PHD Thesis Rajiv Vashisht Submitted
PHD Thesis Rajiv Vashisht Submitted
PHD Thesis Rajiv Vashisht Submitted
by
Doctor of Philosophy
Rotors are extensively used to transfer energy from one part to other parts of a mechanical
system. During the service life of the rotors, they are subjected to different faults like the rotor
crack, coupling misalignment, rotor bow, rotor-stator rubbing and unbalance. One of the important
aspects of these faults is their interconnection. Presence of one fault can accelerate the growth of
the other. If these faults are not identified timely, costly breakdown of the rotor system can occur.
The presence of these faults changes the overall nature of the system dynamics from the linear to
nonlinear one. This aspect can be used to identify the presence of these faults. Various system
components like ball bearings and journal bearings also make the overall dynamics of a healthy
rotor system nonlinear. Hence, it is not easy to detect the presence of faults based on the vibration
signal. Transient response-based crack detection technique has been developed and verified using
simulation techniques.
Rotor dynamics with transverse crack, coupling miss-alignment and rotor-stator rubbing can
intermittently varying coefficients. A new fault detection strategy based on short time Fourier
transformation has been developed to detect the presence of crack and rotor stator rubbing even if
the both coexist simultaneously. Due to the presence of unbalance, rotor systems with constant
spinning speed are subjected to steady-state vibrations that can accelerate the crack growth or
chances of rotor/stator rubbing action. During changes of the spinning speed, transient vibrations
ii | P a g e
generate required control forces. From the spectrum of the control signal, rotor faults like
Applications of rotor dynamics can also be seen in machine tools. The spindle and ball screw
drives are important examples of their applications. In the spindle, transverse and torsional
vibrations are produced. However, in the ball screw drives, axial vibrations are generated. For
precision control of the drives, a hybrid controller is proposed to increase command tracking
efficiency and reduce the dynamic deflection of the carriage during metal cutting process.
Experimental verification of the proposed controller has been done to verify its effectiveness. The
dynamics of cutting tools is represented in the form of time-delay systems. Different time-domain
algorithms are used to identify stable regions of machining operations. To improve the depth of
cut, the active chatter control of turning and boring operations is carried out. Experimental results
verify the effectiveness of the proposed active chatter control strategies. Based on the relative
motion between the workpiece and cutting tool (during boring operation), cutting operations can
be divided into different categories. The dynamics of such systems is represented using delay
differential and time-periodic delay differential equations. Feasibility of active chatter control for
each category is analyzed in detail. The effect of eccentricity and workpiece ovality on the
are proposed to compensate for the changing dynamics of the system due to continuous metal
results verify that these controllers save the control energy in active chatter control applications.
Machine learning techniques are used for online chatter detections in milling operations.
Various chatter indicators are proposed. Improved dynamic equations are proposed for milling
iii | P a g e
operations to find the behaviour of vibrations during chatter conditions. Simulated data are used
to train Convolutional Neural Networks that are then used to detect chatter in an online manner.
iv | P a g e
Acknowledgments
I would like to take this opportunity to express my sincerest thanks to my supervisor, Professor
Qingjin Peng, for his motivation, academic support, and guidance through this work. Without his
support, this work would not have been possibly done. Many thanks are also directed to the
advisory committee members, Professors Nan Wu and Babak Mehran for their constructive advice.
This research has been supported by the University of Manitoba Graduate Fellowship (UMGF)
and the Discovery Grants from the Natural Sciences and Engineering Research Council (NSERC)
of Canada. I would also like to express my appreciation to awards from the Faculty of Graduate
Edward R. Toporeck Graduate Fellowship in Engineering for years 2017 and 2018
Faculty of Graduate Studies research complete scholarship for year 2019
Philip and Marjorie Eckman Scholarship in Engineering for year 2020
Institute of Technology, New Delhi, India for his kind support in conducting the experimental
work in his laboratory. I would also like to thank my dear friends in the Indian Institute of
I would also like to thank my dear friends in Virtual Manufacturing Laboratory at the
v|Page
Dedication
vi | P a g e
Table of Contents
Abstract .......................................................................................................................................... ii
Acknowledgments ......................................................................................................................... v
Dedication ..................................................................................................................................... vi
Table of Contents ........................................................................................................................ vii
List of Tables ................................................................................................................................. x
List of Figures ............................................................................................................................... xi
List of Abbreviations .................................................................................................................. xv
Copyright Notices...................................................................................................................... xvii
1. Introduction ........................................................................................................................... 1
1.1. Introduction to rotor dynamic systems ................................................................................. 1
1.2. Faults in rotor dynamic systems ........................................................................................... 1
1.3. Active vibration control of rotor dynamic systems .............................................................. 3
1.4. Applications of rotor dynamic systems in machine tools .................................................... 4
1.5. Research problems……………………………………………………………………… . 7
1.6. Research objectives and methods ........................................................................................ 8
1.7. Connectivity of various parts of the dissertation .................................................................. 9
1.8. Structure of the dissertation................................................................................................ 10
2. Literature Review ................................................................................................................ 13
2.1. Overview ............................................................................................................................ 13
2.2. Faults in rotor dynamic systems ......................................................................................... 13
2.3. Active vibration control of rotor dynamic systems ............................................................ 17
2.4. Applications of rotor dynamic systems in machine tools .................................................. 19
2.5. Research gaps and directions ............................................................................................. 28
3. Modelling and fault detection in rotodynamic systems .................................................... 30
3.1. Overview ............................................................................................................................ 30
3.2. Mathematical Modelling .................................................................................................... 31
3.3. Conventional crack detection techniques based on the steady-state dynamic response .... 45
3.4. Proposed Crack Detection Technique based on Transient Response ............................... 45
vii | P a g e
3.5. Active Vibration Control using Robust Control Methodology .......................................... 49
3.6. Switching Control Strategy ................................................................................................ 50
3.7. Application of Switching Control and STFT based Spectrum ........................................... 54
3.8. Concluding remarks ........................................................................................................... 55
4. Modelling and detection of rotor stator rubbing action .................................................. 57
4.1. Overview ............................................................................................................................ 57
4.2. Mathematical Modelling .................................................................................................... 58
4.3. Simulation model ............................................................................................................... 63
4.4. Detection of transverse crack and rubbing action .............................................................. 64
4.5. Effect of different factors on rotor/stator rubbing action ................................................... 65
4.6. Concluding remarks ........................................................................................................... 72
5. Active vibration control of rotor dynamic systems .......................................................... 74
5.1. Overview ............................................................................................................................ 74
5.2. Mathematical modelling of a rotor system ......................................................................... 74
5.3. Mu-synthesis using D-K iterations ..................................................................................... 78
5.4. Conventional Adaptive Feedforward Control with an internal estimation of frequency ... 79
5.5. External Frequency estimation ........................................................................................... 81
5.6. Results with conventional Adaptive Feedforward controller ............................................. 82
5.7. Proposed Hybrid controller with an external estimation of frequency .............................. 85
5.8. Experiment ......................................................................................................................... 90
5.9. Results ................................................................................................................................ 92
5.10. Concluding remarks ......................................................................................................... 96
6. Precision control of ball screw drives ................................................................................ 97
6.1. Overview ............................................................................................................................ 97
6.2. Mathematical Modelling .................................................................................................... 98
6.3. Design and Implementation of Hybrid Controller ........................................................... 104
6.4. Experiment ....................................................................................................................... 107
6.5. Theoretical and experimental results................................................................................ 108
6.6. Concluding remarks ......................................................................................................... 116
7. Active chatter control of turning operations ................................................................... 118
7.1. Overview .......................................................................................................................... 118
7.2. Active chatter control using an electromagnetic actuator ................................................ 118
viii | P a g e
7.3. Dynamic modelling for turning of the stepped shaft........................................................ 121
7.4. Linear parameter varying controller design ..................................................................... 134
7.5. Results and Discussion ..................................................................................................... 134
7.6. Concluding remarks ......................................................................................................... 138
8. Active chatter control of boring operations .................................................................... 139
8.1. Overview .......................................................................................................................... 139
8.2. Mathematical Modelling of stationary bar boring operation ........................................... 140
8.3 Mathematical modelling of rotating bar boring operation ................................................ 146
8.4. Simulation model and formulation of delay equations .................................................... 147
8.5. Active chatter control of stationary bar boring operation ................................................ 150
8.6. Active chatter control for rotating bar boring operation .................................................. 162
8.7. Concluding remarks ......................................................................................................... 166
9. Chatter detection in milling operations using Neural Networks ................................... 167
9.1. Overview .......................................................................................................................... 167
9.2. Dynamic modelling of milling operations ....................................................................... 167
9.3. Comparison of simulation results with literature and experimental results ..................... 171
9.4 The amplitude of tool vibrations and tool forces during chatter ....................................... 178
9.5. Presence of multiple chatter frequencies ......................................................................... 180
9.6. Chatter indicators ............................................................................................................. 182
9.7. Online chatter identification based on Convolutional Neural Networks ......................... 190
9.8. Concluding remarks ......................................................................................................... 193
10. Research conclusions, contributions and future recommendations .......................... 195
10.1. Conclusions .................................................................................................................... 195
10.2 Research contributions .................................................................................................... 198
10.3 Further recommendations ................................................................................................ 199
Papers published/submitted from dissertation ...................................................................... 201
Bibliography .............................................................................................................................. 203
Appendix A ..................................................................................................................................... i
ix | P a g e
List of Tables
Table 3-1 Physical parameters and mechanical properties of the rotor bearing system ............... 34
Table 3-2 Damping parameters of the discrete mass representation of rotor bearing system ...... 35
Table 3-3 The comparison of natural frequencies ........................................................................ 39
Table 3-4 Geometric properties of 6014 Ball Bearing (Tiwari and Gupta, 2000) ........................ 43
Table 5-1 The physical parameters and mechanical properties of the rotor bearing system ........ 78
Table 5-2 Physical Parameters of Electromagnetic actuator ........................................................ 92
Table 6-1 Important parameters of the ball screw drive ............................................................. 102
Table 7-1 Physical Parameters of Electromagnetic actuator used .............................................. 120
Table 7-2 Parameters of the flexible shaft .................................................................................. 122
Table 8-1 Physical Parameters of Electromagnetic actuator ...................................................... 151
Table 8-2 Comparison of theoretical and experimental results .................................................. 152
Table 9-1 Confusion matrix for STFT based detection………………………………………...190
Table 9-2 Confusion matrix for tool locus-based detection…………………………………….191
x|Page
List of Figures
Figure 1-1 Rotor with crack (Lalanne and Ferraris, 1990) ............................................................. 2
Figure 1-2 Rotor/stator rubbing ...................................................................................................... 2
Figure 1-3 Schematic of the experimental setup for AMB ............................................................. 4
Figure 1-4 Conventional applications of rotor dynamic systems and applications in machine tools
......................................................................................................................................................... 5
Figure 1-5 Schematic diagram of the ball screw drive ................................................................... 5
Figure 1-6 Conventional turning operation..................................................................................... 6
Figure 1-7 Active chatter control in boring operation .................................................................... 7
Figure 1-8 Connectivity of different parts of the dissertation....................................................... 10
Figure 3-1 Schematic of the finite element model of a rotor-bearing system .............................. 33
Figure 3-2 Equivalent physical model of the discrete mass model of the rotor bearing system in
one plane (not to scale) ................................................................................................................. 36
Figure 3-3 Phenomenological model of the discrete mass model of the rotor bearing system in
one plane (not to scale) ................................................................................................................. 37
Figure 3-4 Bode plot for the different modeling techniques at Disk1 and 2 ................................ 39
Figure 3-5 Ball bearing system ..................................................................................................... 41
Figure 3-6 Cracked Rotor model .................................................................................................. 42
Figure 3-7 Horizontal and vertical responses of the rotor bearing (Healthy and Cracked) system
at disks 1 and 2.............................................................................................................................. 46
Figure 3-8 Spectrogram of the rotor-bearing system with rigid bearing support (at disk2) ......... 47
Figure 3-9 Spectrogram of the rotor Ball Bearing system with flexible bearing support (at disk1)
....................................................................................................................................................... 48
Figure 3-10 Open and closed-loop responses of the 40% Cracked rotor with the robust controller
....................................................................................................................................................... 49
Figure 3-11 Spectrogram of the healthy rotor and 10% cracked rotor with ball bearings at disk 1
(closed-loop response) (in region II)............................................................................................. 50
Figure 3-12 Frequency domain response of the rotor ball bearing system at disk 1 (in region II)
using negative velocity feedback controller as a function of crack severity ................................ 51
Figure 3-13 Architecture of the switched control ......................................................................... 52
Figure 3-14 Response of the 10% cracked flexible rotor with active and switching control ....... 53
Figure 3-15 Spectrogram of the healthy rotor and 5% cracked rotor with ball bearings at disk 1
(closed-loop response) (in region II) using switching control strategy ........................................ 54
Figure 3-16 Spectrogram of the healthy rotor and 30% cracked rotor with ball bearing at disk 1
(closed loop response) (in region II) ............................................................................................. 55
Figure 4-1 Schematic of the 2-DOF rotor stator system (Front view) .......................................... 60
Figure 4-2 Schematic of the 2-DOF rotor stator system (side view) ............................................ 60
Figure 4-3 Cracked rotor model .................................................................................................... 61
Figure 4-4 Comparison of vibration characteristics at different rotor spin frequencies for vertical
and horizontal rotors ..................................................................................................................... 65
xi | P a g e
Figure 4-5 Effect of gravity and presence of crack on vibration characteristics of the horizontal
rotor (spin frequencies above the resonance)................................................................................ 67
Figure 4-6 Effect of gravity and presence of crack on vibration characteristics of the horizontal
rotor (spin frequency below the resonance) .................................................................................. 67
Figure 4-7 Vibration characteristics of the vertical rotor during coasting up as varied clearance
amplitudes ..................................................................................................................................... 69
Figure 4-8 Vibration characteristics of the horizontal rotor during coasting up as varied clearance
amplitudes ..................................................................................................................................... 69
Figure 4-9 Vibration characteristics of the cracked horizontal rotor during coasting up as varied
clearance amplitudes (part 1) ........................................................................................................ 71
Figure 4-10 Vibration characteristics of the cracked horizontal rotor during coasting up as varied
clearance amplitudes (part 2) ........................................................................................................ 71
Figure 5-1 Rotor/bearing/coupling system ................................................................................... 76
Figure 5-2 Comparison of conventional and proposed control architecture................................. 83
Figure 5-3 Time domain response by conventional Adaptive Feedforward Control at 20 Hz
Spinning speed .............................................................................................................................. 83
Figure 5-4 Time domain response by Adaptive Feedforward Control with time varying frequency
component applied to rotor system ............................................................................................... 84
Figure 5-5 Comparison of actual and estimated frequencies by Adaptive Notch Filter at sampling
frequency of 200 Hz...................................................................................................................... 84
Figure 5-6 Flow chart for proposed adaptive feedforward controller........................................... 87
Figure 5-7 Frequency response function of different systems ...................................................... 88
Figure 5-8 Robust stability, performance and other parameters ................................................... 89
Figure 5-9 Multi-harmonic hybrid control architect ..................................................................... 90
Figure 5-10 Experimental setup for active vibration control for rotor system ............................. 91
Figure 5-11 Comparison of Mu- controller and Derivative controller for Disk 2 ........................ 93
Figure 5-12 Response of rotor system with the proposed hybrid controller (Experimental) ....... 94
Figure 5-13 The response of the rotor system with coupling misalignment using proposed
controller (Experimental) .............................................................................................................. 95
Figure 6-1 Two-dimensional representation of the ball screw drive ............................................ 99
Figure 6-2 Simplified model of the ball screw drive .................................................................. 100
Figure 6-3 Effect of location and mass of the carriage on the first natural frequency (with bearing
clearance) .................................................................................................................................... 103
Figure 6-4 Simulink implementation of Hybrid controller ......................................................... 107
Figure 6-5 Experimental setup for ball screw drive.................................................................... 108
Figure 6-6 Loop sensitivity with different controllers ................................................................ 109
Figure 6-7 Loop magnitude with different controllers and loop shaping factor ......................... 110
Figure 6-8 Time domain analysis for command tracking with different controllers (Simulation
results) ......................................................................................................................................... 112
Figure 6-9 Performance of various controllers during metal cutting (Experimental results) ..... 114
Figure 6-10 Comparison of performance with LPV and other controllers (Simulation results) 116
xii | P a g e
Figure 7-1 Active chatter control architecture for flexible stepped shaft ................................... 119
Figure 7-2 Schematic of the overall control system for turning operation ................................. 121
Figure 7-3 Mode shapes of flexible stepped shaft ...................................................................... 123
Figure 7-4 Natural frequencies of the flexible stepped shaft as a function of tool location and
pass number ................................................................................................................................ 124
Figure 7-5 Stiffness and critical depth of cut as a function of tool location for the stepped flexible
shaft ............................................................................................................................................. 125
Figure 7-6 Nomenclature and schematic of the cutting tool ....................................................... 126
Figure 7-7 Stability lobe diagram as a function of tool location ................................................ 130
Figure 7-8 First natural frequency as a function of the number of cutting passes and tool location
..................................................................................................................................................... 131
Figure 7-9 Frequency domain presentation of the perturbed system with LPV controller ........ 135
Figure 7-10 Time-domain performance of the perturbed system (the 10th pass of the tool) with
the nominal H-inf and LPV controller ........................................................................................ 136
Figure 7-11 Performance of the perturbed system with a gain scheduled/LPV controller with
sensor and actuator in the loop.................................................................................................... 137
Figure 7-12 Performance of the perturbed system with a gain-scheduled/LPV controller with
sensor and actuator in the loop (feedrate 400 microns/second) .................................................. 137
Figure 8-1 Different types of boring operations ......................................................................... 139
Figure 8-2 Schematic of the overall boring operation ................................................................ 142
Figure 8-3 Analysis of ovality and eccentricity .......................................................................... 144
Figure 8-4 Forces during revolving bar boring (not to scale) ..................................................... 146
Figure 8-5 Physical schematic for active chatter control ............................................................ 150
Figure 8-6 Design parameters and frequency response of the loop shaping design procedure
controller ..................................................................................................................................... 152
Figure 8-7 Simulink presentation of the control system ............................................................. 155
Figure 8-8 Flowchart for fractional order PDλ control system synthesis.................................... 156
Figure 8-9 Closed loop response of the boring process with fractional order PD controller
(λ=0.2) (simulations)................................................................................................................... 157
Figure 8-10 Comparison of the semi-discretization approach and Pseudo spectral approach ... 158
Figure 8-11 Open-loop Stability lobe diagram for boring without and with ovality .................. 159
Figure 8-12 Closed loop response the boring process with eccentricity using PD controller
(experimental) ............................................................................................................................. 160
Figure 8-13 Performance of Fractional order PDλ controller at 20 microns/rev feed rate
(experimental) ............................................................................................................................. 161
Figure 8-14 Open-loop Stability lobe diagram for boring with revolving bar ........................... 163
Figure 8-15 Closed loop response of the boring operation (with revolving bar) using PD control
..................................................................................................................................................... 164
Figure 8-16 Closed loop response of the boring operation (with revolving bar) using PDλ
controller ..................................................................................................................................... 164
xiii | P a g e
Figure 8-17 Closed-loop response of the boring operation using switching control and time-
varying feedback (with revolving bar) ........................................................................................ 165
Figure 9-1 Geometric parameter representation of the milling tool ........................................... 169
Figure 9-2 Geometric parameters of the cutting surface when the tool leaves the workpiece ... 169
Figure 9-3 Schematic of the milling process with corresponding sensors .................................. 173
Figure 9-4 Stability lobe diagram for a specified range ............................................................. 175
Figure 9-5 Comparison of simulation results with experimental results at 3300 rpm ................ 176
Figure 9-6 Comparison of simulation results with experimental results at 3650 rpm. ............... 177
Figure 9-7 Stability lobe diagram for the system under analysis................................................ 178
Figure 9-8 Amplitude of vibration (in x-direction i.e. feed direction) at a different spinning speed
of milling tool ............................................................................................................................. 179
Figure 9-9 Chatter frequencies as a function of spinning speed of milling cutter ...................... 182
Figure 9-10 Instantaneous frequencies as a function spinning speed of milling tool (axial depth of
cut is 4 mm) ................................................................................................................................ 184
Figure 9-11 Effect of noise on instantaneous frequencies .......................................................... 185
Figure 9-12 STFT as a function of spinning speed of milling tool (at 4 mm axial depth of cut) 186
Figure 9-13 Effect of noise on power spectral density ............................................................... 187
Figure 9-14 Effect of noise on Spectrogram ............................................................................... 187
Figure 9-15 Milling tool locus as a function spinning speed of milling tool (noise free cases) . 189
Figure 9-16 Effect of noise on locus of milling tool at different spinning speeds...................... 189
Figure 9-17 Architecture of convolutional Neural Networks ..................................................... 190
Figure 9-18 Training points for simulation assisted deep learning............................................. 191
xiv | P a g e
List of Abbreviations
xv | P a g e
MHG Multiple harmonics generation
MP Multi period
AFFC Adaptive feedforward controller
DFT Discrete Fourier transformation
EA Electromagnetic Actuator
VA Voltage amplifier
IGSLS Interpolating gain scheduled loop shaping
LSDP Loop shaping design procedure
ODE Ordinary differential equations
FOPD Fractional order proportional derivative
PS Pseudo spectral
CNN Convolutional neural network
DWT Discrete wavelet transformations
SLD Stability lobe diagrams
xvi | P a g e
Copyright Notices
2. With kind permission from the American Society of Mechanical Engineering (ASME):
Rajiv Kumar Vashisht, Qingjin Peng, Nonlinear dynamic modelling of the cracked rotor
ball bearing system with emphasis on damage detection capabilities, 2018, ASME
Publication, Journal of Vibration and Acoustics, 140, pp 041018-10.
Rajiv Kumar Vashisht, Model based determination of size and location of a crack in a rotor
system using wavelet transformation and pattern search optimization. ASME 2019
International Design Engineering Technical Conferences and Computers and Information
in Engineering Conference, IDETC2019-97072.
Rajiv Kumar Vashisht, Qingjin Peng. Dynamic modelling and diagnosis of transverse
crack and rotor/stator rub in a flexible rotor system. ASME 2019 International Design
Engineering Technical Conferences and Computers and Information in Engineering
Conference, DETC2019-97047.
Rajiv Kumar Vashisht, The effect of gravity and rotor configuration on dry friction whip
and safe rotor/stator clearance. ASME 2019 International Design Engineering Technical
Conferences and Computers and Information in Engineering Conference, DETC2019-
97048.
xvii | P a g e
4. With kind permission from Springer Publishers:
Rajiv Kumar Vashisht, Qingjin Peng. Improving the cutting rate in turning operations of
flexible stepped shafts with a linear parameter varying controller using the electromagnetic
actuator, 2019, Springer Publication, International Journal of Advanced Manufacturing
Technology,104, pp. 1-16.
Rajiv Kumar Vashisht, Qingjin Peng. Feasibility analysis of active chatter control for
stationary and revolving bar boring operations based on magnitude of control forces using
Fractional order PD-λ controller. 2020, Springer Publication, International journal of
Advanced Manufacturing Technology, 106, pp.3957–3974.
xviii | P a g e
Chapter 1
1. Introduction
Rotors are machine elements transferring torque and power from one component to the other.
They are integrated with components like bearings, couplings, belts and pulleys. The rotor in
combination with other components is called a rotor system. Rotors are subjected to torque and
bending moments either constant in magnitude or time-varying. Due to the flexibility of rotors,
they are subjected to vibrations. To find the rotor response at a particular point, modelling of the
dynamic rotor system is necessary. Analytical, semi-analytical and numerical techniques are
generally used to set up dynamic governing equations of motion of the system. Normally, the
dynamics of healthy rotor systems can be represented using ordinary differential equations.
Due to the cyclic nature of the applied torque or bending moment, certain types of cracks are
developed in rotors. Dynamic torque produces ‘Slant’ cracks while dynamic bending moments
cause ‘Transverse’ cracks. Rotor bearing systems are generally encountered with ‘Transverse’
cracks. With the passage of time, these cracks propagate and pose other instability problems as
shown in Figure 1.1. This is the first type of faults in rotating machinery. However, there can be
other faults like rotor/stator rub, coupling misalignment, rotor bow and bearing malfunctions. For
the purpose of condition monitoring, different techniques are applied to detect the presence of
various faults. However, vibration signal-based condition monitoring is best suitable due to its
online nature. Due to the presence of these faults, the dynamics of rotor systems become nonlinear.
1|Page
In the presence of multiple faults, it becomes difficult to detect the type of faults. The dynamics
of the cracked rotor can be represented using ordinary differential equations with time-periodic
coefficients.
The clearance between the rotor and stationary housing or stator should be reduced to increase
the efficiency and leakage prevention in rotating machinery. Unbalanced masses are unavoidable
in the rotating machinery. This raises the chances of contacts between the rotor and housing
(stator), which is called rotor-stator rubbing. This is the second main fault in rotating machinery
as shown in Figure 1.2. Depending on the type and severity of rubbing, the amount of damage can
Flexible rotor
Rigid disc
maintained actions. In the first kind, the contact disappears with the passage of time. In the
2|Page
divergent mode (second kind), the amplitude of vibration increases constantly with time. In the
self-maintained mode, the contact does not disappear with time. When the clearance between the
rotor and stator is fixed and the rotor spins at different frequencies, different characteristics are
observed. There is a need for early detection of rotor/stator rub if it occurs. Dynamics of the rotor
in the presence of rub must be thoroughly studied. The dynamics of the rotor system under these
conditions is represented as a piecewise linear or hybrid system. There is a strong need to develop
rub identification strategies based on the vibration response of the rotor dynamic system.
For safety as well as service life considerations, it is important to reduce the vibration level of
any rotor-bearing system. The most common cause of the rotor vibrations is the unbalanced mass
which remains up to a certain extent even with precision balancing. The application of damping at
rotor supports acts, as a mean of reducing the rotor vibration. Active as well as passive damping
techniques have been reported in the literature to reduce the vibration level. In passive techniques
flexible damped supports, squeeze-film dampers and viscoelastic bearing supports are common.
Active magnetic bearing (AMB) supports a load using the magnetic levitation. AMB supports
moving parts without physical contacts as shown in Figure 1.3. For instance, the moving parts are
able to levitate a rotating shaft and permit relative motion with very low friction and no mechanical
wear. Magnetic bearings support the highest speed of all kinds of bearings. Apart from supporting
the load, they reduce rotor vibrations to some extent. In these conditions, they work like
electromagnetic actuators.
3|Page
AMB
Ball bearing Ball bearing
Eddy current sensor
Crack
Flexible rotor
Rigid disc
refinement, machine tool operation and natural gas handling. They are also used in the Zippe-type
centrifuge, for uranium enrichment and in turbomolecular pumps, where oil-lubricated bearings
would be a source of contamination. In case of the active vibration control, transient as well as
Flexible rotors are also applied in machining operations. Principles of rotor dynamics are
applicable for modelling and design of these systems (Figure 1.4). The dynamics of these systems
4|Page
Figure 1.4 Conventional applications of rotor dynamic systems and applications in machine
tools
1.4.1. Applications in ball screw drives of Computer Numerical Control (CNC) machines
CNC machines are extensively used in the manufacturing industry. Ball screw drives (BSD)
and spindle units are the main components of CNC machines. BSD is used to convert the rotary
motion to the linear motion. Due to the finite stiffness of ball bearings and driving shafts, axial
vibrations reduce the command tracking efficiency of BSD (Figure 1.5). When designing
controllers for this application, the axial flexibility of the driving shaft must be considered.
5|Page
1.4.2. Active chatter control in turning operations
While turning flexible shafts, the flexibility of the cutting tool and shaft (i.e. workpiece) plays
an important role (Figure 1.6). Due to this flexibility, for certain combinations of spinning speed
of the shaft and depth of cut, self-excited vibrations of the tool and workpieces are produced. This
phenomenon is called chatter. It can significantly reduce the surface finish of the workpiece and
reduce the useful life of the cutting tool. Hence active as well passive chatter prevention techniques
Chatter vibrations reduce the depth of cut at a particular spindle speed of machine tools
and hence the machining efficiency. Figure 1.7 shows an application of electromagnetic actuators
for actively controlling self-excited vibrations produced during machining in boring operations.
The dynamics of the boring operation can be represented using ordinary differential equations with
time delay. If the eccentricity or ovality is taken into consideration, these systems are changed into
6|Page
Holder Turret
Workpiece Electromagnetic
actuator
Boring bar
Electromagnetic
actuator
Sliding bearing
Spindle
Lathe bed
In milling operations, due to the intermittent nature of cutting forces, forced vibrations are
always present in the system. Suitable online techniques are desirable to detect the presence of
chatter. Once chatter is detected, suitable measures can be taken to move the system into the
chatter-free zone from chatter conditions. In the past, different online and offline chatter detection
I. There is a lack of improved techniques for the modelling of transverse rotor crack and
robust online crack detection strategies in case of rotor systems. The multiple harmonics
are generated in the presence of rotor faults when the rotor is spinning at a particular rotor
spin frequency. It becomes difficult to pin point the type of the fault based on these steady
state vibrations. There are no techniques available to differentiate between rotor crack and
7|Page
other rotor faults (like rotor/stator rubbing action and coupling mis-alignment) based on
II. It is known that feedback controllers are partially effective in suppression of harmonic
steady state vibrations in rotor systems, but there are very few control design techniques
III. High control forces are required to impart active damping in case of active chatter control
applications. There are no active chatter suppression techniques which require small size
IV. There are very less online chatter detection strategies available for milling operation to
increase the productivity. Machine learning techniques are not applied for these types of
applications.
V. There is a lack of applying concepts of rotor dynamics to machine tools and metal cutting
dynamics for the precision control of machine operations and improvement of machining
efficiency. There is a lack of control design strategies for robust, energy efficient and
precision control of machine tools (like ball screw drive unit of the CNC machines).
detection strategy for rotor systems based on vibration signals. A switching control strategy
in combination with short time Fourier transform (STFT) will be investigated for the crack
detection.
8|Page
II. Development of a hybrid controller that combines feedback and feed forward controls to
reduce the transient and steady state rotor vibrations. Mu-synthesis based robust control
technique will be applied to obtain the robust stability and robust performance.
III. Controllers based on H-infinity control theory, interpolation of controllers at the grid points
(i.e. Linear Parameter Varying controller) and fractional calculus will be developed for
IV. Development of an online chatter detection strategy using machine learning techniques to
V. Development of high precision and energy efficient control of machine tools (ball screw
Figure 1.8 shows the connectivity of different parts of the dissertation. The modelling, fault
detection and active vibration control of rotor systems were investigated under classical
applications of rotor dynamic systems. These applications are high-end applications and can be
seen in the power generation, jet engines and other aerospace structures. However, applications of
rotor dynamic systems can also be found in machine tools. The spinning cutting tool of a milling
or boring machine represents a rotor dynamic system with cutting forces acting on the tip of an
over-hanging rotor. The dynamics of such a rotor system can be represented with time-delay
differential equations. To enhance the material removal rate during metal cutting operations, active
damping needs to be imparted in the form of active chatter control. The control forces are provided
using electromagnetic actuators. Apart from transverse vibrations, sometimes axial vibrations are
also generated in rotor systems. The ball screw drive of CNC machines is an important example
9|Page
in this regard. For precise command tracking, the active vibration control of these resonant axial
modes is required.
