang, X., Fu, Y.P.: Research of changing type sensor precision influence of stress of sluggish
error. Coal Mine Mach. 28(4), 102–104 (2007)

Robotics and Mechatronics: Richard (Chunhui) Yang Yukio Takeda Chunwei Zhang Gu Fang Editors

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Robotics and Mechatronics: Richard (Chunhui) Yang Yukio Takeda Chunwei Zhang Gu Fang Editors

Uploaded by

Copyright:

Available Formats

Mechanisms and Machine Science 72

Richard (Chunhui) Yang

More information about this series at http://www.springer.com/series/8779

Yukio Takeda Chunwei Zhang

Robotics and Mechatronics

ISSN 2211-0984 ISSN 2211-0992 (electronic)

1 Architecture Choice of a Robotic Hand for Deep-Sea Exploration

9 Foot Reaction Analysis of Whole Body Dynamic via Screw

21 Inﬂuences of Angular Position of Unbalanced Force on Crack

C. Mizera, M. A. Laribi, D. Degez, J. P. Gazeau, P. Vulliez and S. Zeghloul

Keywords Design · Robotic hand · Motion capture · Submarine · Archaeology

C. Mizera · M. A. Laribi (B) · D. Degez · J. P. Gazeau · P. Vulliez · S. Zeghloul

© Springer Nature Switzerland AG 2019 1

1.2 Analysis of the Archaeologist Gesture

1.2.1 Recording of the Motions

In order to have a better understanding of the submarine archaeology, and to design a

Fig. 1.1 Experimental site

Fig. 1.2 Position of the markers on the hand

1.2.2 Obtained Results

1.2.2.1 Hand Used as a Tool

Fig. 1.4 Range of abduction

Table 1.1 Maximal Joint Maximum rotational speed recorded (°/s)

1.2.2.2 Hand Used to Grasp Objects

1.2.3 Data Analysis

Fig. 1.7 Total workspace of the archaeologist fingertips

Table 1.2 Total angular range

workspace. It should also be able to be in direct opposition with each of the

Based on this analysis, we built a taxonomy of the submarine archaeology activity

1.3 Choice of the Hand Architecture

Table 1.3 Schemes of the possible hand architectures

1.3.1 Total Workspace of the Hand

Pa(i) ∈ W A, i = 1, . . . , n, n ∼ 20000 (1.1)

Ph(i) ∈ W H, i = 1, . . . , m, m ∼ 25000 (1.2)

Table 1.4 Dimensions and joint limits

Table 1.5 Workspaces of the fingers

1.3.2 Ability to Make a Stable Grasp

dk = R for k ∈ [Finger 1, Finger 2, Finger 3, Thumb] (1.5)

Fig. 1.9 Contact between

αmin > 0 (1.8)

Fig. 1.10 Architecture and parameters of the robotic hand

1.4 Conclusion and Future Work

Annex: Taxonomy of the Submarine Archaeology Task

3 fingers 2 fingers 1 finger

Packaging Medium wrap

C. Y. Weng and I. M. Chen

Abstract This paper aims to provide a vision-based self-adaptive strategy to cope

Keywords Computer vision · Robotic assembly · Intelligent system

Based on the fulfillment of sustainability, the demands of industrial robot technol-

C. Y. Weng · I. M. Chen (B)

© Springer Nature Switzerland AG 2019 21

with a simpler and cost-effective measure and peripheral equipment to implement a

2.3 Methodologies and Results

Based on the assumptions made previously, we design the vision-based strategies

5. Determine how to pick up the charging terminal.

2.3.1 Global Detection

where [X L , X U ], [Y L , YU ], [Z L , Z U ] are the lower and upper bounds of x-axis,

3. Remove noise by the radius outlier removal filter:

where Pk is the rough position of cluster Ck ; n k is the number of points in cluster

Fig. 2.1 Results of global detection

2.3.2 Local Detection

Fig. 2.2 Results of local detection

2.3.3 Error Detection

Fig. 2.3 Scenario of error detection