Modelling and
detection of rotor/
stator rubbing
Active vibration
Modelling and control of steady
detection of crack state and transient
in rotor systems rotor vibrations
Online chatter
Rotor dynamics
Precise control of
detection in + Ball Screw Drives
milling
Active control
Active chatter
Active chatter
control of boring
control in turning
operation
operation
The dissertation contains 10 chapters. Chapter 1 introduces the research background and
Chapter 2 discusses the literature survey on various aspects of rotor dynamics and applications in
machine tools. The research directions based on the literature review are then presented.
10 | P a g e
Chapter 3 introduces modelling of rotor dynamic systems along with various faults like transverse
Chapter 4 presents modelling of rotor systems in rotor/stator rubbing actions. Various signal
processing techniques are investigated to detect the presence of rotor/stator rubbing actions based
Chapter 5 presents the active vibration control of rotor systems for coupling faults as well as rotor
cracks. The amplitude of vibrations is reduced in case of transient and steady-state vibrations. The
Mu-synthesis based robust control technique is proposed along with the feed-forward cancellation
technique.
Chapter 6 discusses the hybrid (LPV+PPI) control of ball screw drives of CNC machines. The
Chapter 7 describes the active chatter control of turning operations. An LPV controller is
developed to obtain robust stability for the total span of the shaft.
Chapter 8 introduces the active chatter control of the boring process using fractional order PD
controllers. It can significantly reduce control forces compared to robust controllers. Small-sized
electromagnetic actuators are proposed to maintain closed-loop stability. The effect of eccentricity
and ovality on the stability lobe diagrams is investigated. Feasibility of active chatter controls for
Chapter 9 presents the online chatter detection in milling operations using machine learning
techniques. An automatic spindle speed adjustment algorithm is designed and implemented to drag
11 | P a g e
Chapter 10 summarizes this research and outlines possible extensions for further research.
12 | P a g e
Chapter 2
2. Literature Review
2.1. Overview
Rotors are used to transmit the mechanical energy from one part of the system to the other. A
rotor system consists of bearings and couplings along with rotors. In this chapter, literature is
reviewed regarding different faults in rotor systems, active vibration control of rotor systems and
applications of rotor systems in machine tools. Gaps in the literature are presented with research
In the last three decades, researchers have developed different crack detection techniques. Chen
and Zhang (2011) presented a comprehensive introduction to this subject. Wauer (1990), Gasch
(1993) and Dimarogonas (1996) analyzed the nonlinear dynamics of rotor systems with fatigue
cracks. It was observed that nonlinear vibrations are produced in rotors in the presence of a
transverse crack. The permanently open cracks represent asymmetrical rotor and pose stability
problems. These types of cracks are generally found in rotors with the low weight. Under the
weight dominance, the crack opens as well as closes repeatedly representing a breathing crack.
Hinge models of the cracked rotor were reviewed (Gasch, 1993). Breathing crack model of the
flexible rotor was proposed in the early time (Jun et al., 1992). The comparison of these two models
is available in papers by Patel and Darpe (2008a, 2008b) and Cheng et al. (2011). In case of a
cracked rotor system, the stability of periodic motion was used to detect the presence of cracks
(Sinou, 2007, Chen et al. 2007, Han and Chu, 2012, Collins et al., 1991, Babu et al., 2008). Apart
13 | P a g e
from simple two-DOF systems used to model the rotors, various analytical, semi-analytical
techniques, as well as finite element techniques have been used to model rotors with cracks.
In double row angular contact ball bearings, the restoring forces are a function of quadratic
terms of displacement. Ishida et al. (1990) investigated forced oscillations of rotor-bearing systems
in the presence of these nonlinearities. A simple Jeffcott type of rotors was considered in this work
and experimental results validated theoretical results regarding the shape of resonance curves. A
detailed model of the ball bearing system considering the parametric excitation effect was however
considered. Ji et al. (1998) used perturbation methods for the analysis of transient and steady-state
vibrations of the nonlinear rotor-bearing system. A nonlinear model of restoring forces generated
by bearings was considered. However, parametric excitation considerations and other important
Chen (2009) studied an ideal Jeffcott rotor considering three nonlinear factors of ball bearing
systems. First is the nonlinear Hertzian contact force between balls and races, second is the
clearance of bearings, and third is the varying compliance (VC) vibrations. Villa et al. (2008)
studied the stability of nonlinear systems in the frequency domain. A finite element method was
used to discretize equations of the motion. In the above study, the steady-state behaviour of
nonlinear systems with a very large number of the degrees-of-freedoms was studied using
condensation techniques. However, no attention was paid to calculate the transient response of the
overall system.
Hou et al. (2016) studied the dynamics of a simple Jeffcott rotor mounted on ball bearings in
the presence of a transverse crack. In case of a Jeffcott rotor, bearings are mounted on the
shaft/rotor edges, giving less contribution of nonlinear bearing effects to the overall dynamics of
the system. This is probably the most recent paper in the field of rotor dynamics, investigating the
14 | P a g e
combined effect of bearing nonlinearities and rotor cracks. It is also observed that the Jeffcott
rotor does not represent the real rotor-bearing system with multiple rigid disks, couplings, flexible
bearing supports and centrally located ball bearings. All these parameters demand a high order
model of the system to represent the exact dynamics of the real rotor.
The presence of cracks in a revolving rotor converts a linear system into a nonlinear system
with varying system parameters due to the parametric excitation. Historically, the 2X (where X is
the critical or resonance frequency) response of a revolving rotor system in sub-harmonic and
super-harmonic regions has been used to detect the presence of cracks. This process can be used
in both frequency and time domains. Markert et al. (2001) applied the same approach in the time
domain to identify a crack and its location in a flexible rotor. In this regard, Kulesza et al. (2011)
designed a set of observers to find out the presence and location of a crack in a rotor. Sawicki et
al. (2011a) used auxiliary harmonic excitations (AHE) to detect the presence of cracks in the rotors.
They observed the dynamic response of the rotor under sinusoidal excitation. They noticed that
the spectrum of vibration response of a cracked rotor contains multiple frequencies. These
frequencies are a function of the rotor spin speed and external excitation frequency. The parametric
excitation effect caused by rotor cracks is responsible for the generation of these frequencies. This
is a form of nonlinearity. The same concept was used by the authors and their co-workers in other
publications (Sawícki et al., 2011, Kulesza, 2014). These methods are quite recent and are being
used frequently. Some of the researchers detected the rotor crack based on the transient response
of the revolving rotor. Darpe et al. (2004b) studied the transient vibration behaviour of a cracked
Jeffcott rotor while crossing the critical or resonance speed. They studied the orbit plots near sub-
harmonic resonances to analyze the effect of critical speeds on the orbit orientation. Babu et al.
(2008) applied Hilbert-Huang Transformations (HHT) to analyze the transient response of the
15 | P a g e
cracked Jeffcott rotor. They emphasized that HHT gives better results compared to the continuous
wavelet transform and Fast Fourier transform (FFT) for crack detection in a rotor based on the
transient response. The main emphasis was on the transient response of the rotor when passing
through a critical frequency. Silani et al (2013) investigated the ability of the short-time Fourier
transform (STFT) to detect the presence of small cracks based on the transient response
calculations. Yang et al. (2015) investigated the role of Empirical Mode Decomposition for
analyzing cracks based on the transient response. The existing research in references (Babu et al.,
2008, Silani et al., 2013, Yang et al., 2015) is mainly based on the application of STFT for the
response in resonance regions. This high amplitude in resonance regions can increase the crack
propagation speed. However, working in pre-resonance regions (where the vibration amplitude is
The rubbing of the rotor and stationary stator is also a common fault in rotating machinery. In
this concern, Jacquet-Richardet et al. (2013) presented a detailed literature survey on the
rotor/stator rub phenomena. Elastic collision theory and impact energy models are quite helpful in
the analysis of rotor/stator rub. Based on that, Cong et al. (2011) evaluated the severity of the rub
impact fault. Some researchers presented numerical simulations along with experimental results
and established the correlation between them (Torkhani et al., 2012). Light, medium and heavy
rub conditions were analyzed theoretically and experimentally in their work. Behzad et al. (2013)
also used contact mechanics principles in modelling rotor/stator contact. Ma et al. (2013, 2015)
modelled a rotor-bearing system by using the Lagrange Multiplier method for the contact
formation. Ma et al. (2014, 2015) extended his previous work and modelled the rotor-stator rub
under different loading conditions and stator configurations. Recently, Varney et al. (2015)
16 | P a g e
the effect of gravity. They obtained the various type of vibration responses including periodic,
There are other physical phenomena related with the rotor/stator rub like jump phenomena,
synchronous full annular rub (FAR), partial forward whirl (PFW), partial backward whirl (PBW),
chaotic response and dry whip (WHIP) (Jiang and Ulbrich, 2004). In a WHIP, the frequency
remains constant and the amplitude increases dramatically. This is a dangerous state of affairs and
It has been observed that WHIP can be initiated without the external stimulus if the right
combination of rotor spin frequency and unbalanced mass is achieved. Shang et al. (2011) studied
analytically the global vibration characteristics of rotor/stator rubbing systems. Patel and Darpe
(2008b) investigated rotor-stator rubbing in a cracked rotor using a full spectral analysis. Lee and
Han (1998) used the concept of planar motion to diagnose the rubbing induced faults. Patel et al.
(2009) and Zhao et al. (2012) studied the application of Hilbert-Huang transformations and
wavelet transformations to vibrations produced during coasting up of a rotor with rotor/stator rub.
Generally, the rub phenomena are reproduced by simulations without considering the effect of
gravity. The effect of gravity needs to be studied in full detail along with the presence of a
transverse breathing crack. Also, there is an urgent need to derive new techniques of diagnosing
cracks and rubbing faults based on the vibration response, even if these two faults coexist.
Due to the absence of frictional contacts, active magnetic bearings (AMB) have a long service
life. These bearings are robust in nature and require very less maintenance. Hence, they are used
in many areas like flywheels (Park 2014; Ren et al. 2013; Fang et al. 2013a, 2013b), control
moment gyros and turbopumps (Abrahamsson et al. 2014; Yi et al. 2008). The optimal control is
17 | P a g e
an important technique of feedback controllers. An optimal solution was obtained in terms of
control energy and error (Jastrzębski et al. 2009). However, system uncertainties were not
considered in the design of the controller. Robust control techniques like H∞ and µ-synthesis are
worth considering when designing the controller for an uncertain system (Mushi et al. 2012).
Stability properties of the closed-loop system with such controllers are better than that with optimal
controllers. Proportional – Integral – Derivative (PID) controllers tuned using the multi-objective
A gain phase modifier was used to adaptively suppress vibrations caused by the unbalance
(Fang et al. 2013a). This method was capable of handling the trouble created by low pass
characteristics of the power amplifier. Chen et al. (2015) used a double loop compensation method
Zheng et al. (2016) presented an unbalanced vibration reduction scheme using the PID
controller for a magnetically suspended centrifugal compressor. An adaptive notch filter (ANF)
was used for the frequency estimation. Other methods of the frequency estimation include adaptive
observer structures, frequency locked loop (Rodriguez et al. 2011), marginal median discrete
Fourier Transform (Djurovic, 2007), and neural network theory. Adaptive Notch Filter is well
known in the field due to its simple structure and small computational burden (Mojiri et al. 2004).
The biggest advantage of this technique is that its efficiency remains intact even in the presence
of power system disturbances (Yazdani et al. 2009; Chu et al. 2011). Yazdani et al. (2009) and
Chu et al. (2011) used ANF for reducing the unbalanced vibrations of magnetically levitated rigid
rotors.
18 | P a g e
2.4. Applications of rotor dynamic systems in machine tools
CNC machines have three important units: i) ball screw drives (BSDs) for the linear motion
of the table or carriage, ii) spindle units for giving rotary motion to cutting tools, and iii) control
unit for the overall control of the CNC machine. In BSDs, the rotary motion of the motor is
converted into a linear motion of the carriage by a ball screw nut system. The BSD shaft is mounted
on ball bearings. It can be preloaded to avoid any backlash. Due to axial flexibility of the shaft and
finite stiffness of balls of the nut and bearings, axial vibrations are generated in the BSD system.
The coupling between the motor and shaft also contributes to these vibrations. These are also called
structural resonances. Structural resonances inhibit the performance of BSD in many ways
(Altintas et al., 2011, Varanasi and Nayfeh, 2004, Pritschow and Croon, 2013). Mainly, they lower
the bandwidth of the control system by lowering the gain and phase margins of the overall system
(Gordon and Erkorkmaz, 2013). The particular shape of the required reference motion trajectories
excites these flexible structural modes. The ultimate goal of a servo control system is to dampen
these flexible modes and to increase the bandwidth of the overall system (Kamalzadeh and
Erkorkmaz, 2007).
Different active, as well as passive techniques, are available to reduce the structural resonance.
In passive techniques, the main approach is to reduce the excitation of flexible modes by
aggressive motion commands. The reference trajectories are optimized and pre-filtered to generate
desirable frequency contents (Sencer et al., 2015). Input pre-shaping is an important passive
damping technique to reduce undesirable structural vibrations (Singhose et al., 1996). In this
technique, smooth trajectory commands are generated to reduce undesirable structural vibrations.
Optimal reference commands can be generated using filtered B-splines (Duan et al., 2018). The
19 | P a g e
robustness of the overall system can also be enhanced by optimizing the frequency spectrum of
In active techniques, active damping can be provided to dampen flexible modes. For such
systems, a pole placement control strategy provided quite favourable results (Gordon and
Erkorkmaz, 2013). The sliding mode control strategy was used to control the rigid body and
flexible mode of the BSD system (Kamalzadeh and Erkorkmaz, 2007, Sencer and Tajima, 2016).
In addition, the sophisticated LPV control strategy was used to compensate time-varying dynamics
generated due to removal of mass during metal cutting (Sepasi et al. 2012). Combination of the
pole placement and loop shaping was used (Hanifzadegan and Nagamune, 2015) to safeguard the
controller was implemented for the stiffness variation and mass removal due to cutting.
Feedforward control is another control strategy used for the control of ball screw drives. A
zero-phase error tracking controller (ZPETC) is widely used for this purpose (Tomizuka, 1987).
In this direction, Chen et al. (2006) used a hybrid controller made by combining the ZPETC
controller with cross-coupling controller. Erkorkmaz and Altintas (2001) used a controller made
by the combination of disturbance estimator, friction compensator and a ZPETC. The main
problem with feedforward controllers is that the plant dynamics must be known with certainty.
This problem was solved by Uchiyama (2008). He used an adaptive two-DOF controller to feed
tool drives. The system parameters were updated to achieve desired performance in the presence
of uncertain reference signals. To compensate for position-dependent dynamics of the ball screw
drive, a controller composed of modal characteristic modifier and adaptive tuning algorithm was
used (Liu et al., 2017a, 2017b). Artificial neural networks were used for adaptive tuning. It is
20 | P a g e
obvious from the above work that hybrid controllers are better suitable for satisfying multiple
requirements.
Although robust H∞ controllers require no special hardware for the control implementation,
they have not gained wide acceptance in the industry. In industry, a cascade PPI controller is
normally used (Sun et al., 2016). Here P is the proportional controller and I is the integral
controller. PI controller is implemented in the motor speed loop and P controller is implemented
in a table position loop. Pritschow et al. (2003) modified this controller by additional acceleration
feedback. A semi-active damper between the table and machine bed was used to enhance the
closed-loop performance (Verl and Frey, 2012). A feed drive mechanism with soft axial bearing
and strong parallel damper was used to enhance the performance (Pritschow and Croon, 2013).
Using the concept proposed in references (Pritschow et al., 2003, Verl and Frey, 2012, Pritschow
and Croon, 2013), the increased bandwidth was obtained but additional sensors or actuators were
required. Very recently, Sun et al. (2018) increased the bandwidth of the position loop without
using any additional actuator, with the help of a disturbance observer. Li et al. (2018) used a
feedforward compensation method for dynamic mechanical tracking errors. A dynamic model was
proposed to offset position commands given to the controller. Dumanli et al. (2018) used the
acceleration and jerk feedback to improve the bandwidth of controllers. Linear Quadratic
Merritt (1965) described the basics of the regenerative chatter (Marui et al., 1983, Kato and
Marui, 1974). Jen et al. (1996) described the dynamic interaction of workpiece and lathe structure.
In this work, the flexibility of the cutting tool and workpiece was studied in combination to
generate stability lobe diagrams. Shawky et al. (1998) used a model predictive control-based
21 | P a g e
strategy for enhancing the accuracy of bar turning. They used the Kalman filter-based state
estimation with online measurements. Yu et al. (2008) conducted the stability analysis of flexible
shafts (uniform and stepped). They emphasized on the location-dependent stiffness of the shafts
and related it with the stability lobe diagrams of stepped shafts. Normally the cutting dynamics is
represented as a single-DOF system. However, the two and three-DOF systems are more suitable
to represent the cutting dynamics of flexible shafts. Rao et al. (1999) used a 3-DOF model to
represent the dynamics of the turning process. This analysis was more suitable than that presented
by Wang and Cleghorn (2002). The work on stability analysis of turning of a long slender shaft
was extended by Li et al. (2012). In certain circumstances, the flexibility of workpiece and cutting
tools need to be considered. In this case, the compliance between the two was considered by Otto
et al. (2015). Recently, the stability analysis of long flexible shafts was also presented by Stepan
et al. (2017).
Active chatter control (ACC) for turning operations was presented by Ganguli et al. (2007).
They used frequency-domain methods for checking the stability of turning operations. To validate
theoretical results, experiments were conducted and presented (Ganguli et al., 2005). The concept
of an active dynamic absorber was introduced by Tewani et al., (1995). Apart from turning
operations, the ACC of the boring process has undergone tremendous progress. Recently magnetic
actuators were used in ACC of the boring process (Lu et al., 2014). ACC for the turning process
with the experimental validation is reported by Shiraishi et al. (1991) to implement an optimal
control method. A combination of frequency and time domain methods was used to study the
stability of cutting systems (Mei et al., 2005, Lu et al., 2018). Ma et al. (2017) emphasized that it
is difficult to measure the structural parameters of the cutting system. With the passage of time,
these parameters are bound to change also. They used an Adaptive Sliding Mode Control for ACC.
22 | P a g e
They theoretically proved the stability of the proposed controller and experimentally validated the
effectiveness of the controller. These show state of the art of techniques for chatter prevention in
turning operations.
There is a trend in combining passive and active techniques to enhance the overall efficiency
of the process. Meng et al. (2015) used a sinusoidal spindle speed variation in the combination
with a state feedback controller. After the thorough study, it was observed that the ACC control
has been implemented for preventing chatter for fixed-parameter systems. The dynamics of
stepped shafts with the location-dependent flexibility is quite complicated. As the material is
removed, natural frequencies of the shaft change drastically. Hence, the ACC designed for one
From time to time several passive techniques were devised to enhance the depth of cut without
chatter in boring operations. The research of Klein et al. (1975) provided the theoretical foundation
for enhancing the chatter stability by means of the active chatter control strategy. Tanaka et al.
(1994) used piezoelectric actuators for enhancing the chatter stability of boring operations. Wang
et al. (2001) used electrorheological fluids for actively controlling the chatter in boring operations.
Pratt et al. (2001) investigated theoretically and experimentally the stability of a slender boring bar
under regenerative chatter conditions. Andréne et al. (2003) installed a piezo actuator into a turning
tool holder for vibration control. Chen et al. (2014) used magnetic actuators for controlling the
chatter in boring bars. Mei et al. (2009) used a magnetorheological fluid controlled boring bar for
Experimental results were also discussed (Ganguly et al., 2005). Parameters of machines are
subjected to change during their service life. Designed controllers must be robust enough to bear
23 | P a g e
these uncertainties. In this regard, Sliding Mode Control (SMC), Linear Quadratic Regulators
(LQR) and H∞ controllers have been applied for active control applications (Tewani et al., 1995).
Ma et al. (2017) observed that in designing all these controllers, the proper estimation of machine
The sensor cost is decreasing day-by-day giving a push to online chatter detection technology.
Researchers have developed methods for online chatter detection using vibration, sound and power
Rahman and Ito (1986) diagnosed the onset of chatter based on the horizontal deflection of
the cutting tool. Liao and Young (1996) proposed the spindle speed regulation to control chatter
when it has just started. They collected the dynamic cutting force signal from the dynamometer.
They used fast Fourier transform of collected signals to determine the new spindle speed. (During
chatter, due to the interaction of cutting tool and workpiece, a particular sound is generated. A
microphone is a good sensor to catch that sound. Delio et al. (1992) used the audio signal for
detection and control of chatter. Schmitz et al. (2001) extended this idea because microphones are
(Schmitz et al. 2002, 2003), they worked on the once per revolution audio signal variance as an
indicator of chatter. They also worked on the statistical evaluation of the audio signal. The energy
of the measured sound signal, when exceeds a certain threshold level, can be considered as an
indicator of chatter (Weingaertner et al., 2006). Ismail and Ziaei (2002) worked on a 5-axis milling
machine. They conducted a sound analysis of the machining process in combination with offline
scheduling of process parameters and online spindle ramping. Tsai et al. (2009, 2010) worked on
24 | P a g e
the signal obtained from acoustic sensors and proposed a spindle speed compensation technique.
Okuma (2006) presented a commercial chatter detection solution. They used microphones that
were connected to NC machines through a USB port to give acoustic feedback. He also presented
a software downloadable tool to NC machines, which can suggest the new spindle speeds for
online adaption. However, the limitation of the audio-based signal techniques is that the sound
from other machines can interact with the machine under consideration.
Faassen et al. (2006) used accelerometers for the detection of chatter in machine tools.
Doppenberg et al. (2006) used the algorithm proposed by Faassen et al. (2006) to adjust the spindle
speed to move the system into a chatter-free zone. Tansel et al. (2012) worked on the chatter
detection on a milling machine with index-based reasoning using a torque-based signal from a
rotary dynamometer. They used a lookup table approach for classification of input signals. The
chatter detection in case of turning is easy compared to milling operations. For the turning
operation chatter detection, Tangjitsitcharoen and Moriwaki (2008), and Tangjitsitcharoen (2009)
proposed the use of power spectral density of the tool deflection. Kuljanic et al. (2008, 2009)
developed a multi-sensor chatter detection system. They compared the sensitivity of displacement
sensors, accelerometers and dynamometers for chatter detection applications. Vela et al. (2010)
used a detrended fluctuation analysis (DFA) technique adopted for the time-frequency domain to
detect chatter instabilities from data obtained from accelerometer sensors. DFA has additional
advantages over the traditional spectrum analysis. These can deal with nonstationary and nonlinear
signals efficiently. Yao et al. (2010) used wavelets and support vector machines for chatter
detection. Zhang et al. (2010) used a combination of hidden Markov models (HMM) and artificial
neural networks (ANN) for chatter monitoring. In their work, HMM were used for temporal
modelling of features extracted from time-varying signals and ANN were used for integrating
25 | P a g e
inputs. Eppel et al. (2010) used optical sensors for measuring the milling tool deflection. Based on
the signal obtained, they developed a technique for differentiating between chatter and non-chatter
conditions.
Cao et al. (2013) combined wavelet packets and Hilbert Huang Transformations (HHT) for
the chatter identification in end milling operations. Wavelets were used to denoise the signal and
HHT were used to analyse the time-frequency distribution of the signal obtained from
accelerometers mounted on the spindle unit of the milling machine. Wang et al. (2009) emphasized
that Fourier transform is not capable of differentiating transition in non-stationary signals, hence
is not suitable for online detection of chatter onset. They presented a statistical chatter detection
algorithm based on the statistical analysis of discrete wavelet transforms. Choi and Shin (2003)
presented a wavelet-based maximum likelihood parameter estimation technique for the chatter
identification in turning and milling operations. Yoon and Chin (2005) emphasized that detailed
coefficients in the wavelet transform can give some indications of the chatter in milling operations.
The use of accelerometers and other sensors increases the cost and complexity of CNC
machines. Hence researchers have tried to detect the chatter based on the signal obtained from the
servo system of the spindle unit. Kakinuma et al. (2011) used disturbance observer theory to detect
the self-excited and forced vibrations developed during chatter. Shimoda et al. (2017) also used
disturbance observer theory to detect the presence of chatter from the signal obtained from the
drive unit (i.e. linear encoders) of CNC machines. In these techniques, they used bandpass filters
to detect chatter frequency. With this method, it becomes easy to isolate the structural dynamics
of the cutting tool. However, the chatter frequency is not known a priory and hence it is difficult
to determine the cut-off frequencies of these filters. Manual intervention becomes necessary based
on cutting conditions. This is however detrimental to online chatter detections. To overcome this
26 | P a g e
difficulty, frequency domain chatter detection techniques were developed to use the total spectrum
of the available signal from sensors (Liu et al. 2018). Recently, Aslin and Altintas (2018) used the
machine tool drive unit current obtained from a servo system for the detection of chatter. They
removed the forced vibration component and used a Fourier Transform of the remaining signal to
For applications of machine learning in machine health monitoring, Hoang et al. (2019)
presented a thorough survey of applications of deep learning techniques for the roller bearing fault
diagnosis. For more detail, references therein can be referred to Zhao et al. (2019). They discussed
the applicability of different machining learning methods for machine health monitoring. They
considered the gearbox fault detection, induction motor fault detection and tool wear prediction of
CNC milling machines. They used statistical, frequency domain and time-frequency domain
results in the analysis. Oleaga et al. (2018) proposed a machine learning-based solution for chatter
prediction in a heavy-duty milling machine. They used regression trees for prediction of the chatter
basis. It is a known fact that collecting experimental data is a costly affair. Bearing this in mind,
Sobie et al. (2018) used a simulation-driven machine learning approach for bearing fault
classification. They used data collected from high-resolution dynamic simulations of ball bearings.
These data were used to train machine learning algorithms. They used convolutional neural
networks for the bearing fault classification. After an extensive search, it was realized that the
literature available in the field of chatter detections using the machine or deep learning is scant.
27 | P a g e
2.5. Research gaps and directions
Based on the literature survey, following gaps are noticed and research directions are
identified.
There is a need to show the impact of the ball bearings in the rotor system on the efficiency
There is a need to devise a transverse crack detection strategy, which can be used in the
The effect of gravity needs to be studied in full detail along with the presence of a transverse
breathing crack for rotor/stator rubbing. Also, there is an urgent need to derive new
techniques of diagnosing cracks and rubbing faults based on the vibration response, even
An effort should be made such that the fault detection can be done in an online manner.
vibrations. The control strategy should also be capable of identifying the rotor system faults
(like rotor crack and coupling misalignment) using the power spectral density of applied
control inputs.
There is a great need to implement principles of rotor dynamics to machine tools. There is
a need for applying hybrid controllers to enhance the command tracking efficiency as well
There is a need to apply Linear Parameter Varying (LPV) controllers for updating
controller parameters during the total span of the shaft during turning operations.
28 | P a g e
Due to the flexibility of the boring bar in all directions, there is a need for the 3D model
for the cutting process in case of boring operations. Heavy forces are required to maintain
stability during active chatter control applications. There is a need for controllers to require
small control forces for ACC applications. There is also a need to consider the effect of
ovality of the workpiece as well as the eccentricity of the boring bar during the modelling
There is a need to derive the dynamic model for stationary bar (SB) and rotating bar (RB)
boring operations and to investigate the difference between the two. Feasibility analysis of
There is strong need of developing an online chatter detection technique for turning, boring
and milling operations. An algorithm is required for the automatic spindle speed regulation
so that a system can be dragged to the chatter-free zone from chatter conditions (without
using ACC).
Based on the gaps found in literature and identified research directions, research has been
conducted. Following chapters will introduce theoretical as well as experimental work done in this
29 | P a g e
Chapter 3
3. Modelling and fault detection in rotodynamic systems
3.1. Overview
In case of simple Jeffcott rotors, bearings lie on edges of the rotors, hence bearing
nonlinearities are not properly absorbed in the dynamic response. Real rotors containing multiple
disks, flexible bearing supports, centrally located ball bearings and couplings are far more complex
in terms of their dynamic response than simple Jeffcott rotors. Nonlinear dynamic modelling of
the flexible rotor with centrally located ball bearing has been developed along with transverse
crack and used for simulation purposes. It has been observed that crack detection techniques based
on the steady-state vibration response work well for simple Jeffcott rotors only. The presence of
ball bearings and other system nonlinearities inhibit the direct application of steady-state dynamic
response-based crack detection techniques for real rotor-bearing systems condition monitoring.
However, transient response-based crack detection techniques can be successfully applied to work
in the presence of system nonlinearities. Crack propagation is enhanced during high amplitude
vibrations, hence, instead of working in the resonance region, working in pre-resonance regions is
combination of active vibration control and Short Time Fourier Transform is proposed in this
work. The proposed technique basically operates on the signal in pre-resonance regions (with low
vibration amplitude) hence avoids chances of the crack propagation. This technique can be used
to detect the presence of even small cracks and works well in the presence of bearing nonlinearities
and flexible bearing supports that are inherently present in real rotor-bearing systems and
30 | P a g e
3.2. Mathematical Modelling
Many techniques are available for mathematical modelling of cracked flexible rotors.
Analytical techniques are slightly difficult to apply; however, the generated mathematical models
are low in the order. They can only be applied to simple geometries. Finite element methods can
be applied to complex geometries, but the resulted models are high in the order. In this research, a
discrete mass spring damper element-based model (DMSDM) of the real rotor system is developed
compared to the finite element model (FEM). To obtain better accuracy, parameters of the low
order DMSDM are fine-tuned to match its results with a high order FEM. The resulted model is
A shaft is discretized into several finite elements. Each element contains two nodes with 4
degrees-of-freedom at each node. For discretization, Timoshenko beam elements are used.
Elemental governing equations of the motion are as follows (Lalanne and Ferraris, 1990, Vashisht
et al., 2018)
M e
T + M eR xei + CeR - Ω Rotor G eR xei + K eB xei = Qei (3.1)
where M eT and MeR are translational and rotary elemental mass matrices. CeR , G eR and K eB are
elemental damping, gyroscopic, and stiffness matrices respectively. Qei defines a gravity force
vector for the element. xei is the displacement vector for the element. Ω Rotor is the rotor spin
frequency. For the sake of simplicity, classical damping is considered by using Rayleigh’s
expression: CeR (M Te M Re ) K Be , where and are constants. The dynamic model of the
rigid discs is given by the following relations (Lalanne and Ferraris, 1990).
M d
T + M dR xdi + -Ω Rotor G dR xdi = Qdi + Fid (3.2)
31 | P a g e
where M dT , MdR and G dR are the translational mass, rotary mass, and gyroscopic matrices,
respectively. Qdi and Fid are the weight of the disk and the unbalance forces respectively. The
unbalanced mass is attached to the disk 2. The bearings are mounted on flexible bearing supports
as shown in Figure 3.1 (Sections A and B). A coupling shown in Section C of Figure 3.1 is located
at the right extreme of the flexible shaft. Bearing supports are modelled as two-node linear elastic
spring elements. These bearing supports considerably change the dynamics of the overall system
and are formulated in the form of following equations of motion (Lalanne and Ferraris, 1990,
where M s , K s and Fis are the elemental mass matrix, stiffness matrix, and external load vector,
respectively. The effect of bearing stiffness is accommodated by placing the stiffness coefficients
at appropriate locations. The global dynamics of the complete rotor system is represented in the
(3.4)
M x + C- ΩRotor G x + K x = FQ + FW (3.5)
where M and G are the global mass and gyroscopic matrices respectively. The global damping
and stiffness matrices are represented by C and K respectively. The global vector of the gravity
32 | P a g e
Figure 3.1 Schematic of the finite element model of a rotor-bearing system
Equations (3.1-3.5) are used in this work to obtain the FEM of a rotor system. The rotor is lying
horizontally and has two rigid disks. The effect of gravity is given due consideration. The rotor is
divided into 20 discrete elements. The bearings, disks and coupling are assumed to be located on
nodes. However, due to the purpose of clarity, some of these components cover some of the
elements in Figure 3.1. The overall length and diameter of the rotor are 1478 mm and 40 mm
respectively. The mass of the Disk 1 is 5 Kg and Disk 2 is 19.4 Kg (Figure 3.2). The rotor is
mounted on two bearings. A flexible coupling connects the rotor at the right-hand side. Left-hand
33 | P a g e
side bearing support contains a linear bearing with stiffness 7x108 N/m with a mass of the bearing
pedestal or support as 6 Kg. The stiffness of the bearing support is 3.8x106 N/m. Right-hand side
bearing support contains a ball bearing with bearing stiffness 4.6x107 N/m and a mass of the
bearing pedestal as 6 Kg having the same stiffness of the bearing as the left-hand side support. The
rotor-bearing system is assumed being made of various spring-mass dampers connected. Figure
3.2 represents the equivalent discrete spring-mass system and shows various physical parameters
of the rotor-bearing system. This figure represents these connected elements in one plane. The
important physical parameters of the complete system along with mechanical properties are shown
Table 3.1 Physical parameters and mechanical properties of the rotor bearing system
34 | P a g e
Table 3.2 Damping parameters of the discrete mass representation of rotor bearing system
The DMSDM has reduced order with high computational efficiency. Figure 3.3 represents the
DMSDM of the overall system in the horizontal direction. For the vertical direction, a similar
system can be generated. Equations of motions can be derived using the force method or energy
method.
In this research, the energy method is used. The kinetic energy T and potential energy U subjected
to lateral vibrations in the horizontal direction can be written as follows (Lalanne and Ferraris,
1
T m1 u12 m2 u22 m3 u32 m4 u 24 m5 u52 m6 u62 m7 u72 (3.6)
2
U (3.7)
2 k6 (u5 ) 2 k7 (u4 u6 ) 2 k8 (u6 u7 ) 2 k9 (u7 ) 2
The kinetic and potential energy in the vertical direction can be obtained in a similar manner.
Using the Lagrange method, equations of motions in horizontal and vertical directions can be
calculated. Lagrangian (L) is defined as follows (Lalanne and Ferraris, 1990, Vashisht et al., 2018).
L=T-V (3.8)
35 | P a g e
The Euler-Lagrange equation is given as follows (Lalanne and Ferraris, 1990, Vashisht et al.,
2018).
d L L
(3.9)
dt q q
Using this equation and incorporating the appropriate damping (shown in Table 3.2), the following
M2=1.5 Kg M3=5Kg
M4=1.5 Kg k7 M6=19.4 Kg k8 M7=1Kg
k4
k3
M1=6Kg M5=6Kg
Figure 3.2 Equivalent physical model of the discrete mass model of the rotor bearing system in
one plane (not to scale)
36 | P a g e
x2
M2=4.5 Kg
k1 c1 M4=3.5 Kg
x4
c7
c5 K7=6.35 x 106 N/m
k5
x5 21.4 Kg
m5 m6
x6
c9 k9
Figure 3.3 Phenomenological model of the discrete mass model of the rotor bearing system in
one plane (not to scale)
where us are deflections, cs are damping coefficients, ks are the stiffness’s and Fu s are the forces
acting in the horizontal direction. The equations of motion in the vertical direction are given as
follows.
where vs are deflections in the vertical direction and Fv s are forces acting at individual masses in
the vertical direction. Equations (3.10-3.11) are used to construct the DMSDM of the rotor-bearing
system used in this research. To match the modal parameters of the full and reduced order model,
37 | P a g e
(for r=1…H) are modal frequencies and mode shape components at the N0th degrees-of-freedom
respectively (for the first H modes), corresponding to parameter x, we need to solve following
where K(x) and M(x) represent the stiffness and mass matrices respectively of the reduced-order
DMSDM with parameter x. The optimization problem can be set up to minimize the following
objective function.
2
ˆ i ( x ) i
2
ˆi ( x ) i
where i ( x ) , i 1,..., H . Here z T z is the usual Euclidian
2
z
i
2 2
i
norm. H is the number of modes considered. The modal frequencies and mode shapes for the ith
mode of the full order system are represented by i and i , respectively. The original shaft
stiffness parameters were calculated using the strength of material formulas. These modified
coefficients are available in Figure 3.3. Figure 3.4 shows the frequency response functions of both
models at Disks 1 and 2 for the vertical response. It is evident that only the first two modes are
sufficient to calculate the overall response of the system. The contribution of higher modes is
negligible. Table 3.3 shows the comparison of un-damped natural frequencies of both models. The
38 | P a g e
Table 3.3 The comparison of natural frequencies
Figure 3.4 Bode plot for the different modeling techniques at Disk1 and 2
3.2.3. The Ball Bearing Model
Figure 3.5 shows a ball bearing. Ball-bearing 6014 is used in this work. Table 3.4 shows the
parameters of the ball bearing. The ball bearing is modelled as a two-degrees-of-freedom system
(one in horizontal and the other in the vertical direction) with the radial clearance and Hertz contact
between races and spherical balls. While formulating the dynamical equations of motion it was
assumed that the outer race of the ball bearing is fixed to the flexible bearing support. However, it
is assumed that the inner race is fixed to the flexible rotor. The rotor and cage rotate at different
39 | P a g e
frequencies. The purpose of the cage is to maintain a constant separation between balls. The
relation between the cage rotational velocity and shaft rotational velocity is given by the following
Ri
cage = rotor (3.14)
Ri Ro
where Ri and Ro represent the inner and outer cage radius of the ball bearing respectively. rotor
is the rotor spin velocity. cage is the spinning velocity of the cage. Nb is the number of balls in
2
the bearing. The angular space between two balls is equal to . The angle between balls
Nb
is given by the following relation (Harsha et al., 2004, Tiwari and Gupta, 2000, Vashisht et al.,
2018a).
2
k (k 1) caget , k=1,....,Nb (3.15)
Nb
The relative radial distance k between the inner and the outer races at the kth ball position can
be expressed as follows (Harsha et al., 2004, Tiwari and Gupta, 2000, Vashisht et al., 2018a).
40 | P a g e
Y
Outer Ring
Stator centre
Inner Ring
Ball
Ro Ri X
Radial
Clearance
SHAFT
r
Rotor centre
u i , u 0 , vi and v0 respectively. k is the position at the kth ball. Based on the position of each ball
and the local Hertzian contact, the restoring force can be estimated by using the following relation
( k ) n k
k (3.17)
0 k
where is the radial clearance between races and balls. is the effective stiffness. It can be
obtained by combining the effects of all the balls. It shows in Figure 3.5 that the ball bearing
stiffness Z actually represents stiffness k5 as follows (Harsha et al., 2004, Tiwari and Gupta, 2000).
1
(3.18)
1 1
3/2
3/2
Ki Ko
Global bearing reaction forces can be calculated by adding the individual restoring force from each
ball of bearings. Afterwards, the total restoring force components u and v in X and Y directions
can be represented respectively as follows (Harsha et al., 2004, Tiwari and Gupta, 2000).
41 | P a g e
Nb
u k cos( k ) (3.19)
k=1
Nb
v k sin( k ) (3.20)
k=1
Equations of the motion for the DMSDM in the horizontal direction are modified as follows.
m4u4 c4u3 (c4 c5 c7 )u4 c5u5 c7u6 k4u3 (k4 k7 )u4 ( u ) k7u6 Fu4
(3.21-3.22)
m5u5 c5u4 (c5 c6 )u5 +k6u5 ( u ) Fu5
Similarly, equations of motion of the DMSDM in the vertical direction are modified as follows.
with k =(u 4 -u 5 )cos( k ) +(v4 -v5 )sin(k ) k=1,....,Nb . Equations (3.21-3.24) are used to
v
Figure 3.6 Cracked Rotor model
42 | P a g e
Table 3.4 Geometric properties of 6014 Ball Bearing (Tiwari and Gupta, 2000)
3.2.4. Cracked rotor model compatible with Discrete Mass Spring Damper based model
m 0 u c 0 u u cos(rotor t ) Fu
0 m v 0 c v K (t) m 2
(3.25-3.26)
sin(rotor t ) Fv
rotor
v
where u and v are displacements of the disc in the horizontal and vertical directions, respectively,
in a stationary coordinate system. The stiffness matrix of the cracked shaft is represented as K(t).
‘m’, ‘c’, ‘ε’ and ‘β’ are the equivalent mass of the system, viscous damping coefficient, eccentricity
and attitude angle of the disc imbalance force vector, respectively. ‘Ωrotor’ represents the rotational
speed of the rotor system. Figure 3.6 shows all these parameters. When a cracked rotor rotates
slowly under its own weight, the crack will open and close once periodically. Therefore, the
stiffness matrix of the shaft K (t) varies periodically during the operation due to the crack presence.
The stiffness matrix K (t) in the inertial coordinates reads as follows (Mayes and Davies, 1984,
k kuv
K (t ) uu
kvv
(3.27)
kvu
In rotating coordinates, the stiffness matrix for a cranked shaft can be written as follows (Mayes
43 | P a g e
k 0 k 0
K R (t ) f (rotor t )
k
(3.28)
0 k 0
where k, represents the stiffness of the uncracked rotor. Variations of the rotor stiffness in and
directions can be represented as k and k respectively (Figure 3.6). The crack function
f (rotor t ) depends on the angular orientation of the crack as follows (Mayes and Davies, 1984).
1 cos(rotor t )
f (t ) (3.29)
2
The stiffness matrix K (t) in the inertial coordinates reads as follows (Mayes and Davies, 1984).
cos(rotor t ) sin(rotor t )
K (t ) T K R (t )T 1 , T . (3.30)
sin(rotor t ) cos(rotor t )
By incorporating the appropriate element into the overall system, the dynamic equations are given
as:
m4u4 c4u3 (c4 c5 c7 )u4 c5u5 c7u6 k4u3 k4u4 ( u ) kuu (u4 u6 ) kuv (v4 v6 ) Fu4
m6u6 c7u4 (c7 c8 )u6 c8u7 k8u6 k8u7 kuu (u4 u6 ) kuv (v4 v6 ) Fu6
(3.31-3.32)
Similarly, the dynamic equations of motion in the vertical direction are given as follows.
m4v4 c4v3 (c4 c5 c7 )v4 c5v5 c7 v6 k4v3 k4v4 ( v ) kvv (v4 v6 ) k vu (u4 u6 ) Fv4
m6v6 c7 v4 (c7 c8 )v6 c8v7 k8v6 k8v7 kvv (v4 v6 ) kvu (u4 u6 ) Fv6
(3.33-3.34)
Equations (3.31-3.34) are used to accommodate the effect of cracks on the overall dynamics of
the rotor-bearing system. The purpose to construct the DMSDM is that theoretical results should
well match experimental results while maintaining the low order of the overall system. This low
order model will be useful for obtaining the transient response of the system in Section 3.4. The
44 | P a g e
effect of ball bearing nonlinearities is accurately absorbed in this low order model. Results
3.3. Conventional crack detection techniques based on the steady-state dynamic response
By using certain excitation signals externally and observing the produced response, various
conceptual conclusions can be drawn. These techniques are discussed in Appendix A. The negative
effect of the presence of ball bearing on the crack detection efficiency is also discussed in detail.
The steady-state response of flexible bearings can be affected by ball bearing nonlinearities.
Transient response methods provide better crack detection capabilities without losing performance
in the presence of ball bearing nonlinearities. The torque is applied to the rotor system starting
from rest conditions in such a manner that it provides a particular acceleration to the rotor. In this
study, the rotor spin frequency is increased from zero to 60 Hz in 15 seconds. This acceleration
can be varied depending upon the transient response of the system. When the system starts at zero
initial conditions, due to the presence of gravity, the system vibrates in the vertical direction. It
comes to a certain lower amplitude in a certain time interval depending on the inherent damping
of the system. For convenience, the total time period of 15 seconds is divided into three regions.
The first region (denoted as Region I) is that a portion in which initial transients die in 0 to 4
seconds. The second region (Region II) is that a portion in which the average excitation frequency
covers half the first and second critical frequency of the overall system. This region is also termed
as a pre-resonance region ( 4 to 7.5 seconds). The third region is the resonance region in which
the system vibrates at its first and second critical frequencies. This region is denoted as Region III
( 7.5 to 15 seconds). Since the present system contains the first two closely spaced modes (34.75
Hz and 45.54 Hz), region II should cover half of these critical frequencies. Region III covers the
45 | P a g e
first two critical frequencies. Figure 3.7(c) shows these regions on the timescale. It is also obvious
from the figure that when the spin frequency of the rotor passes through half of the critical
frequencies (Region II), the vertical amplitude of the system increases (in case of the cracked rotor
only). However, the amplitude in the horizontal direction remains constant (for both healthy and
cracked rotors). It is also obvious that, in the resonance region, the vibration amplitude of the
cracked rotor is much more than the amplitude of the healthy rotor. The figure is self-explanatory.
Silani et al., (2013) shows the application of STFT for the vertical response at Disk 1 or 2. First
of all the STFT of the signal for Disk 2 (with the rigid bearing supports) is presented in Figure 3.8.
Figure 3.7 Horizontal and vertical responses of the rotor bearing (Healthy and Cracked) system
at disks 1 and 2
46 | P a g e
(a) Healthy rotor with Linear Bearing (b) 40% Cracked rotor with Linear Bearing
Frequency (Hz) 200 200
Frequency (Hz)
150 150
100 100
50 50
0 0
2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time(s) Time(s)
(c) Healthy rotor with Ball Bearing (d) 40% Cracked rotor with Ball Bearing
200 200
Frequency (Hz)
Frequency (Hz)
150 150
100 100
50 50
0 0
2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time(s) Time(s)
Figure 3.8 Spectrogram of the rotor-bearing system with rigid bearing support (at disk2)
The reason for this is that the crack is near Disk 2 and the rigid bearing supports have been
considered so that our rotor nearly represents the Jeffcott rotor. The healthy bearing shows a single
inclined line in the spectrogram. Figure 3.8 represents the system response corresponding to the
linearly varying rotor spin frequency. In case of a cracked rotor, multiple such lines are observed
(Part b). The findings of this work are matched that of reference (Silani et al., 2013). However,
Part (c) shows that an additional line is also observed due to the presence of bearing nonlinearities.
But that line is not continuous. For the cracked rotor with ball bearings, two additional lines are
observed (Part d). For the cracked rotor, the number of inclined lines is decreased. The intensity
of the lines is also decreased. This shows the negative effect of the bearing nonlinearity on the
crack detection capability based on transient response also. This problem can be partially tackled
by considering the response at Disk 1 instead of Disk 2 which contains fewer distortions caused
47 | P a g e
To further enhance the crack detection capability, the segmented approach is used in this
work. As discussed earlier, the total response has been divided into three different regions. Now,
to minutely study the system response, we draw the spectrum by using STFT for the data pertaining
to region II as shown in Figure 3.9. For a healthy rotor, there are two lines: one corresponding to
the linearly varying rotor spin frequency and the other line showing the prolonged effect of initial
transients excited due to initial conditions of the system. To clearly see the response due to the
(a) Healthy rotor with Ball Bearing (b) 10% Cracked rotor with Ball Bearing
200 200
180 180
160 160
First and second mode excited
due to rotor crack
140 Response corresponding 140
Frequency (Hz)
Frequency (Hz)
100 100
Transients corresponding to
80 initial conditions 80
60 60
40 40
20 20
0 0
0.5 1 1.5 2 2.5 3 3.5 4 0.5 1 1.5 2 2.5 3 3.5 4
Time(s) Time(s)
Figure 3.9 Spectrogram of the rotor Ball Bearing system with flexible bearing support (at disk1)
48 | P a g e
Figure 3.10 Open and closed-loop responses of the 40% Cracked rotor with the robust
controller
Figure 3.10 shows the time response of the 40% cracked rotor during the run-up. To reduce the
amplitude and vibrations in horizontal and vertical directions, a controlled signal through AMB is
Due to the presence of nonlinearities caused by ball bearings and rotor cracks, the overall
system is not purely linear. There is a need to apply the robust control theory to compensate for
49 | P a g e
(a) Spectrogram of the Healthy Rotor (b) Spectrogram of the 10% Cracked rotor
200 200
180 180
Frequency (Hz)
120 120
Excitation of the
100 100
Second Mode
80 80
Excitation of the
60 60 First Mode
40 40
20 20
0 0
0.5 1 1.5 2 2.5 3 3.5 4 0.5 1 1.5 2 2.5 3 3.5 4
Time(s) Time(s)
Figure 3.11 Spectrogram of the healthy rotor and 10% cracked rotor with ball bearings at disk 1
(closed-loop response) (in region II)
3.6. Switching Control Strategy
By applying the robust controller and increasing the speed of the rotor progressively, it is
observed that the initial transients die out quickly. The spectrum of the system in region II is shown
in Figure 3.11. A single line corresponding to the spin frequency of the rotor is shown for the
healthy rotor. For a 10% cracked rotor, an additional line is observed that shows the excitation of
first and second modes when the spin frequency passes through half of the first and second critical
frequencies. The frequency response spectrum of the vertical vibration response of Disk 1 (in
region II) is obtained by using the simple Fourier Transform of data as shown in Figure 3.12. The
system nonlinearities are obvious in this figure. The dominant part of the graph represents the rotor
spin frequency. The second mode excitation is more predominant during the less percentage of the
crack depth in the rotor system. However, with the increase in the crack percentage, both the modes
50 | P a g e
Rotor with Ball Bearing with Closed loop conditions
(a) (b)
Amplitude (dB)
Amplitude (dB)
-60 Healthy rotor -60 10% Cracked rotor
-70 -70
-100 -100
Frequency component Frequency component
-110 corresponding to -110
corresponding to
rotor spin frequency rotor spin frequency
-120 -120
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Amplitude (dB)
-60 20% Cracked rotor -60 30% Cracked rotor
-68dB Second mode
-70 -70 -59dB
Second mode
-80 -80
-90 -90
-100 -100
Frequency component
Frequency component
-110 corresponding to -110 corresponding to
rotor spin frequency
rotor spin frequency
-120 -120
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Figure 3.12 Frequency domain response of the rotor ball bearing system at disk 1 (in region II)
using negative velocity feedback controller as a function of crack severity
This excitation of the first and second modes can be further enhanced by the reverse application
purpose is to reduce the vibration amplitude. By converting this ‘negative feedback controller’ to
‘positive feedback controller’, instead of suppressing the amplitude of vibrations in region II, these
modes get excited. To do this, a switching controller concept is used. According to this concept,
the same ‘negative feedback controller’ used in regions I and III is converted to ‘positive feedback
controller’ in region II. The important thing worth mentioning is that, for the healthy rotor, no
signal is generated in region II, hence the feedback controller is not effective at all i.e. it does not
Figure 3.13 shows a flowchart of the switching controller. For a given set of mass, stiffness and
damping matrices, the system is simulated using ‘ode45’ command of MATLAB. The relative and
absolute tolerances are settled as 1x10-9 for obtaining the accurate numerical results.
51 | P a g e
Start
Feedback Receiver
Feedback Receiver
Increment
t=t + dt and w=d+ dw
Yes
Is t>Tfinal
No
Calculate vertical response at disk 1 and disk 2
Find the STFT of the response for vertical response for region II
End
52 | P a g e
(a) Response of the flexible rotor with Active Control (b) Response of the flexible rotor with Switching Control
-60 -60
Amplitude (dB)
-80 -80
-90 -90
-100 -100
-110 -110
-120 -120
0 50 100 150 0 50 100 150
Frequency(Hz) Frequency(Hz)
Figure 3.14 Response of the 10% cracked flexible rotor with active and switching control
The gyroscopic matrix is not considered in the system as it works on very low spinning speeds. As
time passes, the rotor spin frequency is linearly increased as a function of time in the form of a
linear chirp. The rotor acceleration term is chosen in such a manner to obtain the well-separated
regions I, II and III. Data are analyzed for different rotor accelerations. However, these parameters
have a negligible effect on results as long as all the three regions are well separated. A robustly
designed controller using ‘negative feedback’ is applied to the system. The maximum simulation
time is settled as 15 seconds for the well-separated regions I, II and III. For regions, I and III, a
‘negative feedback controller’ is applied to suppress vibrations. For region II, a ‘positive feedback
controller’ obtained by reversing the sign of the ‘negative feedback controller’ is applied. Here
instead of suppressing vibrations, the controller excites the system based on the vibration
amplitude in this region. It is worth noting that the ‘positive feedback controller’ only excites the
system in this region for a cracked rotor. In case of a healthy rotor, no vibrations are produced in
this region hence no excitation of the system takes place. This fact makes the difference.
53 | P a g e
200 200
180 180
Frequency (Hz)
120 120
Excitation of
100 100
Second Mode
80 80
Excitation of
First Mode
60 60
40 40
20 20
0 0
0.5 1 1.5 2 2.5 3 3.5 4 0.5 1 1.5 2 2.5 3 3.5 4
Time(s) Time(s)
Figure 3.15 Spectrogram of the healthy rotor and 5% cracked rotor with ball bearings at disk 1
(closed-loop response) (in region II) using switching control strategy
3.7. Application of Switching Control and STFT based Spectrum
With the application of active control, the amplitude of vibration decreases in the pre-resonance
region. However, with the application of switching control strategy, the vibration amplitude
increases in the pre-resonance regions. Figure 3.14 shows the comparison of the vibration response
of the 10% cracked rotor with the application of active control and with switching control. It is
obvious that the vibration amplitude at the first mode is enhanced with the switching control
Generally, the measurement signal is corrupted by the noise that is made up of a high-frequency
signal. Using a low pass filter with a certain cutoff frequency, the required signal can be made
noise-free for analysis. Figure 3.15 shows the spectrum of a 5% cracked rotor. Initial transients are
suppressed in region I. In region II, the cracked rotor system is excited. First and second modes
get more excited revealing the presence of a small crack. The first and second mode excitation is
obvious in the STFT based spectrum. For a healthy rotor, this effect is not obtained. Results show
54 | P a g e
in Figure 3.16 with a 30% crack depth (with switching control strategy). The third inclined line
also appears in the picture although faint. The colour of the lines depends upon the amplitude of
vibrations. For a lower amplitude, the blue colour is dominated. As the amplitude increases, the
colour changes from blue to white and then to yellow. Based on these colours, the amount of crack
(a) Spectrogram of the Healthy rotor (b) Spectrogram of the 30% Cracked rotor
200 200
180 180
160 160
Frequency (Hz)
to spin frequency of the rotor
120 120
100 100
80 80
Excitation of First Mode
Excitation of Second Mode
60 60
40 40
20 20
0 0
0.5 1 1.5 2 2.5 3 3.5 4 0.5 1 1.5 2 2.5 3 3.5 4
Time(s) Time(s)
Figure 3.16 Spectrogram of the healthy rotor and 30% cracked rotor with ball bearing at disk 1
(closed loop response) (in region II)
3.8. Concluding remarks
A Jeffcott rotor mounted on bearings with the linear restoring force characteristics differs
significantly in the dynamic response from real rotors. The real rotors have multiple sources of
nonlinearity like ball bearings, rotor-stator rub, rotor bow and misalignment. By modelling the ball
bearing nonlinearities as an integral part of the system dynamics and then simulating the overall
55 | P a g e
Ball bearings or other system nonlinearities inhibit the direct application of crack detection
Flexible bearing supports negatively affect the crack detection capabilities of the AHE and MSE
Transient dynamic response-based crack detection techniques are more suitable compared to
nonlinearities.
If the rotor spin frequency is varied from zero to resonance frequencies, certain amazing
characteristics are observed in the vertical dynamic response of the flexible rotor system.
Different time-frequency analysis techniques like Short Time Fourier Transform have a great
potential for detecting the interesting characteristics hidden in a frequency varying signal.
strategy.
It is obvious that there is a need to investigate the practical utility of the proposed technique. Also,
there is a strong need to develop other signal processing techniques working on the signal in pre-
resonance regions to reveal another interesting dynamic characteristic related to the cracked rotor.
To practically implement the proposed methodology on a real-life rotor, the experiment is required.
state direct current components in the signal. The use of High-Pass and Low-Pass filters can
56 | P a g e
Chapter 4
4. Modelling and detection of rotor stator rubbing action
4.1. Overview
Due to unbalance present in a rotating machinery, fluctuating stresses are generated leading
to the formation of transverse cracks in rotors. The cracks propagate with the passage of time and
increase the amplitude of vibration. High vibration amplitudes can produce rotor/stator rubs.
During the rubbing phase, the crack propagation gets enhanced due to the interconnected nature
of these faults. If left unattended, these faults can cause the premature failure of machine
components. Hence, there is a need to develop fault detection mechanisms based on the vibration
response so that these faults can be diagnosed during initial stages. The effect of gravity and the
presence of cracks significantly change vibration characteristics of the rotor, which is thoroughly
investigated in this research for a two-degrees-of-freedom Jeffcott rotor. It has been observed that
during rubbing, higher harmonics are excited. These harmonics are integer multiple of the rotor
spin frequency. A similar type of response is also observed due to the presence of a transverse
rotor crack. It is difficult to distinguish the type of faults based on the steady-state dynamic
response only. Instead of working on a steady-state vibration response, the transient vibration
response during coasting up of the rotor is considered. During coasting up of the rotor, high
harmonics are excited for both the crack as well as rotor/stator rubbing. The excitation of higher
harmonics starts at much earlier stage in the spectrogram of the vertical response for the cracked
rotor compared to that of rubbing. This fact is used in the development of a fault diagnosis
technique based on Short Time Fourier Transform of the vibration response. The proposed
technique can efficiently distinguish different types of faults even if multiple faults coexist.
57 | P a g e
Unbalance is not avoidable even in a very precise rotating machinery. During the service life
of revolving machinery, it is prone to the development of other faults like cracks and rotor-stator
rubs. The cracks are generated due to fatigue stresses caused by fluctuating bending moments.
Faults in rotating machinery are inter-related and the generation of one may lead to the
development of the other fault. Unbalanced mass tries to enhance these defects. For example, if a
crack is developed, the vibration amplitude increases and rubbing action between the rotor and
stator start due to the limited amount of clearance between rotating parts and the stationary
housing. The rubbing action in some cases can dramatically increase the vibration amplitude which
facilitates the crack growth. The other type of faults in rotating machinery are coupling
misalignment and rotor bow. The existence of these faults cannot be denied. It is important to study
A flexible rotor supported on rigid supports is shown in Figure 4.1. The rotor position is
y. The shaft rotates at a rotor spin frequency n. Governing equations of the motion of such a
M x c x k xx x m n2 cos(nt ) (4.1)
M y c y k yy x m n2 sin(nt ) Mg (4.2)
where M is the mass of the rigid disk on a flexible massless rotor, m is the unbalanced mass lying
at a distance from the central axis, c is the damping coefficient, k xx is the stiffness in the x-
direction and k yy is the stiffness in the y-direction and g is the acceleration due to gravity. There
58 | P a g e
is a stationary housing called stator covering the rotor with a clearance . The rotor/stator contact
occurs when the rotor deflection r x 2 y 2 exceeds the clearance. When the contact occurs,
two forces are generated. One being the normal force FN and other is tangential frictional force FT.
0 if r
FN (4.3)
ks (r - ) if r
FT FN (4.4)
x
cos( ) (4.5)
r
y
sin( ) (4.6)
r
Decomposing tangential and normal forces in x and y-directions as follows (Varney and Green,
2015).
r
Fx ks ( y x) (4.9)
r
r
Fy k s ( y x) (4.10)
r
59 | P a g e
Y
crack
G
C z
rotor
stator
Figure 4.1 Schematic of the 2-DOF rotor stator system (Front view)
Y Y
FN
Stator
Rotor
FT
G
G
r C
Ɵ
C O
x x
δ
Figure 4.2 Schematic of the 2-DOF rotor stator system (side view)
60 | P a g e
u
v
Figure 4.3 Cracked rotor model
where ks is the stiffness of the stator. By considering Eqs. (4.1-4.10), we obtain (Varney and Green,
2015):
r
M x c x k xx x k s ( x y ) m n 2 cos(nt ) (4.11)
r
r
M y c y k yy y k s ( y x) m n 2 sin(nt ) M g (4.12)
r
Since we have to consider the dry friction backward whirling and whip phenomena, the governing
equations of self-excited vibrations are given as follows (Jiang and Ulbrich, 2004, Vashisht et al.,
2019b).
r
M x c x k xx x ks ( x sign(v ) y ) m n 2 cos(nt ) (4.13)
r rel
r
M y c y k yy y k s ( y sign(v ) x) m n 2 sin(nt ) M g (4.14)
r rel
61 | P a g e
where v n r nwr with r x 2 y 2 is the relative velocity between the disk and stator
rel disk
at the contact point. r is the radius of the disk and nw is the whirling frequency of the rotor.
disk
This frequency is positive when the rotor whirls in the forward direction and negative when whirls
in the backward direction. In some cases, the relative velocity can even become negative. During
this stage, the direction of frictional force changes. In case of a dry friction backward whirl, the
rotor whirls at a constant backward whirl frequency. The supplied rotational energy will be
continuously transferred to the rotor in the form of the energy required for the lateral vibration of
the rotor. In this case, the amplitude of the rotor response increases to a high value. This stage is
In case of a cracked rotor, equations of the motion of the rotor system in stationary Cartesian
coordinates can be written as follows (Mayes and Davies, 1984, Vashisht et al., 2018).
M 0 x c 0 x x 2 cos(nt ) 0
0 K (t) m n (4.15)
M y 0 c y y sin(nt ) Mg
In the above expression, x and y represent horizontal and vertical displacements of the disc in
the stationary coordinate system. K (t) is the time-varying stiffness matrix of the cracked rotor.
Parameter β represents the angle of the disc imbalance force (Figure 4.3). The crack opens and
closes periodically under the effect of gravity and behaves like a breathing crack. Therefore, the
stiffness matrix of the shaft (i.e. K (t)) varies periodically due to the crack effect. The stiffness
matrix K (t) in the stationary coordinates is given as follows (Mayes and Davies, 1984, Vashisht
et al., 2018).
k xx k xy
K (t ) (4.16)
k yx k yy
62 | P a g e
The stiffness matrix of a cracked shaft, in rotating coordinates, can be written as follows [30].
k xx 0 k 0
K R (t ) f (nt ) (4.17)
0 k yy 0 k
where diagonal elements in the first matrix represent the stiffness of the uncracked rotor in
horizontal and vertical directions. Diagonal elements in the second matrix represent variations in
rotor stiffness 𝑘𝜁 𝑎𝑛𝑑 𝑘𝜂 𝑖𝑛 𝜁 𝑎𝑛𝑑 𝜂 directions, respectively (Figure 4.3). Function 𝑓(𝑛𝑡) is a
crack function. Its value depends on the angular orientation of the crack as follows.
1 cos(nt )
f (nt ) (4.18)
2
where represents the crack orientation. Its value is taken as zero in this research. Stiffness matrix
K (t) in the stationery coordinates is as follows (Mayes and Davies, 1984, Vashisht et al., 2018).
cos(nt ) sin(nt )
K (t ) T K R (t )T 1, T (4.19)
sin(nt ) cos(nt )
For a two-degree-of-freedom rotor system, it is assumed that the 5 Kg mass disk is supported
by a massless rotor of stiffness 1x105 N/m. An eccentric (unbalanced) mass of 3x10-3 Kg lies on
the disk at a distance of 150 mm from the centroid. The damping coefficient of the system is 225
Ns/m. The stiffness of the stator is 10 times that of the rotor. The clearance of 15x10 -6 m ( ) is
provided between the rotor and stator. The natural frequency of the rotor is 22.5 Hz. When the
rotor and stator are in contact with each other, the natural frequency of the combined rotor/stator
system is 71.17 Hz. In this way, the rotor stiffness varies as a function of time. These systems are
called hybrid systems. The dynamics of these systems is inherently nonlinear in nature.
63 | P a g e
4.4. Detection of transverse crack and rubbing action
Responses of the system are evaluated at rotor spin frequencies of 30 Hz, 37 Hz, 47 Hz and
61 Hz. There is no frequency below the natural frequency of the rotor. Figure 4.4 shows the
spectrum and locus of the rotor at these frequencies. When the rotor is in the vertical position, the
effect of gravity is not considered. At 30 Hz of the rotor spin frequency, the response is forward
annular rub (FAR). The spectrum shows a single frequency in the spectrum i.e. the rotor spin
frequency. At 37 Hz, the response changes to partial forward whirl (PFW). The response contains
two frequencies, the rotor spin frequency and other components of 62 Hz (whirling frequency).
This frequency component changes with the rotor spin frequency. At 47 Hz, the spectrum contains
two frequencies i.e. the rotor spin frequency and 58 Hz (whirling frequency) component. The
amplitude of the system is moderate at this rotor spin frequency. This is a case of the partial
At 61 Hz, the amplitude increases dramatically. The system vibrates solely at a frequency of
73 Hz, near the natural frequency of the combined rotor/stator system. This is a dangerous state as
a dry whip (WHIP) is very harmful to the equipment due to its high amplitude vibrations
characteristics.
64 | P a g e
(a) Spectrum of vertical rotor response
Log10(Amplitude(m)) (b) Locus of the vertical rotor (c) Spectrum of horizontal rotor response (d) Locus of the horizontal rotor
Log10(Amplitude(m))
3 5 -2 -0.85
FAR FAR MHG MHG
2X
Amplitude/r0
Amplitude/r0
2 -0.9
-3
3X
1 0 4X -0.95
-4
0 -1
X=30Hz X=30Hz
-1 -5 -5 -1.05
0 50 100 150 200 -5 0 5 0 50 100 150 200 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06
Log10(Amplitude(m))
Log10(Amplitude(m))
3 5 -2 -0.8
PFW PFW X=37Hz MHG
Amplitude/r0
Amplitude/r0
-0.85
2 2X
62Hz -3 MHG
3X -0.9
1 0
4X -0.95
-4
0
X=37 Hz -1
-1 -5 -5 -1.05
0 50 100 150 200 -5 0 5 0 50 100 150 200 -0.06 -0.04 -0.02 0 0.02 0.04
Log10(Amplitude(m))
Log10(Amplitude(m))
3 5 -2 -0.92
PBW PBW MHG
58 Hz X=47Hz
Amplitude/r0
Amplitude/r0
-0.94
2
-3 MHG
-0.96
1 0 2X
-0.98
X=47 Hz -4 3X
0
-1
-1 -5 -5 -1.02
0 50 100 150 200 -5 0 5 0 50 100 150 200 -0.04 -0.02 0 0.02 0.04
Log10(Amplitude(m))
Log10(Amplitude(m))
5 20 -2 -0.7
73Hz WHIP WHIP MP MP
4 X=61 Hz
Amplitude/r0
Amplitude/r0
10 -0.8
3 -3
2 0 2X -0.9
1 -4
-10 X/2 -1
0
X=61 Hz 3X/2
-1 -20 -5 -1.1
0 50 100 150 200 -20 -10 0 10 20 0 50 100 150 200 -0.06 -0.04 -0.02 0 0.02 0.04
Frequency(Hz) Amplitude/r0 Frequency(Hz) Amplitude/r0
Figure 4.4 Comparison of vibration characteristics at different rotor spin frequencies for
vertical and horizontal rotors
4.5. Effect of different factors on rotor/stator rubbing action
The primary aim of this work is to establish the effect of gravity and rotor cracks on the
vibration characteristics of the overall system. Due to the presence of gravity, there is a static
deflection of 490x10-6 m. For the analysis of the rotor/stator system, =490x10-6 m+15x10-6m is
taken for the system. It has been observed through simulations that responses of the system at 30
Hz to 47 Hz rotor spin frequencies contain multiple harmonics that are an integer multiple of the
rotor spin frequency. We have named this state as the multiple harmonics generation (MHG). If X
is the spin frequency of the rotor, the spectrum of the horizontal rotor contains X, 2X, 3X…,
frequency components. Column (d) in Figure 4.4 shows the locus of the rotor. However, at 61 Hz,
65 | P a g e
the response of the system changes to multi-periodic (MP). At this rotor spin frequency, the
X 3X
response appears not only at X, 2X, 3X…, frequencies but at , ..., frequencies also. It
2 2
A breathing crack model is considered in this work. This action takes place in a situation
where weight dominance is present. The 30% crack depth is taken for the analysis. Again, the
vibration characteristics are investigated at 28 Hz, 47 Hz and 61 Hz rotor spin frequencies as shown
in Figure 4.5. All the frequencies are above the resonance frequency. At 28 Hz, the system behaves
as MP with the period 2 vibration. However, as it moves away from the resonance i.e. (22.5 Hz),
this behaviour diminishes and gradually converts into MHG. At 61 Hz rotor spin frequency, the
crack presence converts the vibration characteristics from MP to MHG as shown in Figure 4.5
with the presence of gravity only. For results in Figures 4.6, the clearance was settled as 15x10-
6
m, however, the vertical static deflection was considered.
66 | P a g e
(a) Spectrum of horizontal rotor with crack at 28 Hz (b) Spectrum of horizontal rotor with crack at 47 Hz (c) Spectrum of horizontal rotor with crack at 61 Hz
-1.5 -1.5 -1.5
-2 X=28Hz -2 -2
X=61 Hz
2X 2X
Log10(Amplitude(m))
-4 -4 -4
X=47Hz
3X/2
-5 -5 -5
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
Frequency (Hz) Frequency (Hz) Frequency (Hz)
(d) Locus of horizontal rotor with crack at 28 Hz (e) Locus of horizontal rotor with crack at 47 Hz (f)Locus of horizontal rotor with crack at 61 Hz
-0.7 -0.7 -0.7
-1 -1 -1
-0.09 -0.08 -0.07 -0.06 -0.05 -0.04 -0.03 -0.02 -0.01 0 -0.09 -0.08 -0.07 -0.06 -0.05 -0.04 -0.03 -0.02 -0.01 0 -0.09 -0.08 -0.07 -0.06 -0.05 -0.04 -0.03 -0.02 -0.01 0
Amplitude/r0 Amplitude/r0 Amplitude/r0
Figure 4.5 Effect of gravity and presence of crack on vibration characteristics of the horizontal
rotor (spin frequencies above the resonance)
Figure 4.6 Effect of gravity and presence of crack on vibration characteristics of the horizontal
rotor (spin frequency below the resonance)
67 | P a g e
Below the resonance frequency, for clearance of 15x10-6m, different types of responses are
possible for the horizontal rotor in the presence of a breathing crack and rubbing. The rotor was
spun at a frequency of 12 Hz. With the presence of rubbing, the response of the system is MHG
with a significant amplitude at first 6 harmonics as the integer multiple of rotor spin frequency
(Figure 4.6, c and d). With a breathing crack but no rubbing, MHG takes place as shown in Figures
4.6e and 4.6f. However, the first two harmonics have a significant amplitude. Under the effect of
both rubbing and transverse crack, the response is shown in parts g and h.
The secondary aim of this research is to develop a crack and rub diagnosing system based
analysis for non-stationary signals (whose parameters vary with time). Short-Time Fourier
Transform is one of the methods used for the time-frequency analysis to present the signal
(transient response) in a graphical format as a function of time. Normally, this graph is called the
spectrogram.
68 | P a g e
Figure 4.7 Vibration characteristics of the vertical rotor during coasting up as varied clearance
amplitudes
Figure 4.8 Vibration characteristics of the horizontal rotor during coasting up as varied
clearance amplitudes
69 | P a g e
Figure 4.7 shows the response, as well as spectrogram of the vertical rotor (without considering
gravity) during coasting up. With a 10-microns clearance, the vibration amplitude increases during
the time period of 6 to 12 seconds as shown in parts c and d of Figure 4.7. However, by further
reducing the clearance level to 6 microns, the vibration response converts into WHIP as shown in
parts e and f of Figure 4.7. The spectrogram of the response shows that after 16 seconds, the
whirling frequency of 72 Hz comes into picture and system vibrates at this frequency only,
For the vertical response of the horizontal rotor without cracks, the effect of gravity comes
into the picture shown in Figure 4.8. It is obvious that even with 10- and 4-microns clearance, the
system response never turns into a WHIP. Here the static deflection is considered due to gravity.
The system response is MHG and the higher harmonics at 2X, 3X.., appear in the system response,
which is obvious in the form of inclined lines (acute angle) in the spectrogram. These inclined
lines start appearing in the spectrogram at 4 seconds and remain up to 7 seconds for the 10 microns
clearance. By further reducing the clearance to 4 microns, these lines are available up to 23
seconds.
70 | P a g e
Figure 4.9 Vibration characteristics of the cracked horizontal rotor during coasting up as varied
clearance amplitudes (part 1)
Figure 4.10 Vibration characteristics of the cracked horizontal rotor during coasting up as
varied clearance amplitudes (part 2)
71 | P a g e
Figure 4.9 shows the response of the system with a breathing crack. In the first row of this
figure, the rubbing is absent. MHG takes place from the time period 4 to 7 seconds. Multiple
inclined lines appear in the spectrogram. This response appears similar to that obtained from
rubbing. However, there is a difference. The first few inclined lines start from zero in case of a
cracked rotor. If both the crack and rubbing, with a 40-microns clearance, are present in the system,
more and more inclined lines start from the origin. For the cracked rotor, if the clearance level is
further reduced to 10 and 4 microns (Figure 4.10), the number of the inclined lines further
increases. Apart from inclined lines at an acute angle, inclined lines at an obtuse angle with respect
to the horizontal also appear in the spectrogram in case the crack and rubbing co-exist. This aspect
Based on the simulated results, the following conclusions has been drawn.
The dry whip phenomena can easily be generated in the absence of gravity through simulations
but it is difficult to produce under the effect of gravity (even with extremely high unbalanced
mass).
Generation of the multiple harmonics is quite general in case of rotor/stator rubbing for a wide
range of rotor spin frequencies and clearance levels under the effect of gravity. Hence this
aspect can be used to detect the presence of rotor/stator rub in case of a healthy rotor based on
In the presence of a breathing crack, multiple harmonics are also generated in the vertical
response of a flexible horizontal rotor. But this feature is also very common in the presence of
rotor/ stator rub. Hence, the steady-state response of the system alone cannot be used to
distinguish between the two types of faults and there is a strong need to work on the analysis of
72 | P a g e
the transient response of a flexible rotor under coasting up, the situation in which the rotor spin
vibrations (during coasting up) presents several diagnosis features, which can be used to detect
the presence of multiple types of faults in rotating machinery. Multiple inclined lines starting
from zero time appear in the spectrogram of the vertical response of the cracked flexible rotor.
Multiple offset lines inclined with respect to the horizontal axis, starting at large distances
from zero time, appear in the spectrogram for the vertical response of the horizontal rotor in
In case of a cracked real rotor, when the rotor spin frequency reaches half of the resonance
frequency, the corresponding mode gets excited. This particular phenomenon is beautifully
captured in the spectrogram of the vertical response of the flexible rotor. An inclined line with
There is a need to improve the model by considering the Hertzian contact and viscous friction
(as well as the nonlinear friction model) between the rotor and stator in future. The inherent
nature of the considered system is the multi-physics and multi-scale. It involves dynamical and
thermo-mechanic responses, effects of a high speed of rotation and strong coupling between
dynamics and tribological characteristics. There is also a need to integrate the coupling of
thermal, mechanical and wear components. Although experimental results for simple Jeffcott
rotors have been compared to theoretical results through simulations, it is necessary to extend
73 | P a g e
Chapter 5
5. Active vibration control of rotor dynamic systems
5.1. Overview
An adaptive hybrid controller is proposed for reducing the unbalanced vibration response of
feedforward controller (AFFC) normally used in the active noise control are very sensitive in
performance for changes in rotor spin frequencies. Although frequency updating is a part of their
architecture, a small variation in the rotor spin frequency can reduce its effectiveness drastically.
A smart combination of the adaptive notch filter and Goertzel filter is proposed for the frequency
estimation. During changes of the rotor spin frequency, fundamental harmonics of the flexible
rotor are excited. By using a hybrid controller that combines feedback control and AFFC, the
harmonic hybrid control, the multiple harmonics generated due to coupling misalignment are
compensated efficiently. Fourier Transform of the control signal is further used to detect the
A flexible rotor system contains multiple rigid disks, multiple bearings with flexible bearing
supports (Figure 5.1a). Geometrical and physical parameters are described in Table 5.1. The
structure consists of certain discrete elements, 16 elements in the present case. The global mass
matrix (𝐌) and stiffness matrix (𝐊) for the structure of the rotor system can be derived based on
the geometry of the structure (Villa et al. 2008). The damping matrix (𝐂) can be derived from mass
and stiffness matrices. The effect of rotational spin frequency (Ω) of the rotor can be considered
74 | P a g e
via gyroscopic matrix (𝐆). Dynamic equations of motion of the complete system are as follows
where 𝐱, 𝐱̇ and 𝐱̈ are global vectors of displacement, velocity and acceleration. 𝐅𝐮 , 𝐅𝐀𝐌𝐁 and
𝐅𝐂𝐨𝐮𝐩𝐥𝐢𝐧𝐠 are vectors of forces generated by unbalanced masses, by active magnetic bearings and
at misaligned coupling, respectively. These coupling forces are as follows (Jalan et al. 2009).
:
0
FX sin( t)+FX sin(2 t)+FX sin(3 t)
Fcoupling (5.1b)
FX cos( t)+FX cos(2 t)+FX cos(3 t)
0
:
75 | P a g e
Figure 5-1 Rotor/bearing/coupling system
Details of these forces can be found in reference (Jalan et al. 2009 and Vashisht et al., 2019c).
Figure (5.1b) shows the parallel and angular misalignment in coupling. In this work, it is assumed
in the system. Apart from coupling misalignment, there are other parameters of the system which
76 | P a g e
can be changed during the service life of the rotor-bearing system. For example, the bearing
support stiffness plays an important role (parameters k1 and k6 in table 5.1, for the detailed model,
refer to chapter 3). The bolt looseness is an important factor which can reduce the bearing support
stiffness. With this, the natural frequencies of the rotor system are subject to change. So, there is a
Finite element models are of large order. To reduce the order of the overall system, modal
analysis is carried out. Equation (5.1) can be reduced into the following form.
̃ 𝐪̈ + (𝐂̃ − Ω 𝐆
𝐌 ̃)𝐪̇ + 𝐊
̃ 𝐪 = 𝚽 𝐓 (𝐅𝐮 + 𝐅𝐀𝐌𝐁 + 𝐅𝐂𝐨𝐮𝐩𝐥𝐢𝐧𝐠 ) (5.2)
𝐱=𝚽𝐪 (5.3)
modal analysis. By considering the first ‘r’ modes, Equation (5.2) in the absence of gyroscopic
effect (for a small variation of spinning speed, the change in gyroscopic effect is negligible) can
be written as follows.
𝐅𝐂𝐨𝐮𝐩𝐥𝐢𝐧𝐠 ) (5.4)
In the above matrices, there are ‘m’ number of sensors and ‘n’ number of actuators. r is the rth
natural frequency, r is the rth damping ratio and nr is the mode shape for the rth mode and nth
location.
77 | P a g e
Table 5.1 The physical parameters and mechanical properties of the rotor bearing system
For any structured uncertainty Δ, the stability of the overall system and the performance should
obey the following conditions (Gu et al., 2005, Vashisht et al., 2019c).
Robust Performance M
1 for uncertainty Δ (5.6)
Nominal Performance M 22
1 (5.8)
78 | P a g e
Nominal Stability M is internally stable
For the robust stability and robust performance, it is required to find a stabilizing controller K such
that:
sup M ( P , K )( j ) 1 (5.9)
R
A D-K iteration method for µ-synthesis is used (Gu et al. 2005). It is based on solving the final
The D-K iteration method for µ-synthesis is as follows (Gu et al., 2005).
D( j ) inf DFl ( P , K ) D 1 ( j )
D
4. Curve fit D ( j ) to get a stable minimum phase D(s). Go to step 2 and repeat until a
Effect of impulsive disturbances can be reduced using a feedback type of controllers. In the
case of periodic disturbances, feedforward controllers are suitable. Let m be the mass of the
unbalanced part, be the distance of the unbalanced mass from the centre of rotation and 1 is
rotational frequency of the rotor in radian per second, the unbalanced force is given by:
F m 12 (5.11)
79 | P a g e
This force will be acting in the horizontal as well as vertical directions with phase angle such
that in discrete-time it can be presented as follows (Shimada et al. 1999, Vashisht et al., 2019c).
n n
d (n) (n) cos 1 (n)sin 1 (5.12)
Fs Fs
where Fs is the sampling frequency and n is the instant of time. If u(n) is the control force, the
where G(z) is the transfer function in z-domain. The error signal will be zero when the estimate of
d(n) is equal to u (n) i.e. u (n) dˆ (n) , dˆ (n) is the estimate of disturbing force as follows.
n n
dˆ (n) ˆ (n) cos 1 ˆ (n)sin 1 (5.14)
Fs Fs
For updating coefficients, the instantaneous cost function J e(n) 2 is used. If w(n) at time n is
T
defined as w(n) ˆ (n) ˆ (n) the gradient of the cost function J becomes (Shimada et al.
1999):
1n
2e(n) cos
J Fs
( n) (5.15)
w(n)
2e(n) sin 1n
Fs
Such that
1n
ˆ (n) ˆ (n 1) 2 e(n) cos
Fs
(5.16)
n
ˆ (n) ˆ (n 1) 2 e(n) sin 1
Fs
80 | P a g e
where is the updating coefficient and is the phase angle of the flexible rotor system at
frequency 1 . In case the frequency 1 is not certain or time-varying, it must be updated at each
instant of time. Parameter updating equations are modified as follows (Shimada et al. 1999,
e(n)
1 (n) 1 (n 1) 2 1
s
F
ˆ (n)sin (n 1) ˆ (n) cos (n 1) (5.18)
1 (n)
(n) (n 1) (5.19)
Fs
( n) (n)
u (n) ˆ (n) cos (n 1) 1 ˆ (n)sin (n 1) 1 (5.20)
Fs Fs
Adaptive Notch Filter (ANF) is a second-order adaptive filter used to estimate the frequency
of a sinusoidal signal. Dynamic equations of the filter are as follows (Yoon et al. 2017)
where x(t) is the filter state, 1 (t ) is the natural frequency of the filter or the frequency to be
estimated for the signal e(t), is the damping ratio of the filter, is the estimation gain. The
estimated frequency 1 (t ) converges to a fixed value when the right-hand side of Equation (5.22)
becomes zero. This presentation is for a continuous-time system. However, for the real-life
81 | P a g e
5.6. Results with conventional Adaptive Feedforward controller
Section 5.4 explains conventional AFFC. The schematic is shown in Figure 5.2 (a). Figure 5.3
shows the response of the system with AFFC. The rotor spin frequency is fixed at 10 Hz. Equation
(5.16) is used to apply AFFC. If the rotor spin frequency is kept at 19.995 Hz, the performance of
the closed-loop system deteriorates. If the rotor spin frequency is further changed to 19.95 Hz, the
control effectiveness becomes nearly zero (not shown for the purpose of brevity). This aspect
shows that the effectiveness of AFFC is fully dependent on the frequency. If the frequency tracking
regenerated. If the frequency gap is increased i.e. the difference between the actual frequency and
starting frequency, convergence takes a lot of time. Up to a certain extent, convergence speed can
Based on the real-life scenario, the rotor spin frequency is increased from 20 Hz to 25 Hz.
This increase in value starts at 20 second and ends at 30 second. The conventional AFFC with high
adaptation coefficients is used. It is obvious that the conventional AFFC cannot cope up such a big
Obviously, the next choice is to use frequency estimators in combination with conventional
AFFC. Adaptive Notch Filters are best suited for the frequency estimation. However, they have
limitations in case of the low-cost instrumentation where low sampling rates are available. At
higher sampling frequencies (in several kHz), the error is virtually zero. However, at 200 Hz, the
error in estimation is quite large especially at higher frequencies as shown in Figure 5.5.
82 | P a g e
Rigid Disks Unbalanced
Rigid Disks Unbalanced
mass
mass
Flexible rotor Flexible rotor
Error Error
sensor sensor
Control point AMB
Control point AMB
u(n)
e(n)
u(n) e(n)
Feed forward Secondary
control Feed forward Feedback
frequency
control control
estimator
Adaptation
Adaptation
(a) Conventional adaptive feed forward control (b) Proposed hybrid control
83 | P a g e
Figure 5.4 Time domain response by Adaptive Feedforward Control with time varying frequency
component applied to rotor system
Figure 5.5 Comparison of actual and estimated frequencies by Adaptive Notch Filter at sampling
frequency of 200 Hz
84 | P a g e
The Goertzel filter-based computation of Discrete Fourier Transform (DFT) is free from these
errors. Very accurate estimation is made with this filter. Based on the sequence length of the
signal, the desired resolution in the frequency domain is possible. An online version of the Goertzel
filter is also available (Manngård et al. 2017). However, the computational load of these filters is
quite heavy. They are not suitable for the online frequency estimation with low sampling rates
(e.g. 200 Hz). The combination of ANF and Goertzel filter is beneficial in this situation. Hence
both the frequency estimators are used to design the proposed AFFC.
From Figure 5.4 it is obvious that when the frequency of a signal changes (although at a
moderate rate), the fundamental harmonics are excited. Their amplitude is much higher than the
The proposed AFFC uses the ANF and Goertzel filter (Figure 5.2b). The flowchart shown in
Figure 5.6 describes the complete architecture of the proposed controller. ANF is used for online
frequency estimation. If the rotor spin frequency of the flexible rotor is changed, ANF can detect
this change. There is a steady-state error in the estimation of the frequency but the frequency
change can be well detected. The number of time steps N is fixed based on the general experience.
If frequency variations are frequent, N is small and vice-a-versa. During frequency changes, AFFC
is totally useless in terms of vibration reduction. Hence, the controller is switched off during
If the frequency variation continues, the controller remains switched off and data points are
collected. ANF is used to detect the frequency variation and rough frequency estimation only.
85 | P a g e
When the frequency variation is ceased, the Goertzel filter roughly estimates the current frequency
of the signal (in the vicinity of the estimate made by ANF). When the rough estimate is completed,
the resolution of the Goertzel frequency can be refined. Discrete Fourier transformation (DFT)
coefficients are calculated in the vicinity of the rough estimate made by the Goertzel filter. Not all
of the DFT coefficients need to be calculated. This reduces the computational burden. After the
refined estimate of the current frequency is done, this frequency can be used as an initial point for
86 | P a g e
Start
Set the adaptation coefficients
Feedback receiver
Feedback receiver
Set k=1
CONTROLLER ON
Apply control input
Increment k= k+1
Yes Is
k<N ?
No
No Is
f(N)-f(1)> delta ?
Yes
No
Is
Shut down requested ?
Yes
Stop
87 | P a g e
5.7.2. Mu-synthesis based Robust Feedback Control
wp I GK I GK
1 1
inequality 1. Singular values of sensitivity function over the
1
frequency range must lie below (Gu et al., 2005). This is achieved by using MATLAB’s
wp
Robust control toolbox. From the frequency response function of the flexible rotor system, it is
obvious that the first two modes are dominant. Figure 5.7 shows the frequency response of the
nominal system, required model of the system, control weighting function and performance
weighting function. Singular values of the closed loop system are shown in Figure 5.8. It is obvious
that the closed loop system achieves the nominal performance as well as robust performance
(Figure 5.8). Controller order reduction is completed before implementation. The order of the full
88 | P a g e
Figure 5.8 Robust stability, performance and other parameters
The hybrid controller is shown in Figure 5.2 (b) is used in this work. The proposed controller
is a combination of the feedback and AFFC controller. The proposed controller can attenuate
multiple harmonics at a time. In the case of coupling misalignment, multiple harmonics that are
integer multiple of rotor spin frequency are developed. The proposed hybrid controller shown in
89 | P a g e
Rigid Disks
Unbalanced
mass
Flexible rotor
SENSOR AMB
AFFC 1
AFFC 2
Secondary
Feedback
frequency Adaptation
control
estimator
AFFC 3
AFFC 4
5.8. Experiment
The experimental setup contains the following parts as shown in Figure 5.10.
For acquiring data, a USB based card NI-USB-6212 from National Instruments is used. It is
compatible with software LabView to acquire data as well as signal generation. Its input and
output ranges are from -10 volts to +10 volts. For acquiring the data only, the sampling rate
of a few kHz can be easily obtained. However, for the controller implementation and
calculating the control signal, a certain time is consumed. This restricts the sampling rate to
lower values. In this research, the 200 Hz sampling rate is fixed. All the calculations for the
controller implementation are processed in this period of time. However, using real-time
engines, much higher sampling rates can be obtained but the high cost of these real-time
90 | P a g e
engines restricts their applications. The cost of National Instruments based PXI- Real-time
engines varies from 10 to 15 times for the NI-USB-6212 data acquisition card.
Eddy current probes from Bently Nevada are used as sensors. The gap up to a few microns
can be easily measured with this sensor system. It has 8 volts voltage drop per millimetre of
gap.
An AC amplifier is used to increase the amplitude of current signals emerged from the
computer.
For generating control forces, a home-made electromagnetic actuator is used. Details are
Rotor
AMB AC amplifier
Figure 5.10 Experimental setup for active vibration control for rotor system
91 | P a g e
Here i0 represents the bias current and i is the control current. The force produced by this current
is given by:
0(𝑖+𝑖 )2 0 (𝑖−𝑖 )2
𝐹𝐴𝑀𝐵 = 𝑘 [(𝑔𝑎𝑝−𝑥) 2
− (𝑔𝑎𝑝+𝑥)2
] (5.23)
where x is the displacement of the workpiece from the equilibrium position, ‘gap’ is the distance
between the core and the workpiece under equilibrium conditions. The value of the coefficient k
is given by:
1
𝑘 = 4 𝜇0 𝑛 2 𝐴 (5.24)
where 𝜇0 is the permeability of the air, n is the number of winding turns and projected area of the
core. The value of different parameters of electromagnetic actuators is given in Table 5.2.
5.9. Results
extensively used to remove the steady-state vibrations from literature. However, this technique did
not work in the present rotor system. A proportional and simple derivative controller is proposed
to reduce the steady-state vibrations in the present rotor system with multiple flexible modes. This
technique works well for rigid rotors as well as simple Jeffcott rotor system with the single
dominant mode. However, in the presence of multiple modes, the efficiency reduces drastically.
92 | P a g e
The next question arises regarding the application of advanced robust control techniques like
Loop Shaping and Mu-synthesis controller. The performance comparison is done between Mu-
synthesis and a derivative type feedback controller. If the high gain derivative controller was
applied, the requirement of control forces raises to a high value. The value of gain is chosen in
such a way that the response with this controller matches exactly with that obtained by Mu-
controller. If the low gain derivative controller is used, the control force requirement is
comparatively less. However, with this controller, there is an initial rise in response.
Another advantage of using the Mu-controller is that the vibration response at Disk 2 (notice
that control actuator is applied at Disk 1) is less than that obtained by using a derivative type
feedback controller (Figure 5.11). At lower gains, the amplitude difference is quite significant
(30% fall). These factors prove the superiority of the Mu-controller compared to Derivative
feedback controller. The same trend follows for PID type feedback controllers. Using robust spatial
Figure 5.11 Comparison of Mu- controller and Derivative controller for Disk 2
93 | P a g e
For the application of the proposed hybrid controller, simulation results show open and closed-
loop responses of the rotor system rotating at 20 Hz frequency. At 20 s, the spinning speed starts
increasing and it increases up to 25 Hz in next 10 s. From 20-30 s, the transient vibrations take
place. Figure 5.12 shows the experimental response of the rotor system along with control forces
in terms of amplified control voltages. Since AC voltage amplifier is used, it becomes easy to
measure the control voltage. With the change of temperature, parameters of the electromagnetic
actuator (EA), as well as voltage amplifier (VA), are subjected to change. One of the biggest
advantages of the AFFC controller is that it can take care of changing parameters of EA and VA.
Figure 5.12 Response of rotor system with the proposed hybrid controller (Experimental)
The AFFC controller is efficient for a particular rotor spin frequency. In the case of coupling
misalignment, multiple frequency components (an integer multiple of the rotor spin frequency)
appear in the spectrum. The multi-harmonic hybrid controller shown in Figure 5.13 is used for this
purpose. The frequency of the first harmonic is estimated only. Figure 5.13 shows the open and
closed-loop responses of the rotor system with coupling misalignment at 20 Hz rotor spin
94 | P a g e
frequency. The response contains 40 Hz and 60 Hz signals also. With applications of the proposed
controller, the amplitude at all the harmonics reduces considerably (Figure 5.13 (a)). Figure 5.13
(b) shows the spectrum of applied control signals. It clearly shows three harmonics i.e. 20 Hz, 40
Hz and 60 Hz.
It is obvious that the feedback controller is nearly useless at other frequencies (i.e. other than
natural frequencies). AFFC is suitable for these frequencies. By using the combination of multi-
harmonic AFFC and robust feedback control (i.e. hybrid control), the magnitude reduction at
-10
0
Amplitude (Log10(m))
Amplitude (Log10(N))
-12
-2
-14
-4
-16
-6
-18
-20 -8
0 50 100 150 0 50 100 150
Figure 5-13 The response of the rotor system with coupling misalignment using proposed
controller (Experimental)
95 | P a g e
5.10. Concluding remarks
results have authenticated the superiority of the proposed method. Based on the results following
Conventional AFFC cannot tolerate sharp rotor spin frequency changes. External frequency
ANF provides unacceptable results at low sampling rates, however, it can be used to indicate
Goertzel filter is computationally demanding, however, it can be used efficiently when a small
Combination of ANF and Goertzel filter can be used efficiently to estimate changing rotor
spin frequencies.
96 | P a g e
Chapter 6
6. Precision control of ball screw drives
6.1. Overview
This chapter presents a hybrid controller by combining the interpolating gain scheduled
loop shaping (IGSLS) controller and cascade PPI controller. It achieves active vibration damping
as well as high bandwidth disturbance rejection for a ball screw drive system (BSD). The IGSLS
controller is designed based on loop shaping principles for the non-collocated system from the
motor torque to carriage/table position. It is observed that by increasing gains of controllers, the
closed-loop stability margins are reduced although the command tracking efficiency increases.
High gain controllers also cause a large variation in motor torque when BSD is used in the
machining process. From time and frequency domain analysis, it is found that the proposed hybrid
controller has high command performance efficiency along with better performance during the
machining process. For machine drives in the absence of pre-loading, the system parameters vary
over a large range, depending upon the carriage/table mass and location. The BSD system can be
modelled as a linear parameter varying (LPV) system using the mass and location of the carriage
as scheduling parameters. It is also observed that fixed gain robust controllers cannot maintain the
closed-loop stability for the total range of system parameters of the BSD. Based on the transient
response, the carriage mass can be estimated online with sufficient accuracy. Using this
information and table location, IGSLS controllers can be implemented in real-time using
interpolation techniques. Simulations and experimental results confirm the performance of the
proposed controller in command tracking, the variation of motor torque during machining and
disturbance rejection.
97 | P a g e
6.2. Mathematical Modelling
The ball screw drive is actuated by a servo motor. The torque is applied through the flexible
coupling. The rotary motion of the ball screw is converted into translational motion of the carriage.
The flexibility of the carriage plays an important role in the overall dynamics of the system.
Figure 6.1 shows a two-dimensional representation of the ball screw drive. The balls and the
lead screw are flexible. The stiffness of the ball near the left-hand side stopper has a stiffness 𝑘𝑏1
and stiffness of the ball near the right-hand side stopper 𝑘𝑏2 . The stiffness of the coupling is 𝑘𝑐 .
The stiffness of the carriage balls is 𝑘𝑛 . The total length of the lead screw is L. The lead of the lead
screw is l. If 𝜃(𝑡) is the actual angle of rotation of the motor (𝑥1 (𝑡) is the virtual displacement),
T(t) is the actual torque applied at the motor (𝐹1 (𝑡) is the virtual force), they are related to each
other with the following expressions as shown in Figure 6.1 (Varanasi and Nayfeh, 2004 ).
Figure 6.1 represents the various stiffnesses related to the ball screw drive. The equivalent
stiffness of the overall system is given as follows (Varanasi and Nayfeh, 2004).
1
1 1 l 1
2
c
kt (6.3)
k1 k2 kn 2 kc GJ
where c is the distance of the carriage from the left-hand side stopper, G is the modulus of rigidity,
J is the polar moment of inertia, kn is the stiffness of balls near the carriage (on the left and right
1
c 1
k1 (6.4)
EI kb1
98 | P a g e
1
Lc 1
k2 (6.5)
EI kb 2
where E is the modulus of elasticity, I is the bending moment of inertia, kb1 and kb2 are the
stiffnesses of balls on the left and right sides of the ball screw system. The details of modelling
can be seen in reference (Varanasi and Nayfeh, 2004). This modelling uses a distributed parameter
99 | P a g e
x1 x2
kt
F1 F2
m11 m22
d1 d2
c1 c2
Let x2 (t) be the displacement of the carriage and F2 (t) be forces acting on the carriage, using the
Galerkin type weighted residual method (Varanasi and Nayfeh, 2004), the dynamic equations of
x (t ) x (t ) x (t ) F (t )
M 1 C 1 K 1 1 (6.6)
x2 (t ) x2 (t ) x2 (t ) F2 (t )
Where
m m12
M 11 (6.7)
m21 m22
With
m11 n 2 ( J J11 ) m
m12 n 2 J12 m
(6.8)
m21 m12
m22 n 2 J 22 m mc
where J is the second polar moment of the screw, mc is the mass of the carriage and the value of
other parameters can be found in reference (Varanasi and Nayfeh, 2004). The damping matrix of
n2C 0 c1 0
C 1 (6.9)
0 C2 0 c2
100 | P a g e
Where C1 is the motor damping and C2 is the damping coefficient between slides and the carriage.
1 1
K kt (6.10)
1 1
It is obvious that the overall stiffness of the ball screw system is dependent on parameter c.
Hence the dynamic system represented by Equation (6.6) is dependent on parameter c. In the
transfer function of the system dynamics can be represented as follows (Figure 6.2) in the form of
F1 ( s)
x1 ( s) G11 ( s) G12 ( s) G13 ( s)
x ( s) G ( s) G ( s ) G ( s ) d1 ( s) (6.11)
2 21 22 23 d (s)
2
Where
x1 ( s ) x ( s ) m22 s 2 c2 s kt
G11 ( s ) G12 ( s ) 1
F1 ( s ) d1 ( s ) ( s )
x1 ( s ) k
G13 ( s ) t
d 2 ( s) ( s)
(6.12)
x (s) x (s) k
G21 ( s ) 2 G22 ( s ) 2 t
F1 ( s ) d1 ( s ) ( s )
x2 ( s ) m11s 2 c1s kt
G23 ( s )
d 2 ( s) ( s)
( s ) m11m22 s 4 a1s 3 a2 s 2 a3 s
a1 m11c2 m22 c1
(6.13)
a2 c1c2 (m11 m22 )kt
a3 (c1 c2 )kt
Parameters in Table 6.1 are used for conducting simulations as well as experiments. The
material of the ball screw is steel with Modulus of Elasticity 210x09 N/m2, Modulus of Rigidity
101 | P a g e
78x09 N/m2 and density 7850 Kg/m3. The carriage mass is allowed to vary from 25 Kg to 45 Kg.
The damping coefficients C1 and C2 are measured experimentally as 0.001 and 0.01, respectively.
5 Left hand side ball stiffness at the start of ball kb1 50 MN/m
screw
Based on Equation (6.6), the state-space model of the system can be expressed as follows.
x1 x1
x x
2 2 F1
A B
x1 x1 F2
x2 x2
(6.14 a, b)
x1 x1
x x
2 2
C
x1 x1
x2 x2
where C is a unity matrix. F1 and F2 are control forces at the motor and carriage. A and B are
102 | P a g e
0 I
A 1
M K M 1C
(6.15 a, b)
0
B 1
M
As the carriage load is increased, the first resonance frequency of the BSD is decreased. Under
certain circumstances, the preload applied to the BSD may vanish also. The ball bearing clearance
plays a major role in this regard. If the clearance on the right-hand side ball bearing increases,
there is a drastic change in the resonance frequency of the BSD as shown in Figure 6.3 (because
now it is in the form of a fixed free bar instead of a fixed-fixed bar). As the carriage mass changes,
Figure 6.3 Effect of location and mass of the carriage on the first natural frequency (with
bearing clearance)
Equation (6.6) shows the dynamic model of the BSD. Elements of the mass and stiffness
matrix vary as a function of location and mass of the carriage. It is obvious that the dynamics of
103 | P a g e
x1 x1
x x
2 2 F1
A( m BSD BSD
,l ) B( mBSD ,lBSD )
x1 x1 F2
x2 x2
(6.16 a, b)
x1 x1
x x
2 2
C
x1 x1
x2 x2
where mBSD and lBSD are the mass and location of carriage from the left-hand side respectively, and
C is a unity matrix. Matrices A( mBSD ,lBSD ) and B( mBSD ,lBSD ) are as follows.
0 I
A( mBSD ,lBSD ) 1 1
M( mBSD ,lBSD ) K( mBSD ,lBSD ) M( mBSD ,lBSD ) C
(6.17 a, b)
0
B( mBSD ,lBSD ) 1
M( mBSD ,lBSD )
The next section shows that any fixed-gain controller cannot meet stability requirements for this
LPV system. Hence, an LPV controller (in this case it is IGSLS controller) is designed and
There are different techniques for tracking the desired trajectory. Robust control, adaptive
control and model-free control techniques can be used for desired trajectory tracking. In the next
sub-sections, we will explain cascaded PPI controllers, fixed-gain loop shaping and IGSLS
controllers.
It is a model-free controller (not in the strict sense). Here P is the proportional controller and
I is the integral controller. P controller is applied for the carriage position feedback. Combination
of P and I controllers is applied for the motor speed feedback. The efficiency of PPI can be further
104 | P a g e
enhanced with a table/carriage velocity feedback. This idea has been implemented recently and is
state of the art in case of model-free controllers. The main disadvantage of such a control system
is that there is a need for additional sensors. Even then, these controllers are extensively used in
the industry. These controllers are extremely robust. However, there is a limit to the performance
of these model-free controllers. Model-based controllers are rich in terms of performance, but their
robustness is limited in case of large plant/system parameter variations. Coefficient Kp1 for the
carriage position feedback is 12x106. Coefficients Kp2 and Ki for the motor velocity feedback are
34000 and 1000 respectively. The coefficients are determined by optimization in the frequency
Practical systems deviate considerably from the nominal model derived using system physics.
There are uncertainties entering the nominal system. These uncertainties enter the nominal system
the robust stability of a closed-loop system, one must minimize follows (Gu et al., 2005).
Theoretical details for synthesizing the controller can be found in reference (Gu et al., 2005).
In the present analysis, the weight of the carriage is assumed to vary from 25 Kg to 45 Kg
with grid points [25, 30, 35, 40, 45] Kg. The carriage travel length varies from 20 mm to 920 mm
with grid points [20, 100, 200, 300, 400, 500, 600, 700, 800, 920] mm. Overall 5 x 10 =50 grid
points are generated at which the Loop shaping controller is designed. The compensator with
105 | P a g e
2s 1 2s 1
transfer function 1x107 is chosen for the low gain controller and 1x108 is
0.9 s 0.1 0.9s 0.1
chosen for the high gain controller. For convenience, 1x108 is termed as a loop shaping factor in
the rest of the chapter. Now a general procedure is described for interpolation of the controller.
Any linear parameter varying (i.e. IGSLS) system for interpolation can be represented as follows.
A B
G (6.19)
C D
parameter. Details can be found in reference (Gu et al., 2005). The above algorithm is implemented
Figure 6.4. There are two scheduling parameters viz. the mass and location of the carriage. The
carriage mass can be identified from the transient response of the system.
106 | P a g e
Figure 6.4 Simulink implementation of Hybrid controller
6.4. Experiment
A single-stage ball screw drive shown in Figure 6.5 is used. A home-made ball screw with
16 mm pitch and 950 mm length is used. Guideways are used to support the carriage. The servo
motor is driven by a pulse width modulator amplifier. The carriage position is measured by a linear
encoder with a resolution of 100 nm. An analog DC accelerometer (PCB 3711E1110G) with a
bandwidth of 1000 Hz is attached to the carriage acceleration. In addition, a rotor encoder with
resolution 0.001 degrees is used to measure the angle of rotation. A DSpace board DS-1103 is used
to implement the controller. The stiffness of balls of the bearing and carriage nut is calculated
107 | P a g e
Figure 6.5 Experimental setup for ball screw drive
experimentations, are conducted. The performance of the BSD system for various control
6.5.1. Performance comparison of fixed gain loop shaping design procedure (LSDP)
controller with Cascaded PPI controller
Frequency domain analysis plays a vital role in accessing the stability of any closed-loop
system. The loop transfer function, Nyquist stability criteria, sensitivity function and cross-over
frequency can be used to design controllers based on frequency-domain results. For a fair
comparison between LSDP and PPI controllers, these controllers are tuned to possess similar
stability margins. In the industry, this limit is normally held between 1.5-2. Figure 6.6 shows the
sensitivity magnitude with different controllers. This figure is also used to represent the bandwidth
108 | P a g e
Figure 6.6 Loop sensitivity with different controllers
With the PPI controller, disturbances at the motor input can be effectively suppressed from 0-
40 Hz and >220 Hz. The LSDP controller can suppress input disturbances from 0-35 Hz and >220
Hz. However, with the hybrid controller, the limit is increased up to 0-55 Hz and >220 Hz (Figure
6.6 a). This means the hybrid controller has more overall bandwidth than the other counterparts.
However, disturbances at input get amplified above 230 Hz if the shaping factor is increased
above 1x107 (Figure 6.6b). It can reduce input disturbances above 375 Hz only (this limit varies
based on shaping factor used). Although by increasing the shaping factor above 1x10 7, its
bandwidth increases for lower frequencies i.e. 0-70 Hz only, disturbances at input get amplified in
the range 240-375 Hz. It can reduce or mitigate input disturbances above 375 Hz only. Hence,
109 | P a g e
Figure 6.7 Loop magnitude with different controllers and loop shaping factor
The loop transfer function for the PPI controlled system is given by:
K p 2 s Ki
L
s
G
11 ( s) s G21 ( s) K p1 (6.20)
Based on the loop magnitude results shown in Figure 6.7a, it is clear that the tracking ability
of LSDP is better than PPI controller due to higher low-frequency gains. The cross-over
frequencies are the same (40 Hz-230 Hz). This means the controllers can actively damp the axial
vibrations of the BSD in the same range. The hybrid controller has high crossover frequencies i.e.
There is a limit to the performance of PPI controlled systems (even after controller parameter
optimization). But this limit can be extended in case of the LSDP controlled system by increasing
the loop shaping factor (Figure 6.7b). However, additional problems are developed by increasing
the shaping factor as will be seen shortly during a metal cutting operation, i.e. when the dynamic
One thing worth mentioning here is that as the carriage is moving away from the motor, the
axial resonance frequency of the BSD decreases. The PPI controller is robust in terms of the
closed-loop stability for a large range of parameters. However, LSDP is less robust in terms of
110 | P a g e
closed-loop stability. The stability of the LSDP controller is also dependent on the loop shaping
factor. For higher loop shaping factor values, the LSDP controller is less stable (in terms of the
The open-loop transfer function from table disturbances d2 to carriage position x2 is given by
transfer function G23 (s). The closed-loop disturbance transfer function for the system with the PPI
Gd2 x2
G23 K p2 s Ki G11G23 G13G21
(6.21)
1 L
The hybrid controller is best in achieving lower level amplitude of vibrations for the external
disturbances acting at the carriage (normally the cutting forces). For the time domain command
tracking performance of different controllers, Part a of Figure 6.8 shows the reference trajectory
to be achieved. From this figure, it is obvious that there is an abrupt start. This type of trajectory
is intentionally chosen to quantify the effectiveness of the different control algorithms for non-
smooth trajectories. With the PPI controller, there is a less transient error at the start. The
movement of the carriage is 150 mm from the start (from the motor end). Hence there is very less
change in system parameters up to this carriage movement. Although model base controllers are
high in performance for the robust stability, parameters of the system should remain within a
tolerable limit. For example, the mass of the carriage varies based on the mass of the workpiece
mounted on it. As discussed earlier, the dynamics of the system depends on the location of the
carriage also. Based on these factors, it is a general practice in the industry to use model-free
controllers like cascaded PPI controllers that are robustly stable for very large values of
carriage/table mass and location. To compare the performance of the model-free controllers with
model-based controllers, the trajectory is designed in such a way that the carriage remains near the
111 | P a g e
motor side (maximum amplitude 150 mm). In this way, the effect of change of dynamics with the
Figure 6.8 Time domain analysis for command tracking with different controllers (Simulation
results)
The maximum error between the reference and achieved trajectory is 115 microns (after initial
transient errors are over) with the PPI controller. With the low gain LSDP (if used alone), the error
is much more. However, with the hybrid controller, the error is 73 microns (Figure 6.8). By using
a high gain LSDP controller, the trajectory error further reduces to a small value. But, as discussed
earlier, higher controller gains are not suitable when BSD is undergoing the metal cutting process.
In the above figures, the efficiency of the controllers is tested in the absence of metal cutting
forces (i.e. rapid movement of the BSD). For testing the efficiency of the controller during the
metal cutting process, a milling process was performed. A milling cutter of diameter 19.04 mm
was used for metal cutting. The spinning speed of the milling tool was fixed at 2150 r.p.m. The
milling tool has only one tooth to avoid the problem of runout. The feed rate of the BSD was fixed
at 3.58 mm/s. The tooth pass frequency of 2150/60=35.8 Hz was realized during experimentation.
112 | P a g e
Multiple harmonics that are integer multiple of tooth pass frequency are present in the spectrum.
Figure 6.9 shows the performance of various controllers. Parts a, b and c represent the table
position error, motor velocity and torque applied at the motor respectively with low gain LSDP
controller (with shaping factor 1x107). The performance of the LSDP controller and cascaded PPI
controller is almost the same in terms of table position error, motor velocity variation and applied
torque (Parts d, e and f). With the hybrid controller, the table position error is relatively low and
motor velocity variation is minimum (Parts g, h and i). The table position error was also a
minimum.
Parts j, k and l (Figure 6.9) show that if a high gain LSDP controller is used (with shaping
factor 1x108), the table position error is maximum, motor velocity variation is maximum as well
as applied torque amplitude varies between extreme limits. The torque amplitude variation is 4
times as compared to the low gain LSDP controller (with shaping factor 1x107).
The variation is harmful to the motor life. The energy efficiency also decreases with an
increase in torque amplitude variation. Hence, there is a limit for increasing the shaping factor (in
other words the controller gains). It is obvious that to increase the overall efficiency (i.e. BSD
motion during metal cutting and rapid movements), a combination of cascade PPI and low gain
Afterwards, to check the robustness of PPI and LSDP controllers in the time domain, both
controllers are applied to the BSD when the starting point of the reference trajectory is 350 mm
from the start (i.e. motor side). The PPI controller maintains closed-loop stability under the
changing dynamics. However, The LSDP controller becomes unstable. There is the continuous
chattering of the carriage around some nominal points. The magnitude of chattering depends upon
113 | P a g e
the saturation limit of the motor. Here the need of the IGSLS controller arises i.e. the system should
Figure 6.9 Performance of various controllers during metal cutting (Experimental results)
It is obvious that as gains of the controller are increased (i.e. using a high loop shaping factor),
the stability margins of the closed-loop system for the total range of system parameters are
114 | P a g e
decreased. However, the system performance improves with gains of the controller (for rapid
movement of BSD). The stability of the closed-loop system also depends on various other factors.
These factors include the unmodelled dynamics, accuracy with which the system parameters are
modelled and the actuator nonlinearities. Depending on the accuracy limit of the carriage/table
mass estimation, the loop shaping factor plays an important role. In the present work,
2s 1
1x107 is fixed as the loop shaping factor for the controller design. The closed-loop
0.9 s 0.1
stability is retained even if the carriage mass is estimated as 30 Kg instead of 25 Kg (i.e. exact
value).
In the next simulation, the carriage or table is moved from the extreme left to extreme right
(l=920 mm). To prevent sudden acceleration, the desired trajectory was filtered out by curve
fitting. The curve fitting was done by 7th order sinusoidal such the displacement as a function of
time. Figure 6.10 shows the performance of different controllers. The carriage mass of 25 Kg was
taken for the analysis. The PPI controller is stable for the total travel of the carriage. However, the
performance is not satisfactory as a large difference between reference and obtained trajectory.
With the PPI controller, the maximum trajectory error is 66 microns (after the transients are over
for 0-0.08 seconds). With the IGSLS controller (if used alone), the error is high and with the hybrid
controller, the error is 43 microns, i.e. 54% increase in the command tracking efficiency.
115 | P a g e
Figure 6.10 Comparison of performance with LPV and other controllers (Simulation results)
Based on time and frequency domain results following conclusions can be drawn.
Robust stability of LSDP controllers cannot be maintained for a large variation of system
parameters.
Cascaded PPI controllers are robustly stable; however, their command tracking performance
High gain LSDP controllers cause large variations in the motor velocity and torque applied at
the motor. This factor reduces the useful life of the motor and enhances energy losses at the
motor. Hence controller gains cannot be increased arbitrarily. The only option left to increase
the command tracking performance (during rapid movement of BSD) and decrease torque
variation (during metal cutting operation) is to use hybrid controllers. Command tracking
performance of the hybrid controller made by joining PPI and LSDP controllers in parallel is
much better than the case when they are used individually.
116 | P a g e
The bandwidth and performance (during rapid movement and metal cutting) of the hybrid
controller is much higher than PPI and LSDP controllers (if used alone).
Using interpolation-based IGSLS controllers, the closed-loop stability and performance can
be maintained for the complete range of parameters. The performance and bandwidth are
further enhanced by connecting the PPI controller in parallel with the IGSLS controller (i.e.
117 | P a g e
Chapter 7
7. Active chatter control of turning operations
7.1. Overview
The stability requirement in turning operations limits the cutting depth. The common trend is
to generate stability lobe diagrams to avoid chatter regions. However, flexibility of shaft results in
the low material removal rate due to the limited stable depth of cut. Normally the workpiece is
considered rigid and active chatter control techniques are applied to reduce vibrations of flexible
cutting tools. In this work, both the workpiece and cutting tool are considered flexible in
appropriate directions. The novelty of the work is the combination of theory and practical
applications. As the material is removed during cutting, natural frequencies of the shaft are
changed. Fixed parameter controllers lose their effectiveness under these changing conditions.
Linear Parameter Varying controller is designed and implemented so that the cutting stability can
be maintained in the total range of parameters. Due to the non-contact nature of electromagnetic
actuators, control forces can be applied from a distance and hence stability can be maintained.
Theoretical solutions are investigated for the feasibility of the control process through simulations.
The results are verified experimentally. The productivity is improved as the cutting depth is
For active chatter controls, additional damping has to be provided to the system. In the active
control nomenclature, this is called electronic damping. An electromagnetic actuator (EA) is used
for applying control forces. Figure 7.1 shows a general case for the flexible shaft processed by a
118 | P a g e
lathe machine. The EA is installed on the tool post of the lathe machine along with the cutting tool.
An eddy current probe which acts as the sensor is also applied near the actuator for a collocated
L
ɸ1 ɸ2 ɸ3 ɸ4 ɸ5
L1 L2 L3 L4 L5
Cutting tool
Gap sensor
Electromagnetic Z-axis
actuator
Motor
Lathe dog
Chuck
Tool Post
X-axis
Figure 7.1 Active chatter control architecture for flexible stepped shaft
7.2.2. Dynamics of the electromagnetic actuator
In the case of an EA, control forces are related to the actuator current as follows (Vashisht et
al., 2019d).
(i i0 )2 (i i0 )2
FEA k 2
(7.1)
( x) ( x)
2
where represents the gap between the shaft and EA in the equilibrium position, x is the distance
between the core of the EA and flexible shaft. The instantaneous current is represented as i and the
1
k 0 n 2 A (7.2)
4
119 | P a g e
In the above relation µ0 is the permeability of the air, n is the number of winding turns and A is
the area of the shoe on which winding is attached. Table 7.1 shows the properties of the EA.
Electric resistance R and inductance L can be measured for the circuit. The voltage applied v
di
v R iL (7.3)
dt
The control voltage generated from the computer can be stepped up using the voltage amplifier
so that high voltage and hence the high current can be applied for generation of control forces.
The experimental setup contains the following hardware components (Figure 7.2). The detail
The force generated by the electromagnetic actuator depends on the applied current as well as
gap between the shaft and actuator. To approximate this nonlinear phenomenon (i.e. the relation
between applied current and the generated force), artificial neural networks (ANN) are used. The
input layer of the ANN contains 2 neurons, the output layer contains one neuron and the hidden
layer contains 20 neurons. After training, the ANN is used as a Simulink block.
120 | P a g e
Inverse model
of Eddy
current
sensor
x (in m)
Sensor Computer
Data
voltage
acquition card
/Real time
amplifier engine
LPV controller
x (in m) x (in m)
v(in volts) V(in volts) x (in m) X(in volt)
Longitudinal
Neural AC voltage Electromagne Eddy current length
Flexible rotor Pass counter
Netwok amplifier tic actuator sensor measurement
sensor
- F(in N) FAMB (in N)
Figure 7.2 Schematic of the overall control system for turning operation
7.3. Dynamic modelling for turning of the stepped shaft
Let Fx , Fy and Fz are forces acting in x, y and z axes, respectively on the cutting tool,
displacements caused by these forces are given as follows (in Laplace domain).
121 | P a g e
where xx s is the transfer function of the structural system from input Fx s to output x(s). It is
m
ai si
expressed as xx s , where m are the number of modes considered, ai is
i 1 s 2ii s i2
2
the mode shape at the actuator location, si is the mode shape at the sensor location, i and i are
damping ratio and natural frequency for the ith mode considered. Similarly, the transfer function in
other directions can be formulated. Forces 𝐹𝑥 , 𝐹𝑦 𝑎𝑛𝑑 𝐹𝑧 also act on the workpiece. If the
workpiece is flexible, the corresponding deflections will be experienced by the workpiece. In this
work, it is assumed that the cutting tool has enough rigidity along x-axis and y-axis (see Figure
7.1). The movement of the cutting tool is possible only along the z-axis. However, the shaft is
considered flexible with the shaft movement along x-axis and y-axis due to limited rigidity. The
finite element method is used to determine dynamic equations of motion of the flexible shaft as
below.
where 𝐟(t) is the applied nodal force and z (t) is the nodal displacement vector. M, C and K are
global mass, damping and stiffness matrix of the flexible shaft respectively. Using the modal
analysis, natural frequencies and modes of the flexible shaft can be determined. Initial dimensions
122 | P a g e
The desired diametral dimensions after machining are [30 25 20 15 15] mm. Figure 7.3 shows
the first mode shape of the flexible stepped shaft. As the tool moves from one cutting location to
the other, natural frequencies of the flexible shaft change due to subsequent material removal. At
the same time, the whole material cannot be cut in a single pass. At each pass, some material is cut
and the systems natural frequencies change. If 0.25 mm material is cut in a single pass, the final
Figure 7.4 shows natural frequencies as a function of the tool location and number of tool
123 | P a g e
Figure 7.4 Natural frequencies of the flexible stepped shaft as a function of tool location and
pass number
7.3.2. Location dependent stiffness of the stepped shaft
From the configuration of the stepped shaft, it is obvious that the shaft stiffness is location-specific.
It is minimum near the centre. Hence the critical depth of cut is minimum near the centre as shown
in Figure 7.5. The critical depth of cut ranges from 70 to 370 microns. Hence, we cannot design a
124 | P a g e
Figure 7.5 Stiffness and critical depth of cut as a function of tool location for the stepped flexible
shaft
7.3.3. Nomenclature for the cutting system
To design a controller based on parameters of the system, dynamics of the cutting process
need to be understood. For the turning of the flexible shaft, a single point cutting tool is used. The
125 | P a g e
Thickness
ɣ F
α z z
width W α
β n
α
Direction
n
r
Of chip flow
r
x y,t
x
(a) Schematic of the cutting tool (b) Relation between local and cartesian angles
types of coordinate axis viz. global (i.e. x, y and z-axis) and local (r, n and t). The tool geometry
can be represented using certain angles of the tool. These angles include the approach angle(α),
inclination angle (β) and normal rake angle (γ). The depth of cut (W) and thickness of the chip (F)
are also shown along the local coordinate axis. The feed rate per revolution (f0) is along the
negative z-axis. The actual width or depth of cut (b) is along the x-axis. The chip flows on the
plane (r-n) in an angle (η) with the local n-axis. It is called the chip flow angle. The force acting
It is obvious that the tool moves along the z-axis at a certain angle. The chip thickness (F) and
width of cut (W) are the major parameters. Cutting forces along the local coordinate system are
K n , K r and K t are stiffness coefficients of the workpiece along the respective local coordinate
system. The relation of W and F with other parameters is given by Figure 7.6 (b).
126 | P a g e
F = f cos z(t ) cos( ) + x(t) sin( ) z(t T ) cos( ) + x(t - T ) sin( )
0
(7.7)
where z(t) and x(t) are the displacement of the tool at the instant of time (t) along z-axis and x-
axis respectively. z(t − T) and x(t − T) are displacements of the tool in one revolution before i.e.
delayed by time interval T. T is the time required to complete one revolution. The physical depth
In turning, the workpiece rotates at a certain spinning speed. Forces acting along the global
Let Kn, Kr and Kt be the stiffnesses of the workpiece along the local axes n, r, and t. Following
Fn (t)=K n F W (7.11)
Fr (t)=K r F W (7.12)
Ft (t)=K t F W (7.13)
127 | P a g e
b f cos 2 ( )+b sin ( ) cos( )x(t-T)+b cos 2 ( )z(t-T)
Fr (t)=K r 0 (7.15)
-b sin( )cos( )x(t)- b cos2 ( )z(t)
Putting the results in Equation (7.10), the global forces are given by:
Where,
After simplifying,
Fz (t ) b f g z(t) g z(t-T) g g x(t) + g x(t-T)
0 21 21 21 23 23 (7.20)
Fx (t ) b f g z(t) g z(t-T) g g x(t) + g x(t-T)
0 41 41 41 43 43 (7.21)
Fy (t ) b f g z(t) g z(t-T) g g x(t) + g x(t-T)
0 61 61 61 63 63 (7.22)
Where,
Using these forces, the dynamics of the cutting system can be formulated.
128 | P a g e
7.3.5. Open-loop stability analysis
It is obvious from Figure 7.6 that forces in x-direction and z-direction are coupled. The
workpiece material is considered as steel with Km, Kr and Kt of 544x106 N/m2, 124x106 N/m2 and
2881x106 N/m2, respectively (Eynian and Altintas, 2009). Values of tool angles α, β and γ are
taken as 300 each. Values of m, c and k for the cutting tool is taken as 50 Kg, 2000 Ns/m and 2x107
N/m respectively (Meng et al., 2017). Using Equations (7.5) and (7.20-7.22), the total dynamics
of the system can be represented in a state-space form with delay terms as follows.
where,
0 1 0 0 0 0
k b g 21 c bg
23 1 0 0 0
m m m m
A 0 0 0 1 0 0
bg 411 0 x2 bg 4312 2 x 0 0
0 0 0 0 0 1
bg 611 bg6312 0 0 y2 2 y
0 0 0 0 0 0
0
b g 21 bg 231 0
m
0 0 0 0
m
0
Ad 0 0 0 0 0 0 , B
bg 411 0 bg 4312 0 0 0 2
0
0 0 0 0 0 0
bg bg 6312 0
61 1 0 0 0 0
T
X (t ) z (t ) z (t ) qx (t ) qx (t ) q y (t ) q y (t )
T
X (t T ) z (t T ) z(t T ) qx (t T ) qx (t T ) q y (t T ) q y (t T )
129 | P a g e
where q’s are modal displacements and q ’s are modal velocities, 1 is the mode shape value at the
tool location. Matrix B is composed of mode shapes where the actuator is located. 2 is the mode
shape value at the EA location. By using Equation (7.23), the stability lobe diagrams can be
generated. Pseudo-Spectral methods are used to generate these diagrams. Figure 7.7 shows the
stability boundary from 400-2400 rpm at different tool locations. Different flexibility of the
workpiece at different locations plays a major role in terms of critical depth of cut.
130 | P a g e
7.3.6. Linear parameter varying system-based representation of the stepped shaft
From Figure 7.4 it is obvious that the natural frequency of the flexible shaft is a function of
both the tool location and number of passes of the cutting tool. Figure 7.8 shows this dependency.
Let q1 and q2 represent tool pass and tool location parameters respectively.
Figure 7.8 First natural frequency as a function of the number of cutting passes and tool
location
Dynamics of an open-loop system can be written in the following form.
or
With
131 | P a g e
0 1 0 0 0 0
k c
0 0 0 0
m m
A ( ) A (q1 , q2 ) 0 0 0 1 0 0
0 0 f1 (q1 , q2 ) 2 2 f1 (q1 , q2 ) 0 0
0 0 0 0 0 1
0 0 0 0 f1 (q1 , q2 ) 2 2 f1 (q1 , q2 )
0 0 0 0
1
0 0 0
m
B ( ) B (q1 , q2 ) 0 0 0 0
0 f 2 (q1 , q2 ) 0 f 3 (q1 , q2 )
0 0 0 0
0 0 f 2 (q1 , q2 ) 0
1 0 0 0 0 0
0 0 f 2 (q1 , q2 ) 0 0 0
C ( ) C (q1 , q2 )
0 0 0 0 f 2 (q1 , q2 ) 0
0 0 f3 (q1 , q2 ) 0 0 0
f1 (q1 , q2 ) p00 + p10q1 + p01q 2 + p20q12 + p11q1q 2 + p02q 2 2 + p30q13 + p21q12q2 + p12 q1q2 2 + p03q23
(7.25)
f 2 (q1 , q2 ) r00 + r10q1 + r01q 2 + r20q12 + r11q1q 2 + r02q 2 2 + r30q13 + r21q12q2 + r12 q1q2 2 + r03q23
(7.26)
f3 (q1 , q2 ) s00 + s10q1 + s01q 2 + s20q12 + s11q1q 2 + s02q 2 2 + s30q13 + s21q12q2 + s12 q1q2 2 + s03q23
(7.27)
132 | P a g e
When cutting forces are included, ordinary differential equations (ODE) are changed in delay
T s T 2s 2 -6Ts+12
e (7.28)
T 2s 2 +6Ts+12
Delay differential equations can be again converted into ODEs. The time-varying system can
x A( ) x B1 ( ) w + B2 ( )u
G (., ) z = C1 ( ) x + D11 ( ) D12 ( )u (7.29)
y = C x+ D w+ D u
2 21 22
box with corner i with i =1,2, …, q, and q=2m. The system matrix becomes (Gu et al., 2005):
A( ) B1 ( ) B2 ( )
G (., ) C 1 ( ) D11 ( ) D12 ( ) (7.30)
C D21 D22
2
With ranges in a matrix polytope with vertices G i . Decomposing the parameter vector
q
(t ) p1 1 .... pq q with pi 0 and pi 1 over the corners i of the parameter box,
i 1
133 | P a g e
7.4. Linear parameter varying controller design
A Linear Parameter Varying (LPV) controller is designed using the Linear Matrix Inequality
technique (Gu et al., 2005) for LPV systems represented by Equation (7.29). The linear parameter
AK ( ) BK ( ) y
K (., ) (7.32)
u = C K ( ) + DK ( ) y
AK ( ) BK ( ) q AK (i ) BK (i )
C ( ) D ( ) pi C ( ) D ( ) (7.33)
K K i 1 K i K i
A k ( ) p1A k 1 .... pq A k q
B k ( ) p1B k 1 .... pq B k q
(7.34)
Ck ( ) p1Ck 1 .... pq Ck q
Dk ( ) p1Dk 1 .... pq Dk q
This interpolation scheme yields smooth scheduling of the controller for parameter α(t).
As the natural frequency varies in a range, the Bode plot for all the systems (in an open loop)
is shown in Figure 7.9 (a). The LPV controller has a Bode plot given by part (b) of this figure. The
comparison of open and closed loop systems is represented by part (c) of this figure.
134 | P a g e
Figure 7.9 Frequency domain presentation of the perturbed system with LPV controller
To calculate the time domain response, simulations are conducted in the SIMULINK/
MATLAB environment. In the case study, the open-loop critical depth of cut (for tool location 0.2
m from right) is 72 microns. To enhance the cutting rate, it is decided to increase the depth of cut
to 250 microns. For this LPV system, an H∞ LPV controller is synthesised and applied. The feed
rate is settled as 100 microns/revolution. The spinning speed of the shaft is fixed as 600 rpm. This
corresponds to a time delay of 0.1 seconds. During the first tool cuts the workpiece in dimensions
of [35 30 25 20 15] mm, the response shows the control forces required to overcome the chatter as
shown in Figures 7.10 (a) and (b). The comparison is done with the velocity feedback controller.
It is obvious that the H∞ controller has better noise tolerance. However, if this controller (H∞
controller) is applied during the10th tool pass when dimensions of the shaft are [30 25 20 15 15]
mm, the closed-loop stability of the system is lost (Figures 7.10(c) and (d)). The amplitude of
135 | P a g e
vibration grows as well as the amplitude of control forces. However, if the proposed H∞ LPV
controller is applied, the closed-loop stability is preserved (Figures 7.10 (e) and (f)).
Figure 7.10 Time-domain performance of the perturbed system (the 10th pass of the tool) with
the nominal H-inf and LPV controller
In this work, the EA is proposed for the generation of control forces. Eddy current probe is
used for measuring displacements of the workpiece from the equilibrium position. Section 7.2 is
devoted to actuator dynamics. With parameters shown in Table 7.1, the dynamics of the actuator
is developed. An AC voltage amplifier is used for increasing the amplitude of control voltage
generated by the controller (computer). Figure 7.11 shows the amplitude of control voltages for
the ACC implementation. A feed rate of 100 microns/revolution is maintained during this
experiment. Afterwards, the feed rate is then enhanced to 400 microns/revolution and the
experiment is repeated. Due to the high feed rate, the steady-state deflection of the workpiece is
increased. Figure 7.12 shows the response of the system along the x-axis and the applied control
voltage along the x-axis. To suppress the initial transient vibrations, high control forces are
required which can be seen in the high control voltage applied to the actuator.
136 | P a g e
Figure 7.11 Performance of the perturbed system with a gain scheduled/LPV controller with
sensor and actuator in the loop
Figure 7.12 Performance of the perturbed system with a gain-scheduled/LPV controller with
sensor and actuator in the loop (feedrate 400 microns/second)
It is observed that to maintain stability and avoid the chatter, control forces need to be applied
along x-direction only. However, transient vibrations along the y-axis can be quenched quickly by
137 | P a g e
the application of control forces along the y-direction. The stiffness of the cutting tool is taken
sufficiently high to avoid any control force requirement for quenching vibrations along the z-
direction.
Following conclusions are drawn from the work presented in this chapter
Due to the limited flexibility of long shafts, the chatter cannot be prevented even if the cutting
tool is rigid. On the other hand, the flexibility of the stepped shaft depends on the tool location.
With subsequent cutting passes, the material is lost and natural frequencies are subjected to
change. Hence the controller designed for one particular tool location and cutting pass losses
An LPV controller designed can stabilize the cutting system for the total range. The system
stability is ensured for all cutting passes. Much higher cutting rate (250 microns/r) is obtained
with the proposed method compared to open loop conditions (72 microns/r), enhancing the
overall productivity.
138 | P a g e
Chapter 8
8. Active chatter control of boring operations
8.1. Overview
Boring is a machining process to make a precise hole in a workpiece. Figure 8.1 shows two
types of boring operations. Part (a) shows a case of the stationary bar (SB) boring operation. In this
case, the workpiece is light and is revolving. Part (b) shows the rotating bar (RB) boring case for
the heavy workpiece. There is a tremendous difference in dynamic modelling of these two
processes.
for chatter-free machining. Stability lobe diagrams can truly represent this limit. However, due to
the presence of manufacturing inaccuracies like ovality and eccentricity, these limits are
considerably changed. Hence, considering these in modelling the cutting process, stability analysis
139 | P a g e
and controller design is a novel idea, hence implemented in this work. Dynamic modelling of the
boring process is presented in detail using a 3-DOF model. Dynamics of such systems is
represented using delay differential equations with time-periodic coefficients. The system stability
is enhanced with active chatter control (ACC) techniques. The closed-loop system considering
fractional-order PDλ in the loop is a nonlinear time-periodic delay differential equation system.
Any systematic controller synthesis process for such systems is rarely available in the literature. It
spectral and response optimization techniques is highly efficient in terms of the requirement of the
low amplitude of the peak force and simplicity of implementation. Transient vibrations can also be
quenched in a limited period of time by using this controller. Different control techniques available
in the literature (H∞ Loop shaping and PD control) are tested and compared to the proposed
controller for enhancing the material removal rate and surface finish of the workpiece. By using
the ACC, the chatter can be efficiently reduced and the material removal rate can be increased. The
critical depth of cut is increased from 0.2 mm (open loop) to 0.6 mm (closed-loop) with a limited
It is also observed in this work that the open loop stable depth of cut for the RB boring process
is much more compared to an SB boring process. It is further observed that the RB boring process
is not a suitable candidate for ACC because much higher harmonic control forces are required for
The transfer function from the input (cutting point) to output (cutting point) locations is given
by (Figure 8.2):
140 | P a g e
y( s) x( s ) (1 ) 2
(8.1)
Fy ( s) Fx ( s) s 2 2 s 2
where 1 is the mode shape at the cutting point, is the resonance frequency and is the damping
ratio of the boring bar. The transfer function from the input (electromagnetic actuator) to output
ysensor ( s) xsensor ( s) ( 3 ) 2
2 (8.2)
Fyactuator ( s) Fxactuator ( s) s 2 s 2
The transfer function from the input (electromagnetic actuator) to output (cutting point) is given
by:
y(s) x( s ) 1 3
2 (8.3)
Fyactuator ( s) Fxactuator ( s) s 2 s 2
Forces on the boring bar are entering via cutting point as well as the electromagnetic actuator.
1 1 1 3
y( s) s 2 s 2
2 2
s 2 s Fy ( s)
2
y (8.4)
sensor ( s) 3 1 3 3 Fyactuator ( s)
s 2 2 s 2 s 2 2 s 2
11 1 3
x( s ) s 2 s 2
2 2
s 2 s Fx ( s)
2
x (8.5)
sensor ( s) 31 3 3 Fxactuator ( s)
s 2 2 s 2 s 2 2 s 2
141 | P a g e
This model can be used to simulate the dynamics of boring operations. It actually forms a
multi-input and multi-output system, where cutting forces act as disturbing forces and forces
applied at actuators are controlling forces. Cutting forces to be used in Equations (8.1-8.5) are
given by equations (7.20-7.22) in Chapter 7. By considering the cutting forces, system dynamics
β n
Z-axis
Y-axis
X-axis r x
y,t
Thickness
F z
width W α
F(t-T) Chip
α
Direction
n f 0 dy
f0 Tool
r
Of chip flow
x F(t)
(c) Relation between local and
(d) Regenerative chatter
cartesian angles
Work Electromagnetic
piece actuator
Cutting
tool
µ
Boring bar
142 | P a g e
8.2.2. Consideration of Ovality
It is assumed that there is ovality in the inside diameter of the workpiece as shown in Figure
8.3 (a). The maximum radius of the inside portion is ax and minimum radius is by. The equation of
x2 y 2
1 (8.6)
ax2 by2
where x and y are coordinates of a point moving on the ellipse. At an angle χt, the ratio of different
coordinates is as follows.
y
tan t (8.7)
x
Solving the above equations for x and y and substituting in R i.e. instantaneous radius.
R x 2 y 2 , we get
If R max is the maximum radius, the time varying depth of cut is given as:
where the varying depth or width of cut b(t) keeps on changing as the workpiece revolves i.e. the
width of cut changes periodically. The term periodically means that the width of cut b(t) at a
particular time is same as its value at T time ago, where T is the time period to complete one
revolution.
143 | P a g e
Outside
diameter
Inside
diameter
diameter Rmin
(t)/2
Max. depth
of cut Min. depth
aax of cut (b)
Eccentricity ()
(b) Eccentricity in inside diameter
Another most common problem during metal cutting is eccentricity. The workpiece may not
be mounted in the lathe chuck properly, causing overall eccentricity. It is assumed that there is
eccentricity at the inside diameter of the workpiece as shown in Figure 8.3 (b). The minimum
radius of the inside portion is aax and maximum radius is bby. The equation of the oval system is:
144 | P a g e
x2 y2
1 (8.10)
aax bb
2 2
y
y
tan t (8.11)
x
Solving the above equations for x and y and substituting in R i.e. instantaneous radius.
R x 2 y 2 , we get
R (8.12)
aax tan 2 t / 2 bby
2 2
If Rmin is the minimum radius of the cavity, the time-varying depth of cut is given as:
Putting aax Rmin b and bby Rmin b 2 , where b is the minimum width of cut, R min is
the inside radius of the cavity and ∆ is the eccentricity in the horizontal direction as shown in Figure
b t Rmin (8.14)
Rmin b tan 2 t / 2 + Rmin b 2
2 2
One thing worth consideration at this point is that with time-varying width of cut b(t), forces in
Equations (7.20-7.22) are periodically time-varying. It is observed that this factor changes the
145 | P a g e
ordinary delay differential equations (with constant coefficients) into time-periodic delay
In the case of CNC machining centres, the workpiece remains stationary and the tool
rotates. It is the case of RB boring operations. In terms of a global coordinate axis, cutting forces
Workpiece
Tool insert
XX-axis
Boring bar
YY-axis
146 | P a g e
f 0 cos t g 41 sin t g 61 + sin t g 63 cos t g 43 x(t )
Fxx b cos t g 43 sin t g 63 x(t T ) sin t g 61 cos t g 41 (t ) (8.17)
cos t g 41 sin t g 61 (t T )
By considering Equations (8.17 and 8.18), it is obvious that the resulting system contains time
periodic delay terms as well as forced terms. 𝜒 represents the angular frequency of rotation i.e.
1
2 where T is the time delay, or the time required to complete one revolution is in
T
seconds.
A boring bar of length 0.6 m is considered. The diameter of the bar is 70 mm. The tool
insert is attached at the free end. The density and Young’s modulus of elasticity are taken as 7800
Kg/m3 and 210x109 N/m2. The first four natural frequencies are calculated as 141.15 Hz, 884.5 Hz,
2477.3 and 4858 Hz. The damping coefficient for each mode is 0.0318. For simplicity, only one
mode is considered for analysis. There are four DOF per element i.e. two at each node. First is the
linear displacement and the second is the slope. After solving the eigenvalues, for the first mode,
the mass normalised mode shape coefficient ψ1 for the linear displacement is 0.4712 Kg-1/2. The
slope mode shape coefficient ψ2 is 1.08 Kg-1/2. The control force is applied near the fixed end at a
distance 60 mm from the fixed end. The linear displacement mode shape ψ3 is 7.9048 x10-3 Kg-1/2.
For properties of the workpiece material, it is considered as steel with Km, Kr and Kt as
1544x106 N/m2, 124x106 N/m2 and 2881x106 N/m2 respectively. These values are taken from
147 | P a g e
reference (Eynian et al., 2009) based on practical results. The values of tool angles α, β and γ are
taken as 300 each. Due to force acting along the z-direction, the slope of the tool inserts (i.e. tool
bit) changes dynamically. Apart from that, there is a dynamic movement along x-axis and y-axis.
By considering the first mode only, equations of the motion are as follows.
(t ) 2 (t ) 2 (t ) 2 Fz (t )
2
where θ(t), x(t) and y(t) represent dynamic slope changes of the cutting tool tip, its displacements
in the x-axis and y-axis are in the global coordinate system. In these equations, parameter µ
represents the distance of the cutting edge from the central axis of the boring bar (Figure 8.2 e).
For stability calculations, equations of the system can be represented in the first-order form as
follows.
where
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
A 2 (8.23)
- - g 21b 2 - g 23b 22 -2
2 2
0 0 0
- g 41b 12 - 2 - g 43b 12 0 0 -2 0
- g 61b 1 - g 63b 12 - 2 -2
2
0 0
T
with X (t) = (t ) x(t ) y (t ) (t ) x(t ) y (t )
148 | P a g e
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
Ad (8.24)
g 21b 2 g 23b 22
2 2
0 0 0 0
g 41b 12 g 43b 12 0 0 0 0
g 61b 1 g 63b 12 0
2
0 0 0
T
with X (t - T) = (t T ) x (t T ) y (t T ) (t T ) x (t T ) y(t T )
and
0
0
0
B (8.25)
0
1 3
0
where
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
A(t ) 2 (8.27)
- - g 21b(t ) 2 - g 23b(t ) 22 -2
2 2
0 0 0
- g 41b(t ) 12 - 2 - g 43b(t ) 12 0 0 -2 0
- g 61b(t ) 1 - g 63b(t ) 12 - 2 -2
2
0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
Ad (t ) (8.28)
g 21b(t ) 2 g 23b(t ) 22
2 2
0 0 0 0
g 41b(t ) 12 g 43b(t ) 12 0 0 0 0
g 61b(t ) 1 g 63b(t ) 12 0
2
0 0 0
149 | P a g e
8.5. Active chatter control of stationary bar boring operation
The virtual stiffness and damping of the boring bar can be enhanced using active control
techniques. It is a form of the closed-loop system. An electromagnetic actuator can be applied near
the fixed end. Figure 8.5 shows the physical schematic of the system.
Electromagnetic actuator
Boring bar
rollers
Lathe bed
Electromagnetic
Boring bar AC amplifier
actuator
150 | P a g e
8.5.1. Experimental setup for active chatter control
Components of the experimental setup can be found in Chapters 5, 6 and 7. The physical
The transfer function of the boring bar system in the finite element method is verified
experimentally. For estimating the transfer function at the cutting location, a piezoelectric impact
hammer with the force sensor is used. From input/output data, the frequency response function of
the system is calculated using the MATLAB command ‘tfestimate’. However, to estimate the
transfer function at the actuator location, an excitation signal is applied at the electromagnetic
actuator location. Eddy current probes are used to measure the deflection of the boring bar. From
the frequency response functions, the natural frequency, damping ratio and mode shape are
calculated. Table 8.2 shows the comparison of theoretical and experimental values.
151 | P a g e
Table 8.2 Comparison of theoretical and experimental results
2 (Mode shape at cutting location (slope)) 1.08 Kg-1/2 1.20 Kg-1/2 10.0
3 (Mode shape at actuator location) 8 x10-3 Kg-1/2 9.3 x10-3 Kg-1/2 13.9
For designing the loop shaping design procedure (LSDP) controller, original delay differential
equations are converted into ordinary differential equations by using the 5th order Pade’s
approximation. The order of the approximate system is 11. Figure 8.6 (a) shows the comparison of
Figure 8.6 Design parameters and frequency response of the loop shaping design procedure
controller
152 | P a g e
The LSDP controller is designed using the method in reference (Gu et al., 2015). The
comparison of the open and closed-loop systems is shown in Figure 8.6 (b). The plot of the
By considering the ovality in the workpiece, delay differential equations are converted into
time-periodic delay differential equations i.e. delay differential equations with periodically varying
coefficients. Pade’s approximation of these types of systems is not possible and hence the robust
control synthesis. Systematic controller design procedures for such systems are scant in literature.
Methods for the stability analysis are used to design the controller. Proportional derivative (PD)
controllers can be successfully designed base on this technique. The system of time periodic delay
The integral controller is not useful for the active chatter reduction due to the presence of
steady deflection of the tool resulting actuator saturation. The PD controller is quite suitable. By
choosing u(t) = −K P x(t) − K D ẋ (t), the above equation is modified with the following matrix.
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
A(t ) 2
- - g 21b(t ) 2 - g 23b(t ) 22 -2
2 2
0 0 0
- g 41b(t ) 12 - 2 - g 43b(t ) 12 ( K P ) 1 3 0 0 -2 ( K D ) 1 3 0
- g 61b(t ) 1 - g 63b(t ) 12 - 2 -2
2
0 0
(8.30)
Parameters of the controller can be chosen based on the stability of the overall closed-loop
system. In the section of results, it is described that the derivative controller (i.e. velocity feedback)
153 | P a g e
is more suitable in terms of lower peak force requirements than a proportional controller (i.e.
Fractional order proportional derivative (FOPD) controller can be realised by using the
proportional and fractional-order derivative of output error e(t). In active chatter control
applications, the output error is given as -x(t). The control force applied to the boring bar can be
represented as follows.
where K p and K d are the proportional and derivative gains, respectively. Dλ denotes the
t
1
e(t ) w j e t jh
h
( )
D λ e(t)=lim h0
(8.32)
h j 1
with
1 ( )
w (0 ) 1 and w (j ) 1
j
w j 1 (8.33)
The value of λ ranges generally from 0 to 2. In this work, 0 < λ < 1 is considered. If the value
of λ is taken as 1, it will represent a conventional PD controller. For designing the FOPD controller
(rough estimation), the following algorithm (Figure 8.8) is applied based on the pseudo-spectral
Values of K p for the FOPD controller remain the same as for ordinary PD controllers.
However, K d for FOPD (with λ=0.5) is given as √K d for ordinary PD controllers. K d obtained
using the above method is approximate. It acts as a starting point for determining actual values by
response optimization techniques. Figure 8.7 shows the Simulink model of the system.
154 | P a g e
S1 b(t)* f0 *C1
F1
Time (t) S2 b(t)* f0 *C2
S3 b(t)* f0 *C3 F2
F1 (S1) [bg21{(t T) (t)}bg23{x(t T) x(t)}]
F2 (S2) [bg41{(t T) (t)}bg43{x(t T) x(t)}] F3
F3 (S3) [bg61{(t T) (t)}bg63{x(t T) x(t)}]
x
Output to MATLAB
workspace
delay
+
(t T ) (t )
-
Flexible Boring x(t T ) x(t ) Input from MATLAB
bar + workspace
delay
-
-
y Kp Output to MATLAB
workspace
uv +
Input from MATLAB Kd Fx
workspace λ
S1 b(t ) * f 0 * C1
S2 b(t ) * f 0 * C2 (8.34)
S3 b(t ) * f 0 * C3
and
C1 cos X 1
2
C2 cos X 2
2
(8.35)
C3 cos X 3
2
Values of constants X1, X2 and X3 are dependent on material parameters (Eynian et al., 2009).
A flowchart of designing the FOPD controller (exact parameters) is shown in Figure 8.8. The
155 | P a g e
Figure 8.8 Flowchart for fractional order PDλ control system synthesis
156 | P a g e
A pattern search optimization technique is applied for optimization. The Simulink file shown
in Figure 8.7 is executed along with constraints of the closed loop stability and the root mean
As per the flow chart, parameters Kp, Kd and λ are optimised to find such parameters with the
minimum value of amplitude of control forces along with preserving the closed loop stability.
Parameter λ plays an important role. Figure 8.9 shows the response of the system corresponding to
λ =0.2. It is obvious that much lower values result in control signal fluctuations in region II as
shown in Figure 8.9 (d). There is fluctuation in control forces although the transient vibrations are
Figure 8.9 Closed loop response of the boring process with fractional order PD controller
(λ=0.2) (simulations)
157 | P a g e
8.5.4. Simulation and experimental results
a). Open loop stability analysis
Figure 8.10 shows the stability lobe diagram for the boring operation in open loop conditions.
The semi-discretization method (Insperger et al., 2004) and PS method are compared. The first
column shows results by the PS method. The second column shows results by the semi-
discretization method. It is obvious that PS methods give more accurate results at lower spinning
rates. Figure 8.11 shows the stability lobe diagram for the open loop system with and without
ovality. The figure shows the comparison of the minimum critical width or depth of cut as a
function of spinning speed. The spinning speed from 400 Hz – 1200 Hz is considered. This is the
practical limit for most boring operations. In all these figures, the area below the curve is the stable
zone (chatter free) and area above the curve is unstable region (with chatter) (Breda et al., 2012,
Figure 8.10 Comparison of the semi-discretization approach and Pseudo spectral approach
158 | P a g e
Figure 8.11 Open-loop Stability lobe diagram for boring without and with ovality
In case of the workpiece without ovality, the critical depth of cut is 0.28 mm. To analyse the
effect of ovality, a particular case is used with the cavity radius 100 mm (major radius) and 99.9
mm (minor radius). The minimum depth of cut is settled as 0.2 mm. With these parameters, the
depth of cut varies from 0.2 mm to 0.3 mm periodically. Figures 8.11 (b) and (d) show the stability
lobe diagram for boring operations with ovality. From the figure, it is obvious that as the spinning
speed increases there are bigger regions of stability. If there is an eccentricity of 0.05 mm, the
actual width or depth of cut varies from 0.2 mm to 0.3 mm periodically. The stability lob diagram
is the same as shown in Figure 8.11 (with very minute differences, hence not shown).
For the chatter stability with active damping, control forces are applied along the x-axis only.
This ensures the overall stability of a closed-loop system. There is no need to apply forces in z-
axis and y-axis. The feedback rate of 20 microns/revolution with the depth of cut 0.6 mm is applied.
159 | P a g e
The closed-loop system with a stabilizing LSDP controller produces a control force of 120 Newton
(when the tool engages with workpiece). If higher feed rates are applied, the amplitude of control
forces increases to higher values. Experiments are also conducted with this controller and the
The control system should be applied from the start so that the amplitude of control forces
does not go to very high values. Without a Band – Pass filter, the LSDP controller has integral
action. To remove the steady-state deflection of the cutting tool, the control action is applied in a
single direction. This will cause the actuator saturation. This difficulty is removed by using a Band
With a stabilizing PD controller, the amplitude of control force decreases to 85 Newton. The
PD controller increases the virtual damping of the system. The LSDP controller increases both the
dynamic stiffness and virtual damping of the boring bar. There is a requirement of the high control
Figure 8.12 Closed loop response the boring process with eccentricity using PD controller
(experimental)
160 | P a g e
In simulations, it is assumed that there is no ovality and eccentricity in the boring operation.
However, it is difficult to avoid these faults in real life. The presence of eccentricity makes the
response of the system in x-direction and y-direction sinusoidal even if the transients have died.
The magnitude of eccentricity is the same as used to construct the stability lobe diagrams. Figure
8.12 shows experimentally measured responses of the boring bar in various directions.
Figure 8.13 shows experimental results for the FOPD controller with 20 microns/revolution
feed rate. The response of vibrations in various directions is not sinusoidal after transients are over.
This is due to the reason that eccentricity and ovality were completely eliminated in this
experimental set. The steady-state deflection approaches a constant value here. The applied force
reaches to a virtual zero value (due to the presence of a bandpass filter in the circuit) after dying
out of the transients. The sensor/actuator results are converted into microns/Newtons respectively
for comparison with simulation results. From the simulated and experimental results, it is obvious
that the FOPD controller requires much less amplitude of the control force (35 N). Also, transients
are quenched out much quickly with this controller compared to PD and LSDP controllers. For
Figure 8.13 Performance of Fractional order PDλ controller at 20 microns/rev feed rate
(experimental)
161 | P a g e
8.6. Active chatter control for rotating bar boring operation
8.6.1. Controller design
Similarly, the dynamics of the revolving bar boring operation (in the presence of active
With matrix in Equation (8.27) converted into the following matrix (using PD control).
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
A(t )
- - g 21b 2 22
2
- g 23b 22 0 -2 0 0
-( g 41 cos( t ) g 61 sin( t ))b 12 - 2 - ( g 43 cos( t ) g 63 sin( t ))b 12 ( K P ) 1 3 0 0 -2 ( K D ) 1 3 0
-( g 61 cos( t ) g 41 sin( t ))b 1 -( g 63 cos( t ) g 43 sin( t ))b 12 - 2 -2
2
0 0
(8.37)
In a similar way, PDλ controller can be designed for this system as per section 8.5.3.
For RB boring operations, the critical depth of cut increases to 1.4 mm (at 600 rpm) under
open-loop conditions (Figure 8.14). It is obvious that there is 418% increase (at 600 rpm) in the
critical depth of cut. Hence, it is safe to conclude that, stability margins for RB boring operations
are much higher than SB boring operations. The simplest method to increase the stable depth of
162 | P a g e
Figure 8.14 Open-loop Stability lobe diagram for boring with revolving bar
b). Performance with active control
The spinning speed of the rotor is taken as 600 rpm and width of the cut as 1.4 mm. There are
steady-state vibrations in both horizontal and vertical directions (Figure 8.15). However, the
system is stable. Afterwards, the depth of cut is enhanced to 1.6 mm (i.e. in the chatter zone with
14% increase in depth of cut). The feed rate of 5 microns/revolution is fixed. The controller
designed in Section 8.6.1 is applied. A control force with an amplitude of 290 Newton is applied
by using this stabilizing controller (Figure 8.15, Part (c)). Even for such a low value of feed rate,
much higher control forces are required to prevent chatter. Part (d) of Figure 8.15 shows the
163 | P a g e
Figure 8.15 Closed loop response of the boring operation (with revolving bar) using PD control
Figure 8.16 Closed loop response of the boring operation (with revolving bar) using PDλ
controller
164 | P a g e
Afterwards, fractional-order PDλ controller is applied (Figure 8.16). There is no change in the
magnitude of control forces required for the stabilization purpose. Also, after the transients are
over (nearly after 1.5 seconds), the harmonic component of force (nearly 60 N) remains active.
However, this force can be reduced by switching the PD or PDλ controller gains to lower
values (Figure 8.17) (i.e. after 2s). The feed rate is slowly increased from 5 microns/r to 20
microns/r (from 4s-7s) (Figure 8.17). With this, the steady-state vibration amplitude keeps on
increasing along with the applied control forces. A harmonic force of 100 N is applied to maintain
stability and avoid chatter. Obviously, the ACC for RB boring operations above the critical depth
is not a good candidate for the active damping treatment. For detail the reference (Vashisht et al.,
Figure 8.17 Closed-loop response of the boring operation using switching control and time-
varying feedback (with revolving bar)
165 | P a g e
8.7. Concluding remarks
PS techniques are more efficient in terms of the stability analysis compared to semi-
discretization techniques.
PD controllers are efficient in terms of low peak force requirements. Their efficiency is better
than the LSDP type of robust controllers. However, FOPD controllers are most efficient in
terms of peak force requirements. The transient response with this controller is superior to
other controllers. Hence, small size actuators are required to implement active chatter control
The presence of ovality and eccentricity converts the boring operation dynamics from ordinary
successfully applied for the stability analysis and approximate FOPD controller design for
such systems. Using response optimization techniques, the exact parameters of this controller
can be obtained.
The open-loop critical depth of cut for the revolving bar boring operation is much higher than
the stationary bar boring operation. The easiest way to increase the critical depth of cut and
hence the material removal rate is to make the workpiece stationary and revolve the boring
bar.
However, to further increase the depth of cut, high amplitude sinusoidal forces are required.
ACC is quite suitable for SB boring operations but not for RB boring operations. Even by
166 | P a g e
Chapter 9
9. Chatter detection in milling operations using Neural
Networks
9.1. Overview
In this chapter, three chatter indicators for milling operation are provided based on Hilbert
Huang Transformations, short-time Fourier transform and locus of the milling cutter at sampling
frequency synchronous to per revolution frequency. Convolutional neural network (CNN) system
based on machine/deep learning is proposed for the chatter detection based on these indicators. It
is observed that even in the chatter-free zones, the amplitude of vibrations of the cutting tool varies
a lot and depends on the spinning speed of milling tools. For a better surface finish of the
workpiece, this amplitude should be kept at the minimum level. Theoretically, during chatter, due
to the regenerative effect, the amplitude of vibrations increases infinitely. However, practically,
due to the disengagement of the cutting tool and workpiece, the amplitude of vibrations remains
finite. Certain modifications are presented in the dynamic equations of motion of the milling
operation, which can be used to find the amplitude of vibrations during chatter conditions. The
simulated data generated in this way is used to train a CNN. The trained CNN is then used to detect
chatter in an online manner. The results show the maximum 99.8% accuracy even in the presence
The dynamic model of the milling process can be represented by the following equation
167 | P a g e
x(t ) 2 0 x(t ) 2 0 x(t ) 1 hxx (t ) hxy (t ) a0 x(t ) x(t )
w
y(t ) 0
2 y(t ) 0 2 y(t ) m hyx (t ) hyy (t ) y(t ) y(t )
(9.1)
tool in the feed direction. Similarly, ys represents corresponding quantities in the direction
perpendicular to the feed direction. is the damping ratio and represents the natural frequency
of the milling cutter (tool). w is the axial width of cut and m is the modal mass. a0 is the feed rate
(in meters/tooth pass) of the cutting tool. x(t-τ) and y(t-τ) are displacements of the cutting tool in
60
(9.2)
N
where N is the number of teeth in the cutter and is the spinning speed of the cutter in revolutions
per minute. The other quantities are given below (Insperger et al. 2004).
hxx (t ) g[ j (t )]sin j (t ) K t cos j (t ) K n sin j (t )
N
j 1
hxy (t ) g[ j (t )]cos j (t ) K t cos j (t ) K n sin j (t )
N
j 1
(9.3)
hyx (t ) g[ j (t )]sin j (t ) K t sin j (t ) K n cos j (t )
N
j 1
hyy (t ) g[ j (t )]cos j (t ) K t sin j (t ) K n cos j (t )
N
j 1
With
2 2
j (t ) t j (9.4)
60 N
1 if st j (t ) ex
g[ j (t )] (9.5)
0 otherwise
168 | P a g e
where st and ex are start and exit immersion angles respectively. For up milling (present case)
Workpiece
Cutter
L
o y
Direction of
D
feed
a
cx kx
Cutter
ky w
cy Workpiece
Figure 9.2 Geometric parameters of the cutting surface when the tool leaves the workpiece
169 | P a g e
When simulations are carried out using Equation 9.1, the tool displacement from the
equilibrium position along the feed direction as well as perpendicular to the feed direction can be
calculated. Due to the presence of delay in these equations, the simulation system becomes
unstable for certain combinations of axial width of cut and spinning speed of the milling tool.
Theoretically, the amplitude of vibration becomes infinite under these unstable conditions.
However, under practical conditions, the amplitude of vibration remains finite. This happens
because the tool leaves the workpiece during the certain portion of the cutting period and the
amplitude of applied forces remains finite. This is called a flyover effect. By writing Equation
where Fx (t) and Fy (t) are the cutting forces in the feed and perpendicular to feed direction,
respectively. Time delay terms are important and hence we are writing x T=x(t-τ) and yT=y(t-τ) in
these equations. More precise values for xT and yT should be considered to model the flyover
effect. That means more than one proceeding milling tool revolution must be considered in the
and
170 | P a g e
Following relations are used to calculate cutting forces.
with
x* a0 xT x(t ) (9.12)
and
y* yT y(t ) (9.13)
Figure 9.2 shows the graphical representation of selecting a delay term in the feed direction.
In a similar way, the delay terms in the perpendicular to feed direction can be calculated. Choosing
the cutting forces in this way, simulation results can effectively take into account the flyover effect.
With this arrangement, through simulations, we can effectively predict the behaviour of the system
during chatter conditions. Jemielniak et al. (1989) suggested that this effect must be taken into
consideration for effectively designing any chatter detection strategy. The milling tool behaviour
can be better understood using these sets of equations. The pattern of cutting forces predicted by
To validate the simulation results, the experiment is conducted. Results are also compared
171 | P a g e
9.3.1. Simulation model
A carbide end mill with a diameter 19.05 mm (3/4 inch) with a single flute is used (the second
a
a=4.515 mm is fixed so that the radial immersion ratio of =0.237 is used. The up-milling is
D
conducted for aluminium (i.e. 7075-T6). The cutting coefficients in the tangential and normal
direction are calculated by using Kistler rigid dynamometer. Estimated values of these coefficients
are Kn=2.1x108 N/m2 and Kt=5.6x108 N/m2. These values are similar to those in references (Mann
2003, Insperger et al. 2003). Modal parameters of the cutting tool (i.e. milling cutter) are in good
agreement with these references. These values are intentionally chosen to verify the respective
results with the previously available results. Impact testing using an impact hammer is used to
experimentally estimate the modal parameters of the milling tool. The first modal frequency is
estimated to be 146 Hz with modal damping ratio 0.003. The estimated modal mass is 2.6 Kg. The
The schematic of the experimental setup is shown in Figure 9.3. Components of the
172 | P a g e
Figure 9.3 Schematic of the milling process with corresponding sensors
9.3.3. Stability lobe diagrams
The regenerative process in metal cutting processes is the main source of instability. In
literature, single and two-degree-of-freedom models of the milling tool had been used for analysis.
In this research, a two-degree-of-freedom model is built with the flexibility of the tool in feed and
perpendicular to the feed direction. Equation (9.1) can be written in the first-order form in a state-
173 | P a g e
X (t ) A(t) X (t ) Ad (t) X (t ) (9.14)
with
0 0 1 0
0 0 0 1
whxy (t )
A(t ) 2 whxx (t ) 2 0
m m
whyx (t ) whyy (t )
2 0 2
m m
0 0 0 0
0 0 0 0
whxy (t )
Ad (t)= whxx (t ) 0 0
m m
whyx (t ) whyy (t )
0 0
m m
Such that
x(t ) x(t )
y (t ) y (t )
X (t)= and X (t - )= (9.15a, 9.15b, 9.15c, 9.15d)
x(t ) x(t )
y (t ) y (t )
The simulation model discussed in Section 9.3.1 is used for constructing stability lobe
diagrams. The spinning speed ranging from 2900 r.p.m. to 3800 r.p.m. is considered. Same
parameters in references (Mann 2003, Insperger et al. 2003) are taken so that a valid comparison
can be made between simulations, our experimental results and in these references. The semi-
discretization method discussed in reference (Insperger et al., 2004) is used to construct the
stability lobe diagram based on Equation (9.14-9.15). Two specific points at 3300 rpm and 3650
rpm. are taken in the chatter zone (dark area) to observe the behaviour during chatter conditions.
174 | P a g e
Figure 9.4 Stability lobe diagram for a specified range
9.3.4. Simulation vs experimental results
For comparing simulation results with our experimental results and those in literature (Mann
2003, Insperger et al. 2003), Figures 9.5 and 9.6 are generated.
175 | P a g e
Figure 9.5 Comparison of simulation results with experimental results at 3300 rpm
176 | P a g e
Figure 9.6 Comparison of simulation results with experimental results at 3650 rpm.
177 | P a g e
First of all, results are compared at 3300 rpm. Two different sampling frequencies are used in
the data collection. Firstly, the sampling frequency is taken as 2kHz. Parts (a), (c) and (e) of Figure
9.5 represent the signal in the feed direction. Secondly, per revolution sampling frequency is taken
(3300/60=55 Hz). Since the second tooth of the milling tool is grounded, it also corresponds to the
tooth pass frequency. Parts (b), (d) and (f) represent the signal corresponding to this frequency.
During experimentation, the feed rate is progressively increased so that the feed rate of 0.1 mm/r
is obtained in 4 seconds starting from zero. After the steady-state, the simulated amplitude of
vibration matches well with experimental results as well as results from the literature. Similar is
the case with 3650 rpm, spinning speed. Period double vibrations are obvious in Figure 9.6. With
these comparisons, it is obvious that our simulation technique is correct and can be used to design
9.4 The amplitude of tool vibrations and tool forces during chatter
It is a common myth that chatter corresponds to high amplitude vibrations only. To investigate
it thoroughly, the situation at points marked in a stability lobe diagram shown in Figure 9.7 is
considered.
Figure 9.7 Stability lobe diagram for the system under analysis
178 | P a g e
Figure 9.8 Amplitude of vibration (in x-direction i.e. feed direction) at a different spinning speed
of milling tool
Two stability lobes are considered in this figure. Dark areas represent the unstable zone.
Only points in the range 2125 -2200 rpm are stable (corresponding to 4 mm axial depth of cut).
179 | P a g e
This stability zone is termed as number 1. The other stability zone is termed as number 2. Another
stability zone is in the range of 1700-1750 rpm. At other depth of cuts, there is some variation in
spinning speed. The interest is to work in this narrow range so that that maximum axial depth of
cut can be obtained corresponding to that spinning rate. The unstable points are considered towards
the left and right of stability zone 1. Point B with 2075 rpm is under chatter conditions (Figure
9.8). The amplitude of vibrations is nearly 200 microns. Similarly, at point E with 2225 rpm. the
amplitude of vibration of the milling tool in the feed direction is near 1800 microns. The amplitude
of vibrations is minimum at point C with 2125 rpm (stable point). However, point D with 2200
rpm is also stable with an amplitude of vibration near 1600 microns. Hence, the amplitude of
vibration is not the criterion for chatter. As towards the right, the amplitude of vibration decreases.
The second column of this figure shows the response at per revolution sampling rate (also known
The displacement of the cutting tool in the feed direction is simplified as follows.
w
x(t ) 2 x(t ) 2 x(t ) hxx (t ) a0 x(t ) x(t ) (9.16)
m
x(t ) a0 x p (t ) (t ) (9.17)
where a0 is the feed per tooth pass, xp(t) is the forced periodic response of the system when no
self-excited vibrations are present. For xp(t)= xp(t+τ), (t ) is the response due to self-excited
vibrations at chatter frequencies. During chatter, this response component is as follows (Insperger
et al. 2003).
w
(t ) 2 (t ) 2 (t ) hxx (t ) (t ) (t ) (9.18)
m
180 | P a g e
Stability properties of this equation can be determined by the infinite number of characteristic
multipliers. Floquet theory for delay differential equations can be applied to determine the stability.
If e is a characteristic multiplier, there exists a solution of Equation (9.18) in the following
form.
denotes the complex conjugates. This equation is stable if all the characteristic multipliers have
the modulus less than 1. The characteristic multipliers of Equation (9.14) can be located in the
following ways.
1. In case 1, These can be located on the unit circle as complex pairs (i.e. 1, ). This is
As suggested by Insperger et al. (2003), case 2, 1, can not arise in Equation (9.18). First of all,
ln( )
for case 1, 1 . In this case, i is pure imaginary where . Since the complex
function is periodic, the logarithmic function is not unique in the complex plane. Hence, multiple
chatter frequencies exist in the solution. This shows that frequencies arising in Equation (9.18) for
2
f H n rad / s (9.20)
181 | P a g e
In the above expression, the index f H is for secondary Hopf bifurcations. For period-doubling
ln(1)
bifurcations where 1 (case 3), the characteristic exponent is given . Frequencies,
2
f PD n rad / s (9.21)
where index f PD is for period-doubling bifurcations. Based on the above discussion, Figure 9.9 is
constructed for spinning speeds of the milling cutter in the range 2200 to 2500 rpm.
Based on simulated data, using various signal processing techniques, different characteristics
are observed that can be used to indicate the presence of chatter in the machining processes. First,
182 | P a g e
9.6.1. Hilbert Huang Transformations based indicators
Using Equations 9.1-9.13, for simulation purposes, the displacement of the milling tool in the
feed direction is calculated. Using Hilbert Huang Transformations (HHT), the instantaneous
frequency of the signal as a function of time is calculated at various spinning speeds of the milling
cutter. Figure 9.10 shows that plot. Following spinning speeds are considered: 1950, 2075, 2125,
2200, 2225, 2300, 2400 and 2500 rpm. Under noise-free simulations, instantaneous frequencies as
a function of time are shown in this figure. As the spinning rate of the milling cutter changes, there
are unique instantaneous frequency patterns associated with each spinning rate. This feature can
be used as the chatter indicator. At 2125 rpm. the instantaneous frequency ranges from 120-150
Hz. It is a chatter-free case. The amplitude of vibrations is also minimum at this spinning rate.
Although 2200 rpm case is also chatter-free, the amplitude of vibrations is much higher (1350
microns compared to 175 microns) at this spinning rate (Figure 9.8). From Figure 9.10, it is
obvious that this pattern is independent of the displacement amplitude of the milling tool. The
advantage of this technique is that the instantaneous frequency range can easily be calculated.
However, there are certain limitations to this approach. The sensitivity of this approach is
highly dependent on the noise present in the data. Figure 9.11 shows the effect of noise on the
HHT at different spinning rates. Spinning rate of 2175 rpm. (chatter-free case) and 2225 rpm. (with
chatter) are considered in this figure. It is observed that as the Signal-to-Noise-Ratio (SNR)
approaches 20 dB, it becomes difficult to distinguish between the chatter-free and with chatter
case. Discrete wavelet transformations (DWT) as well as HHT (by selectively choosing the
Intrinsic Mode Functions) can be used to denoise the signal. However, limited success is achieved
183 | P a g e
Figure 9.10 Instantaneous frequencies as a function spinning speed of milling tool (axial depth
of cut is 4 mm)
184 | P a g e
Figure 9.11 Effect of noise on instantaneous frequencies
Short time Fourier Transformation (STFT) of the displacement of the milling tool is generated.
A sampling frequency of 2000 Hz is used to collect the signal as shown in Figure 9.12. Power
Spectral Densities, as well as Spectrograms, are shown in this figure. Spinning speed from 1950
to 2500 rpm is used in this analysis. In the spinning rate 1950-2075 rpm and 2225-2500 rpm chatter
conditions prevail in the system. Multiple chatter frequencies are also present in the signal. Figure
9.9 shows the chatter frequencies in the range 2200 to 2500 rpm. As discussed in the next section,
convolutional neural networks can be used to classify images. A properly trained neural network
However, the measurement noise is inevitable. In the noise presence, the difference in
spectrogram characteristics (with and without chatter) drastically reduces. Figure 9.13 represents
a particular case showing the effect of noise on the power spectral density. Spinning speeds of
2175 rpm (chatter-free case) and 2225 rpm (chatter conditions) are considered in the analysis.
185 | P a g e
Figure 9.12 STFT as a function of spinning speed of milling tool (at 4 mm axial depth of cut)
186 | P a g e
Figure 9.13 Effect of noise on power spectral density
187 | P a g e
A signal with 20 dB SNR is distinguished from the noise-free case. Figure 9.14 shows the
Spectrograms of the signal under the noisy condition and noise-free case. Obviously, it is
comparatively difficult to distinguish between chatter and chatter-free cases even with trained
convolutional neural networks. On the other hand, when the spinning rate gap between chatter and
non-chatter conditions increases, chatter becomes easy to detect even at lower SNR values.
9.6.3. Locus of milling tool (at synchronous per revolution frequency)-based indicators
Chatter indicators indicated in the above two sub-sections require only one sensor measuring
the tool displacement (in the feed direction). Due to the problem of measurement noise, their
efficiency is not up to the mark. In this section, we discuss a two-sensor approach (one along feed
direction and the other perpendicular to feed direction) collecting data at sampling frequency
synchronous to tooth pass frequency. In this work, due to the presence of only one tooth for cutting,
tooth pass frequency is equal to the milling tool spinning frequency. For example, at 2075 rpm the
sampling frequency should be integer multiple of 2075/60 Hz. The integer is chosen as 100, so the
2075
sampling frequency is x100 3458.3 Hz . The collected data in x and y-direction are used to
60
plot the locus of the milling tool. Figure 9.15 shows loci/paths at 2075 (chatter), 2125 (chatter-
188 | P a g e
Figure 9.15 Milling tool locus as a function spinning speed of milling tool (noise free cases)
At 2075 rpm the response of the system is equivalent to period two bifurcations. At 2225 rpm,
the response is equivalent to Hopf Bifurcations. During other chatter-free conditions (2125 and
2175 rpm), the response is a hollow tilted circle. Fortunately, the response pattern in the presence
of measurement noise does not change. Figure 9.16 shows the locus of the cutting tool in the
Figure 9.16 Effect of noise on locus of milling tool at different spinning speeds
189 | P a g e
In this case, neural networks can be used to classify the response of the system.
Convolutional Neural Networks as shown in Figure 9.17 are used for image classification. It
contains convolutional layer, pooling layer, fully connected layer and classification layer (Zhao et
al., 2019).
Classification
Fully connected
Based on the quality of the machine, the response of the system at higher spinning rates can be
predicted. In this work, the rpm of the milling cutter is assumed to vary between 1500-2800. The
axial depth of cut is fixed in the range 0.5-4 mm. From the stability lobe diagram (Figure 9.18),
the cutting space is divided into 12 (A to L) x 20 =240 points. All depths of cuts as well as spinning
speeds are covered in the required range. Using a computer program, the tool displacement data in
the feed as well as perpendicular to feed direction are calculated for a certain period of time. It is
observed that for the first 0.6 seconds, the initial transients are over and the steady-state response
is achieved. Hence simulation data are created for the first 2 seconds. The images of the milling
tool locus are made with a pixel density (227x227). For training of the CNN, 3 categories are
bifurcations), the second category contains type II chatter (corresponding to Hopf Bifurcations)
190 | P a g e
and the third category is chatter-free vibrations. The neural network is trained on Intel core i7-
All the 240 images are classified into these three categories. All the images are infected with
20 dB SNR noise. It is observed that after 400 iterations, there is no significant improvement in
results. It takes almost 30 minutes to complete these iterations. The configuration of the CNN in
the MATLAB toolbox for Deep Learning is used (after certain hit and trials) as follows.
191 | P a g e
An accuracy of 80% is obtained with an SNR of 20 dB after training the network. It is also
observed that it is easy to distinguish between type I chatter and chatter-free conditions (than with
type II chatter and chatter-free conditions) with STFT based images. This is due to a large number
dangerous which is difficult to detect with this method. The confusion matrix for this system
(STFT-based detection) is given as Table 9.1. Class 1 represents no chatter, class 2 represents type
In tool locus-based chatter detection, 240 images are infected with white Gaussian noise
corresponding to the 20 dB SNR. After 500 iterations, CNN is converged to an accuracy of 99.8%.
It takes an average time of 35 minutes to complete the task. All the 240 images are classified into
these three categories. The configuration of the CNN in the MATLAB toolbox for Deep Learning
192 | P a g e
Input [227x227]-64Convolution [3x3]-64Pooling[3x3]-BatchNormalization-ReLU-
128Convolution[3x3]-128Pooling[3x3] -BatchNormalization-ReLU-256Convolution[3x3]-
256Pooling[3x3] -BatchNormalization-ReLU-384Convolution[3x3]-384Pooling[3x3] -
BatchNormalization-ReLU-FullyConnected-FullyConnected-Softmax-Classification.
The biggest advantage of tool locus-based chatter detection is that it is easy to detect type II
chatter which is more dangerous. The confusion matrix for this system (tool locus-based detection)
is given as Table 9.2. Class 1 represents no chatter, class 2 represents type I chatter and class 3
In this work, CNN based algorithms are proposed for online chatter detection. A CNN is
trained using images formed by simulation based on the locus of the milling cutter. This trained
CNN is then used to detect chatter on an online basis. Various chatter indicators are introduced for
193 | P a g e
Mathematical techniques to model the flyover effect can predict the amplitude of vibrations
during chatter conditions with sufficient accuracy hence are useful to create simulated data.
The amplitude of vibrations is not always high during chatter conditions. At certain spinning
rates, the amplitude of vibrations during chatter-free conditions is several times higher than
Type II chatter is much dangerous than type I. But, fortunately, it is easy to detect type II
chatter using a tool locus method. The reverse is also true for the STFT based method.
HHT and STFT based chatter detection requires only one sensor but efficiency is prone to
measurement noise. CNN for detecting chatter is less accurate using the image based on STFT.
The second method based on tool locus at synchronous per revolution sampling frequencies
CNN has remarkable accuracy in terms of the image classification and is quite useful to detect
chatters during milling process on the real-time basis using tool locus method, however, short-
194 | P a g e
Chapter 10
10. Research conclusions, contributions and future
recommendations
10.1. Conclusions
Mathematical modelling, fault detection and active vibration control of rotor dynamic systems
with applications in machine tools have been discussed in this dissertation. First of all, the
mathematical modelling of the cracked rotor is presented. Based on simulation results, it shows
that different components of the rotor system like ball bearings transform the linear dynamics of
the rotor system to nonlinear dynamics. It is observed that the different crack detection techniques
based on the steady-state vibration response start losing their effectiveness in the presence of
nonlinearities caused by ball bearings. The transient vibration response-based techniques are better
suitable under these conditions. Switching control strategy can enhance the dynamic
characteristics related to the transverse crack. Short-time Fourier transform techniques can be
effectively used for detecting the interesting characteristics hidden in a frequency varying signal
during the run-up of the rotor. However, there are certain limitations in the proposed technique
(based on switching control). In the presence of other faults like coupling misalignment and
rotor/stator rub (in certain specific circumstances), similar results are obtained. Under these
Rotor/stator rubbing phenomena are modelled mathematically. The phenomena of the dry
whip, partial forward whirl and partial backward whirl are easily observed in case of vertical rotors
under the absence of gravity. However, for horizontal rotors (under the influence of gravity), it is
very difficult to observe these phenomena. For horizontal rotors, multiple harmonics are generated
195 | P a g e
when the rotor is running at a constant spinning speed. This aspect can be used as a rubbing
detection indicator. However, a similar response is observed in the case of cracked flexible rotors.
Hence, it is concluded that it is difficult to pinpoint the type of fault based on steady-state vibration
response. On the other hand, transient vibration responses during run-up can easily differentiate
between rotor crack and rubbing action using the short-time Fourier transform.
A hybrid controller, made by combining Mu-synthesis based feedback controller and adaptive
vibrations of the flexible rotor system. Due to the external frequency estimator, the controller is
effective when the spinning speed of the rotor is changed to a large extent. The single input, single
experimentally. Excellent experimental results are obtained. The biggest advantage of the
proposed controller is that multiple harmonics can be quenched efficiently. This aspect can be used
to detect the presence of rotor crack and coupling misalignment from the Fourier transform of the
control signal. However, this technique has one shortcoming also i.e. it cannot pinpoint the type
of fault. We may not be able to distinguish whether the fault is coupling misalignment or rotor
crack.
The precise control of ball screw drives used in CNC machines has been carried out. Due to
the axial flexibility of the rotating screw, the resonance modes are excited. For better command
tracking efficiency, the active vibration control of these resonant modes is necessary. In the
absence of pre-loading, there is a drastic change in the dynamics of the ball screw drive as the
table/carriage moves from one extreme end to the other. The conventional robust control design
techniques cannot maintain stability under these conditions. PPI controllers are normally used in
industry for the large variation in system parameters. However, they have limited command
196 | P a g e
tracking efficiency. A hybrid controller made by combining conventional PPI controller and linear
controller is high in bandwidth and efficient in closed-loop command tracking. It is observed that
For the active chatter control of turning operations in flexible shafts. Improved depth of cuts
and hence productivity is obtained using active chatter control applications. The main difficulty in
active chatter control of these systems is that there is a large variation in dynamic parameters as
the material is removed. Linear parameter varying controllers designed with H-infinity
optimization and interpolation techniques provide robust stability and performance under these
For the feasibility analysis of active chatter control for boring operations, stationary and
revolving bar boring operations are analysed in detail. Mathematical modelling of both types of
boring operations is discussed in detail. The fractional-order PD controllers are tested for active
chatter control applications. The experimental implementation of these controllers confirms their
efficiency. The effect of workpiece ovality and eccentricity on the stability of the system has been
actuators are required for maintaining the closed-loop stability during active chatter controls using
these controllers. It is observed that although the rotating bar boring operation is more stable in
Convolutional neural networks are applied for online chatter detections in milling operations.
Various chatter indicators have been suggested for online implementations. It is observed that the
milling tool locus is a better chatter indicator efficiently to distinguish chatter from non-chatter
conditions in an online manner. On the other hand, the spectrogram of the milling tool deflection
197 | P a g e
is not that much efficient for distinguishing chatter conditions from stable machining conditions
in the presence of measurement noise. Apart from that, it is observed that by using certain
modifications in dynamic equations of milling operations, the system behaviour during chatter
conditions can be predicted to a reasonable accuracy. Experimental results confirm the accuracy
of simulation equations. They are successfully used to train the convolutional neural networks for
Efficiency of the state-of-the-art crack detection techniques is verified in the presence of ball
Transient response-based crack detection strategies using the combination of short-time Fourier
The effect of gravity and rotor configurations on the dynamic behaviour of the rotor system in
such as transverse crack and rotor/stator rubbing action for a two-degree-of-freedom rotor
system.
Model-based determination of size and location of a crack in a rotor system is proposed using
Active vibration control of rotor systems in the presence of rotor crack and coupling
joining feedback and adaptive feedforward cancellation control is implemented to reduce the
198 | P a g e
A hybrid controller made by joining the PPI and Linear Parameter Varying controller is
developed theoretically and experimentally to improve the command tracking efficiency of the
ball screw drive used in CNC machines. Better performance is achieved using the proposed
technique for the BSD system during the metal cutting process.
interpolation techniques is developed theoretically and practically for active chatter controls in
turning operations.
experimentally for active chatter controls of the boring process. It is observed that using the
proposed control strategy, small size electromagnetic actuators can be used in robust controllers.
Feasibility of active chatter control for revolving bar boring operation is investigated.
Online chatter detection techniques for milling operations using convolutional neural networks
automatically adjust the spindle speed so that the system moves from chatter zone to chatter-
free conditions.
The experimental verification of the proposed crack detection techniques in rotor dynamic
199 | P a g e
It is needed for developing control strategies to not only quench the steady-state and transient
vibration but pinpoint the type of fault in the rotor system also. Using fractional calculus, certain
controllers can be developed to quench the rotor vibrations with the minimum control energy
There is a need to develop controllers that are high in bandwidth and do not produce large
variations in torque applied at the motor when the ball screw drive is used for metal cutting
operations.
One of the limitations of LPV controllers is that they require high-cost hardware (like DSPACE
There are certain limitations in the proposed fractional-order controllers for active chatter
control in boring operations. These controllers are low in bandwidth. Hence, there is a clear
need for improving the bandwidth of these controllers in future so that experimental
It is recommended for the online prediction of chatter using Long Short-Term Memory
networks. These neural networks can efficiently classify the pattern/sequence of signals and
hence can be used in combination with the current signal from ball screw drives for online
chatter detection.
200 | P a g e
Papers published/submitted from dissertation
1. Vashisht, R. K., Peng, Q., 2018. Crack detection in the rotor ball bearing system using
switching control strategy and Short Time Fourier Transform. Elsevier Publication, Journal of
Sound and Vibration 432, 502-529
2. Vashisht, R. K., Peng, Q., 2018. Nonlinear dynamic modelling of the cracked rotor ball bearing
system with emphasis on damage detection capabilities. ASME Publication, Journal of
Vibration and Acoustics, 140, pp 041018-10
3. Vashisht, R. K., Peng, Q., 2019. Adaptive hybrid control of unbalanced vibrations of a
rotor/active magnetic bearing system with coupling misalignment using low cost
instrumentation. Sage Publication, Journal of Vibration and Control. 25(15), pp. 1-24.
4. Vashisht, R. K., Peng, Q., 2019. Improving the cutting rate in turning operations of flexible
stepped shafts with a linear parameter varying controller using the electromagnetic actuator.
Springer Publication, International Journal of Advanced Manufacturing Technology, 104, pp.
1-16.
5. Vashisht, R. K., Peng, Q., 2020. Efficient active chatter mitigation for boring operation by
electromagnetic actuator using optimal fractional order PD-λ controller. Elsevier publication,
Journal of Material Processing Technology, 276, 116423.
6. Vashisht, R. K., Peng, Q., 2020, Feasibility analysis of active chatter control for stationary and
revolving bar boring operations based on magnitude of control forces using Fractional order
PD-λ controller. Springer Publication, International journal of Advanced Manufacturing
Technology, 106, pp. 3957–3974.
7. Rajiv Kumar Vashisht, Qingjin Peng., 2020. Simultaneous fault detection and multivariable
adaptive hybrid control of unbalanced vibrations of a cracked flexible rotor system using low-
cost instrumentation. ASME Publication, Journal of Dynamic Systems, Measurement and
Control, 142 pp. 121001-3
8. Rajiv Kumar Vashisht, Qingjin Peng., 2021, Fractional calculus-based energy-efficient active
chatter control of milling process using small size electromagnetic actuators. ASME Journal
of Vibration and Acoustics, 143 / pp. 011005-1
9. Rajiv Kumar Vashisht, Qingjin Peng. Online chatter detection for milling operations using
neural networks assisted by motor current signals of ball screw drives. ASME journal of
Manufacturing Science and Engineering. Published online with DOI : 10.1115/1.4048001
10. Rajiv Kumar Vashisht, Qingjin Peng. 2020, Linear parameter varying modelling and
evaluation of a hybrid controller for machine tool ball screw drives. Mechanical Systems and
Signal Processing. Submitted in Feb. 2020
11. Rajiv Kumar Vashisht, Qingjin Peng., 2020, Online chatter detection and spindle speed
regulation for milling operations using simulation assisted 2D convolutional neural networks.
Elsevier publication, Journal of Intelligent Manufacturing. Accepted
201 | P a g e
Conference papers published
12. Vashisht, R. K., 2019. Model based determination of size and location of a crack in a rotor
system using wavelet transformation and pattern search optimization. ASME 2019
International Design Engineering Technical Conferences and Computers and Information in
Engineering Conference, IDETC2019-97072.
13. Vashisht, R. K., Peng, Q., 2019. Dynamic modelling and diagnosis of transverse crack and
rotor/stator rub in a flexible rotor system. ASME 2019 International Design Engineering
Technical Conferences and Computers and Information in Engineering Conference,
DETC2019-97047.
14. Vashisht, R. K., 2019. The effect of gravity and rotor configuration on dry friction whip and
safe rotor/stator clearance. ASME 2019 International Design Engineering Technical
Conferences and Computers and Information in Engineering Conference, DETC2019-97048.
202 | P a g e
Bibliography
Abrahamsson, J., Hedlund, M., Kamf, T., Bernhoff, H., 2014. High-Speed Kinetic Energy Buffer:
Optimization of Composite Shell and Magnetic Bearings. IEEE Transactions on Industrial Electronics
61, 3012-3021.
Altintas, Y., Verl, A., Brecher, C., Uriarte, L., Pritschow, G., 2011. Machine tool feed drives. CIRP Annals
60, 779-796.
Andrén, L., Häkansson, L., Claesson, I., 2003. Active control of machine tool vibrations in external turning
operations. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering
Manufacture 217, 869-872.
Aslan, D., Altintas, Y., 2018. On-line chatter detection in milling using drive motor current commands
extracted from CNC. International Journal of Machine Tools and Manufacture 132, 64-80.
Babu, Ramesh., Srikanth, S., Sekhar, A.S., 2008. Hilbert–Huang transform for detection and monitoring of
crack in a transient rotor. Mechanical Systems and Signal Processing 22, 905-914.
Behzad, M., Alvandi, M., Mba, D., Jamali, J., 2013. A finite element-based algorithm for rubbing induced
vibration prediction in rotors. Journal of Sound and Vibration 332, 5523-5542.
Bozorg, O., Solgi, M., Loáiciga, H., 2017, Meta‐Heuristic and Evolutionary Algorithms for Engineering
Optimization,Wiley, New York.
Breda, D., Maset, S., Vermiglio, R., 2012. Approximation of Eigenvalues of Evolution Operators for Linear
Retarded Functional Differential Equations.
Breda, D., Maset, S., Vermiglio, R., 2014. Pseudospectral methods for stability analysis of delayed
dynamical systems. International Journal of Dynamics and Control 2, 143-153.
Cao, H., Lei, Y., He, Z., 2013. Chatter identification in end milling process using wavelet packets and
Hilbert–Huang transform. International Journal of Machine Tools and Manufacture 69, 11-19.
Chen, C.-S., Fan, Y.-H., Tseng, S.P., 2006. Position command shaping control in a retrofitted milling
machine. International Journal of Machine Tools and Manufacture 46, 293-303.
Chen, C., Dai, L., Fu, Y., 2007. Nonlinear response and dynamic stability of a cracked rotor.
Communications in Nonlinear Science and Numerical Simulation 12, 1023-1037.
Chen, F., Lu, X., Altintas, Y., 2014. A novel magnetic actuator design for active damping of machining
tools. International Journal of Machine Tools and Manufacture 85, 58-69.
Chen, G., 2009. Study on Nonlinear Dynamic Response of an Unbalanced Rotor Supported on Ball Bearing.
Journal of Vibration and Acoustics 131.
Chen, Q., Liu, G., Zheng, S., 2015. Suppression of imbalance vibration for AMBs controlled driveline
system using double-loop structure. Journal of Sound and Vibration 337, 1-13.
Chen, Y.S., Zhang, H.B., 2011. Review and prospect on the research of dynamics of complete aero-engine
systems. Hangkong Xuebao/Acta Aeronautica et Astronautica Sinica 32, 1371-1391.
Chen, Z., Zhang, H.-T., Zhang, X., Ding, H., 2013. Adaptive Active Chatter Control in Milling Processes.
Journal of Dynamic Systems, Measurement, and Control 136, 021007-021007-021007.
Cheng, L., Li, N., Chen, X.-F., He, Z.-J., 2011. The influence of crack breathing and imbalance orientation
angle on the characteristics of the critical speed of a cracked rotor. Journal of Sound and Vibration 330,
Choi, T., Shin, Y.C., 2003. On-Line Chatter Detection Using Wavelet-Based Parameter Estimation. Journal
of Manufacturing Science and Engineering 125, 21-28.
Chu, Z., Ding, M., Du, S., Feng, X., 2011. Exponential Stability, Semistability, and Boundedness of a
Multi-ANF System. IEEE Transactions on Circuits and Systems I: Regular Papers 58, 326-335.
Collins, K.R., Plaut, R.H., Wauer, J., 1991. Detection of Cracks in Rotating Timoshenko Shafts Using Axial
Impulses. Journal of Vibration and Acoustics 113, 74-78.
203 | P a g e
Cong, F., Chen, J., Dong, G., Huang, K., 2011. Experimental validation of impact energy model for the
rub–impact assessment in a rotor system. Mechanical Systems and Signal Processing 25, 2549-2558.
Darpe, A.K., Gupta, K., Chawla, A., 2004a. Coupled bending, longitudinal and torsional vibrations of a
cracked rotor. Journal of Sound and Vibration 269, 33-60.
Darpe, A.K., Gupta, K., Chawla, A., 2004b. Transient response and breathing behaviour of a cracked
Jeffcott rotor. Journal of Sound and Vibration 272, 207-243.
Delio, T., Tlusty, J., Smith, S., 1992. Use of Audio Signals for Chatter Detection and Control. Journal of
Engineering for Industry 114, 146-157.
Dimarogonas, A.D., 1996. Vibration of cracked structures: A state of the art review. Engineering Fracture
Mechanics 55, 831-857.
Djurovic, I., 2007. Estimation of the Sinusoidal Signal Frequency Based on the Marginal Median DFT.
IEEE Transactions on Signal Processing 55, 2043-2051.
Doppenberg, E.J.J., Faassen, R.P.H., Van de Wouw, N., Oosterling, J.A.J. 2006, CIRP (Eds.), Active in-
process chatter control, 12–13, Vancouver, British Columbia, Canada.
Duan, M., Yoon, D., Okwudire, C.E., 2018. A limited-preview filtered B-spline approach to tracking
control – With application to vibration-induced error compensation of a 3D printer. Mechatronics 56,
287-296.
Dumanli, A., Sencer, B., 2018. Optimal high-bandwidth control of ball-screw drives with acceleration and
jerk feedback. Precision Engineering 54, 254-268.
Eppel, A., Enikov, E.T., Insperger, T., Gabor, S., 2010. Feasibility Study of Optical Detection of Chatter
Vibration During Milling. International Journal of Optomechatronics 4, 195-214.
Erkorkmaz, K., Altintas, Y., 2001. High speed CNC system design. Part III: high speed tracking and
contouring control of feed drives. International Journal of Machine Tools and Manufacture 41, 1637-
1658.
Eynian, M., Altintas, Y., 2009. Chatter Stability of General Turning Operations With Process Damping.
Journal of Manufacturing Science and Engineering 131, 041005-041005-041010.
Fang, J., Xu, X., Tang, J., Liu, H., 2013a. Adaptive complete suppression of imbalance vibration in AMB
systems using gain phase modifier. Journal of Sound and Vibration 332, 6203-6215.
Fang, J., Zhou, X., Liu, G., 2013b. Precise Accelerated Torque Control for Small Inductance Brushless DC
Motor. IEEE Transactions on Power Electronics 28, 1400-1412.
Faassen, R.P.H., Doppenberg, E.J.J., VandeWouw, N., Oosterling, J.A.J., Nijmeijer, H., 2006
CIRP(Eds.),Online detection of the onset and occurrence of machine tool chatter in the milling process,
12–13, Vancouver, British Columbia, Canada
Ganguli, A., Deraemaeker, A., Horodinca, M., Preumont, A., 2005. Active Damping of Chatter in Machine
Tools - Demonstration with a ‘Hardware-in-the-Loop’ Simulator. Proceedings of the Institution of
Mechanical Engineers, Part I: Journal of Systems and Control Engineering 219, 359-369.
Ganguli, A., Deraemaeker, A., Preumont, A., 2007. Regenerative chatter reduction by active damping
control. Journal of Sound and Vibration 300, 847-862.
Gasch, R., 1993a. A Survey Of The Dynamic Behaviour Of A Simple Rotating Shaft With A Transverse
Crack. Journal of Sound and Vibration 160, 313-332.
Gasch, R., 1993b. A Survey Of The Dynamic Behaviour Of A Simple Rotating Shaft With A Transverse
Crack.
Gordon, D.J., Erkorkmaz, K., 2013. Accurate control of ball screw drives using pole-placement vibration
damping and a novel trajectory prefilter. Precision Engineering 37, 308-322.
Gu, D. W., Petkov, P. H., and Konstantinov, M. M., 2005, Robust Control Design With MATLAB,
Springer, London
Han, Q., Chu, F., 2012. Parametric instability of a rotor-bearing system with two breathing transverse
cracks. European Journal of Mechanics - A/Solids 36, 180-190.
204 | P a g e
Hanifzadegan, M., Nagamune, R., 2015. Tracking and Structural Vibration Control of Flexible Ball–Screw
Drives With Dynamic Variations. IEEE/ASME Transactions on Mechatronics 20, 133-142.
Harsha, S.P., Sandeep, K., Prakash, R., 2004. Non-linear dynamic behaviors of rolling element bearings
due to surface waviness. Journal of Sound and Vibration 272, 557-580.
Hoang, D.-T., Kang, H.-J., 2019. A survey on Deep Learning based bearing fault diagnosis.
Neurocomputing 335, 327-335.
Hou, L., Chen, Y., Cao, Q., Lu, Z., 2016. Nonlinear vibration analysis of a cracked rotor-ball bearing system
during flight maneuvers. Mechanism and Machine Theory 105, 515-528.
Insperger, T., Stépán, G., 2004. Updated semi-discretization method for periodic delay-differential
equations with discrete delay. International Journal for Numerical Methods in Engineering 61, 117-
141.
Insperger, T., Stépán, G., Bayly, P.V., Mann, B.P., 2003. Multiple chatter frequencies in milling processes.
Journal of Sound and Vibration 262, 333-345.
Ishida, Y., Ikeda, T., Yamamoto, T., 1990. Nonlinear Forced Oscillations Caused by Quartic Nonlinearity
in a Rotating Shaft System. Journal of Vibration and Acoustics 112, 288-297.
Ismail, F., Ziaei, R., 2002. Chatter suppression in five-axis machining of flexible parts. International Journal
of Machine Tools and Manufacture 42, 115-122.
Jacquet-Richardet, G., Torkhani, M., Cartraud, P., Thouverez, F., Nouri Baranger, T., Herran, M., Gibert,
C., Baguet, S., Almeida, P., Peletan, L., 2013. Rotor to stator contacts in turbomachines. Review and
application. Mechanical Systems and Signal Processing 40, 401-420.
Jalan, A. K., Mohanty A. R., 2009, Model based fault diagnosis of rotor bearing system for misalignment
and imbalance under steady state conditions, Journal of Sound and Vibration 327, 604-622.
Jastrzębski, R.P., Pöllänen, R., 2009. Centralized optimal position control for active magnetic bearings:
comparison with decentralized control. Electrical Engineering 91, 101-114.
Jemielniak, K., Widota, A., 1989. Numerical simulation of non-linear chatter vibration in turning.
International Journal of Machine Tools and Manufacture 29, 239-247.
Jen, M.U., Magrab, E.B., 1996. The Dynamic Interaction of the Cutting Process, Workpiece, and Lathe’s
Structure in Facing. Journal of Manufacturing Science and Engineering 118, 348-358.
Ji, Z., Zu, J.W., 1998. Method of multiple scales for vibration analysis of rotor shaft systems with non-
linear bearing pedestal model. Journal of Sound and Vibration 218, 293-305.
Jiang, J., Ulbrich, H., 2004. The Physical Reason and the Analytical Condition for the Onset of Dry Whip
in Rotor-to-Stator Contact Systems. Journal of Vibration and Acoustics 127, 594-603.
Jun, O.S., Eun, H.J., Earmme, Y.Y., Lee, C.W., 1992. Modelling and vibration analysis of a simple rotor
with a breathing crack. Journal of Sound and Vibration 155, 273-290.
Kakinuma, Y., Sudo, Y., Aoyama, T., 2011. Detection of chatter vibration in end milling applying
disturbance observer. CIRP Annals 60, 109-112.
Kamalzadeh, A., Erkorkmaz, K., 2007. Compensation of Axial Vibrations in Ball Screw Drives. CIRP
Annals 56, 373-378.
Kato, S., Marui, E., 1974. On the Cause of Regenerative Chatter Due to Workpiece Deflection. Journal of
Engineering for Industry 96, 179-186.
Klein, R.G., Nachtigal, C.L., 1975. A Theoretical Basis for the Active Control of a Boring Bar Operation.
Journal of Dynamic Systems, Measurement, and Control 97, 172-178.
Kulesza, Z., 2014. Dynamic behavior of cracked rotor subjected to multisine excitation. Journal of Sound
and Vibration 333, 1369-1378.
Kulesza, Z., Sawicki, J., 2011. Auxiliary state variables for rotor crack detection. Journal of Vibration and
Control 17, 857-872.
Kuljanic, E., Sortino, M., Totis, G., 2008. Multisensor approaches for chatter detection in milling. Journal
of Sound and Vibration 312, 672-693.
Kuljanic, E., Totis, G., Sortino, M., 2009. Development of an intelligent multisensor chatter detection
system in milling. Mechanical Systems and Signal Processing 23, 1704-1718.
205 | P a g e
Lalanne, M., Ferraris, G., Rotordynamics 1990, Prediction in Engineering, Wiley, New York.
Lee, C.W., Han, Y.S., 1998. The directional wigner distribution and its applications. Journal of Sound and
Vibration 216, 585-600.
Li, D., Cao, H., Zhang, X., Chen, X., Yan, R., 2019. Model predictive control based active chatter control
in milling process. Mechanical Systems and Signal Processing 128, 266-281.
Li, F., Jiang, Y., Li, T., Ehmann, K.F., 2018. Compensation of dynamic mechanical tracking errors in ball
screw drives. Mechatronics 55, 27-37.
Li, S., Qun Li, Z., Zheng, M., 2012. Chatter Stability Modeling and Simulation for Long Slender Shaft
Turning Operation.
Liao, Y.S., Young, Y.C., 1996. A new on-line spindle speed regulation strategy for chatter control.
International Journal of Machine Tools and Manufacture 36, 651-660.
Liu, C., Zhu, L., Ni, C., 2018. Chatter detection in milling process based on VMD and energy entropy.
Mechanical Systems and Signal Processing 105, 169-182.
Liu, H., Zhang, J., Zhao, W., 2017a. An Intelligent Non-Collocated Control Strategy for Ball-Screw Feed
Drives with Dynamic Variations. Engineering 3, 641-647.
Liu, J., Yang, L., Yang, L., Zhang, X., Chen, X., 2017b. Multiple-harmonic amplitude and phase control
method for active noise and vibration reshaping. Journal of Vibration and Control 24, 3173-3193.
Lu, K., Lian, Z., Gu, F., Liu, H., 2018. Model-based chatter stability prediction and detection for the turning
of a flexible workpiece. Mechanical Systems and Signal Processing 100, 814-826.
Lu, X., Chen, F., Altintas, Y., 2014. Magnetic actuator for active damping of boring bars. CIRP Annals 63,
369-372.
Ma, H., Shi, C., Han, Q., Wen, B., 2013. Fixed-point rubbing fault characteristic analysis of a rotor system
based on contact theory. Mechanical Systems and Signal Processing 38, 137-153.
Ma, H., Wu, J., Yang, L., Xiong, Z., 2017. Active chatter suppression with displacement-only measurement
in turning process. Journal of Sound and Vibration 401, 255-267.
Ma, H., Wu, Z., Tai, X., Wen, B., 2014. Dynamic characteristics analysis of a rotor system with two types
of limiters. International Journal of Mechanical Sciences 88, 192-201.
Ma, H., Zhao, Q., Zhao, X., Han, Q., Wen, B., 2015. Dynamic characteristics analysis of a rotor–stator
system under different rubbing forms. Applied Mathematical Modelling 39, 2392-2408.
Mann, B.P., Insperger, T., Bayly, P.V., Stépán, G., 2003. Stability of up-milling and down-milling, part 2:
experimental verification. International Journal of Machine Tools and Manufacture 43, 35-40.
Manngård, M., Böling, J.M., 2017. Online frequency estimation with applications to engine and generator
sets. Mechanical Systems and Signal Processing 91, 233-249.
Markert, R., Platz, R., Seidler, M., 2001. Model Based Fault Identification in Rotor Systems by Least
Squares Fitting. International Journal of Rotating Machinery 7.
Marui, E., Ema, S., Kato, S., 1983. Chatter Vibration of Lathe Tools. Part 1: General Characteristics of
Chatter Vibration. Journal of Engineering for Industry 105, 100-106.
Mayes, I.W., Davies, W.G.R., 1984. Analysis of the Response of a Multi-Rotor-Bearing System Containing
a Transverse Crack in a Rotor. Journal of Vibration, Acoustics, Stress, and Reliability in Design 106,
139-145.
McCormack, A.S., Godfrey, K.R., Flower, J.O., 1994. The detection of and compensation for nonlinear
effects using periodic input signals, 1994 International Conference on Control - Control '94., pp. 297-
302 vol.291.
Mei, C., Cherng, J.G., Wang, Y., 2005. Active Control of Regenerative Chatter During Metal Cutting
Process. Journal of Manufacturing Science and Engineering 128, 346-349.
Mei, D., Kong, T., Shih, A.J., Chen, Z., 2009. Magnetorheological fluid-controlled boring bar for chatter
suppression. Journal of Materials Processing Technology 209, 1861-1870.
Meng, H., Kang, Y., Chen, Z., Zhao, Y., Liu, G., 2015. Stability Analysis and Stabilization of a Class of
Cutting Systems With Chatter Suppression. IEEE/ASME Transactions on Mechatronics 20, 991-996.
206 | P a g e
Merritt, H.E., 1965. Theory of Self-Excited Machine-Tool Chatter: Contribution to Machine-Tool Chatter
Research—. Journal of Engineering for Industry 87, 447-454.
Mojiri, M., Bakhshai, A.R., 2004. An adaptive notch filter for frequency estimation of a periodic signal.
IEEE Transactions on Automatic Control 49, 314-318.
Monnin, J., Kuster, F., Wegener, K., 2014a. Optimal control for chatter mitigation in milling—Part 1:
Modeling and control design. Control Engineering Practice 24, 156-166.
Monnin, J., Kuster, F., Wegener, K., 2014b. Optimal control for chatter mitigation in milling—Part 2:
Experimental validation. Control Engineering Practice 24, 167-175.
Mushi, S.E., Lin, Z., Allaire, P.E., 2012. Design, Construction, and Modeling of a Flexible Rotor Active
Magnetic Bearing Test Rig. IEEE/ASME Transactions on Mechatronics 17, 1170-1182.
OKUMA.MachiningNavi,MacNavi.Available at:
http://www.okuma.co.jp/english/onlyone/process/index.html.
Oleaga, I., Pardo, C., Zulaika, J.J., Bustillo, A., 2018. A machine-learning based solution for chatter
prediction in heavy-duty milling machines. Measurement 128, 34-44.
Otto, A., Khasawneh, F.A., Radons, G., 2015. Position-dependent stability analysis of turning with tool and
workpiece compliance. The International Journal of Advanced Manufacturing Technology 79, 1453-
1463.
Park, Y., 2014. Design and implementation of an electromagnetic levitation system for active magnetic
bearing wheels. IET Control Theory & Applications 8, 139-148.
Patel, T.H., Darpe, A.K., 2008a. Influence of crack breathing model on nonlinear dynamics of a cracked
rotor. Journal of Sound and Vibration 311, 953-972.
Patel, T.H., Darpe, A.K., 2008b. Vibration response of a cracked rotor in presence of rotor–stator rub.
Journal of Sound and Vibration 317, 841-865.
Patel, T.H., Darpe, A.K., 2009. Study of coast-up vibration response for rub detection. Mechanism and
Machine Theory 44, 1570-1579.
Pratt, J., Nayfeh, A., 2001. Chatter control and stability analysis of a cantilever boring bar under
regenerative cutting conditions.
Pritschow, G., Croon, N., 2013. Ball screw drives with enhanced bandwidth by modification of the axial
bearing. CIRP Annals 62, 383-386.
Pritschow, G., Eppler, C., Lehner, W.D., 2003. Ferraris Sensor – The Key for advanced dynamic Drives.
CIRP Annals 52, 289-292.
Rahman, M., Ito, Y., 1986. Detection of the onset of chatter vibration. Journal of Sound and Vibration 109,
193-205.
Babu, R., T., Srikanth, S., Sekhar, A.S., 2008. Hilbert–Huang transform for detection and monitoring of
crack in a transient rotor. Mechanical Systems and Signal Processing 22, 905-914.
Rao, B.C., Shin, Y.C., 1999. A comprehensive dynamic cutting force model for chatter prediction in
turning. International Journal of Machine Tools and Manufacture 39, 1631-1654.
Ren, Y., Su, D., Fang, J., 2013. Whirling Modes Stability Criterion for a Magnetically Suspended Flywheel
Rotor With Significant Gyroscopic Effects and Bending Modes. IEEE Transactions on Power
Electronics 28, 5890-5901.
Rodriguez, P., Luna, A., Candela, I., Mujal, R., Teodorescu, R., Blaabjerg, F., 2011. Multiresonant
Frequency-Locked Loop for Grid Synchronization of Power Converters Under Distorted Grid
Conditions. IEEE Transactions on Industrial Electronics 58, 127-138.
Sawicki, J.T., Friswell, M.I., Kulesza, Z., Wroblewski, A., Lekki, J.D., 2011a. Detecting cracked rotors
using auxiliary harmonic excitation. Journal of Sound and Vibration 330, 1365-1381.
Sawicki, J.T., Storozhev, D.L., Lekki, J.D., 2011b. Exploration of NDE Properties of AMB Supported
Rotors for Structural Damage Detection. Journal of Engineering for Gas Turbines and Power 133.
Schmitz, T., 2003. Chatter recognition by a statistical evaluation of the synchronously sampled audio signal.
Journal of Sound and Vibration, 262, 721-730.
207 | P a g e
Schmitz, T.L., Davies, M.A., Medicus, K., Snyder, J., 2001. Improving High-Speed Machining Material
Removal Rates by Rapid Dynamic Analysis. CIRP Annals 50, 263-268.
Schmitz, T.L., Medicus, K., Dutterer, B., 2002. Exploring once-per-revolution audio signal variance as a
chatter indicator. Machining Science and Technology 6, 215-233.
Schoukens, J., Pintelon, R., Rolain, Y., Dobrowiecki, T., 2001. Frequency response function measurements
in the presence of nonlinear distortions. Automatica 37, 939-946.
Sencer, B., Ishizaki, K., Shamoto, E., 2015. High speed cornering strategy with confined contour error and
vibration suppression for CNC machine tools. CIRP Annals 64, 369-372.
Sencer, B., Tajima, S., 2016. Frequency Optimal Feed Motion Planning in Computer Numerical Controlled
Machine Tools for Vibration Avoidance. Journal of Manufacturing Science and Engineering 139,
011006-011006-011013.
Sepasi, D., Nagamune, R., Sassani, F., 2012. Tracking Control of Flexible Ball Screw Drives With Runout
Effect and Mass Variation. IEEE Transactions on Industrial Electronics 59, 1248-1256.
Shang, Z., Jiang, J., Hong, L., 2011. The global responses characteristics of a rotor/stator rubbing system
with dry friction effects. Journal of Sound and Vibration 330, 2150-2160.
Shah, P. , Agashe, S., 2016, Review of fractional PID controller, Mechatronics 38, 29–41
Shawky, A.M., Elbestawi, M.A., 1998. Model-Based Predictive Control of Workpiece Accuracy in Bar
Turning. Journal of Manufacturing Science and Engineering 120, 57-67.
Shimada, Y., Nishimura, Y., Usagawa, T., Ebata, M., 1999. Active control for periodic noise wih variable
fundamental. An extended DXHS algorithm with frequency tracking ability.
Shimoda, T., Fujimoto, H., Kumagai, N., Terada, Y., 2017. External sensorless adaptive chatter avoidance
in NC machining by applying disturbance observer using high resolution linear encoder, 2017 IEEE
International Conference on Advanced Intelligent Mechatronics (AIM), pp. 879-884.
Shiraishi, M., Yamanaka, K., Fujita, H., 1991. Optimal control of chatter in turning. International Journal
of Machine Tools and Manufacture 31, 31-43.
Silani, M., Ziaei-Rad, S., Talebi, H., 2013. Vibration analysis of rotating systems with open and breathing
cracks. Applied Mathematical Modelling 37, 9907-9921.
Singhose, W.E., Searing, W.P., Singer, N.C., 1996. Improving repeatability of coordinate measuring
machines with shaped command signals. Precision Engineering 18, 138-146.
Sinou, J.-J., 2007. Effects of a crack on the stability of a non-linear rotor system. International Journal of
Non-Linear Mechanics 42, 959-972.
Sobie, C., Freitas, C., Nicolai, M., 2018. Simulation-driven machine learning: Bearing fault classification.
Mechanical Systems and Signal Processing 99, 403-419.
Stepan, G., Kiss, A.K., Ghalamchi, B., Sopanen, J., Bachrathy, D., 2017. Chatter avoidance in cutting
highly flexible workpieces. CIRP Annals 66, 377-380.
Sun, Z., Pritschow, G., Lechler, A., 2016. Enhancement of feed drive dynamics using additional table speed
feedback. CIRP Annals 65, 357-360.
Sun, Z., Pritschow, G., Zahn, P., Lechler, A., 2018. A novel cascade control principle for feed drives of
machine tools. CIRP Annals 67, 389-392.
Tanaka, H., Obata, F., Matsubara, T., Mizumoto, H., 1994. Active Chatter Suppression of Slender Boring
Bar Using Piezoelectric Actuators. JSME international journal. Ser. C, Dynamics, control, robotics,
design and manufacturing 37, 601-606.
Tangjitsitcharoen, S., 2009. In-process monitoring and detection of chip formation and chatter for CNC
turning. Journal of Materials Processing Technology 209, 4682-4688.
Tangjitsitcharoen, S., Moriwaki, T., 2008. Intelligent monitoring and identification of cutting states of chips
and chatter on CNC turning machine. Journal of Manufacturing Processes 10, 40-46.
Tansel, I.N., Li, M., Demetgul, M., Bickraj, K., Kaya, B., Ozcelik, B., 2012. Detecting chatter and
estimating wear from the torque of end milling signals by using Index Based Reasoner (IBR). The
International Journal of Advanced Manufacturing Technology 58, 109-118.
Tewani, S.G., Rouch, K.E., Walcott, B.L., 1995. A study of cutting process stability of a boring bar with
active dynamic absorber. International Journal of Machine Tools and Manufacture 35, 91-108.
208 | P a g e
Tiwari, M., Gupta, K., Prakash, O., 2000. Effect of radial internal clearance of a ball bearing on the
dynamics of a balanced horizontal rotor. Journal of Sound and Vibration 238, 723-756.
Tomizuka, M., 1987. Zero Phase Error Tracking Algorithm for Digital Control. Journal of Dynamic
Systems, Measurement, and Control 109, 65-68.
Torkhani, M., May, L., Voinis, P., 2012. Light, medium and heavy partial rubs during speed transients of
rotating machines: Numerical simulation and experimental observation. Mechanical Systems and
Signal Processing 29, 45-66.
Tsai, N.-C., Chen, D.-C., Lee, R.-M., 2010. Chatter prevention for milling process by acoustic signal
feedback. The International Journal of Advanced Manufacturing Technology 47, 1013-1021.
Tsai, N.C., Chen, D.C., Lee, R.M., 2009. Chatter prevention and improved finish of workpiece for a milling
process. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering
Manufacture 224, 579-588.
Uchiyama, N., 2008. Adaptive two-degree-of-freedom control of feed drive systems, International Journal
of Machine Tools and Manufacture 48(3-4):437-445.
van de Wouw, N., Dijk, N., Schiffler, A., Nijmeijer, H., Abele, E., 2017. Experimental Validation of Robust
Chatter Control for High-Speed Milling Processes, pp. 315-331.
Varanasi, K.K., Nayfeh, S.A., 2004. The Dynamics of Lead-Screw Drives: Low-Order Modeling and
Experiments. Journal of Dynamic Systems, Measurement, and Control 126, 388-396.
Varney, P., Green, I., 2015. Nonlinear phenomena, bifurcations, and routes to chaos in an asymmetrically
supported rotor–stator contact system. Journal of Sound and Vibration 336, 207-226.
Vashisht, R. K., Peng, Q., 2018. Crack detection in the rotor ball bearing system using switching control
strategy and Short Time Fourier Transform. Elsevier Publication, Journal of Sound and Vibration 432,
502-529
Vashisht, R. K., Peng, Q., 2018a. Nonlinear dynamic modelling of the cracked rotor ball bearing system
with emphasis on damage detection capabilities. ASME Publication, Journal of Vibration and
Acoustics, 140, pp 041018-10
Vashisht, R. K., 2019. Model based determination of size and location of a crack in a rotor system using
wavelet transformation and pattern search optimization. ASME 2019 International Design Engineering
Technical Conferences and Computers and Information in Engineering Conference, IDETC2019-
97072
Vashisht, R. K., Peng, Q., 2019a. Dynamic modelling and diagnosis of transverse crack and rotor/stator rub
in a flexible rotor system. ASME 2019 International Design Engineering Technical Conferences and
Computers and Information in Engineering Conference, DETC2019-97047
Vashisht, R. K., 2019b. The effect of gravity and rotor configuration on dry friction whip and safe
rotor/stator clearance. ASME 2019 International Design Engineering Technical Conferences and
Computers and Information in Engineering Conference, DETC2019-97048
Vashisht, R. K., Peng, Q., 2019c. Adaptive hybrid control of unbalanced vibrations of a rotor/active
magnetic bearing system with coupling misalignment using low cost instrumentation. Sage Publication,
Journal of Vibration and Control. 25(15), pp. 1-24.
Vashisht, R. K., Peng, Q., 2019d. Improving the cutting rate in turning operations of flexible stepped shafts
with a linear parameter varying controller using the electromagnetic actuator. Springer Publication,
International Journal of Advanced Manufacturing Technology, 104, pp. 1-16.
Vashisht, R. K., Peng, Q., 2020. Efficient active chatter mitigation for boring operation by electromagnetic
actuator using optimal fractional order PD-λ controller. Elsevier publication, Journal of Material
Processing Technology, 276, 116423
Vashisht, R. K., Peng, Q., 2020a, Feasibility analysis of active chatter control for stationary and revolving
bar boring operations based on magnitude of control forces using Fractional order PD-λ controller.
Springer Publication, International journal of Advanced Manufacturing Technology, 106, pp. 3957–
3974
209 | P a g e
Vela-Martínez, L., Carlos Jauregui-Correa, J., Rodriguez, E., Alvarez-Ramirez, J., 2010. Using detrended
fluctuation analysis to monitor chattering in cutter tool machines. International Journal of Machine
Tools and Manufacture 50, 651-657.
Verl, A., Frey, S., 2012. Improvement of feed drive dynamics by means of semi-active damping. CIRP
Annals 61, 351-354.
Villa, C., Sinou, J.J., Thouverez, F., 2008. Stability and vibration analysis of a complex flexible rotor
bearing system. Communications in Nonlinear Science and Numerical Simulation 13, 804-821.
Wang, M., Fei, R., 2001. On-line chatter detection and control in boring based on an electrorheological
fluid. Mechatronics 11, 779-792.
Wang, Z.C., Cleghorn, W.L., 2002. Stability analysis of spinning stepped-shaft workpieces in a turning
process.
Wang, L., Liang, M., 2009. Chatter detection based on probability distribution of wavelet modulus maxima.
Robotics and Computer-Integrated Manufacturing 25, 989-998.
Wauer, J.r., 1990. On the Dynamics of Cracked Rotors: A Literature Survey. Applied Mechanics Reviews
43, 13-17.
Wei, C., Söffker, D., 2016. Optimization Strategy for PID-Controller Design of AMB Rotor Systems. IEEE
Transactions on Control Systems Technology 24, 788-803.
Weingaertner, W.L., Schroeter, R.B., Polli, M.L., de Oliveira Gomes, J., 2006. Evaluation of high-speed
end-milling dynamic stability through audio signal measurements. Journal of Materials Processing
Technology 179, 133-138.
Xie, C., Wu, Y., Liu, Z., 2018, Modeling and active vibration control of lattice grid beam with piezoelectric
fiber composite using fractional order PD-μ algorithm, Composite Structures 198, 126–134
Yang, Y., Chen, H., Jiang, T., 2015. Nonlinear response prediction of cracked rotor based on EMD. Journal
of the Franklin Institute 352, 3378-3393.
Yao, Z., Mei, D., Chen, Z., 2010. On-line chatter detection and identification based on wavelet and support
vector machine. Journal of Materials Processing Technology 210, 713-719.
Yazdani, D., Mojiri, M., Bakhshai, A., JoÓs, G., 2009. A Fast and Accurate Synchronization Technique
for Extraction of Symmetrical Components. IEEE Transactions on Power Electronics 24, 674-684.
Yi-Hua, F., Yan-Che, J., Rong-Jhe, C., Ying-Tsun, L., Tzu-Wei, W., 2008. Adaptive variable structure
controller design of turbomolecular pump with active magnetic bearings, 2008 3rd IEEE Conference
on Industrial Electronics and Applications, pp. 1060-1065.
Yoon, J.-M., Bahn, W., Kim, T.-I., Han, J.-S., Lee, S.-H., Cho, D.-i.D., 2017. Discrete derivative method
for adaptive notch filter-based frequency estimators. International Journal of Control, Automation and
Systems 15, 668-679.
Yoon, M.C., Chin, D.H., 2005. Cutting force monitoring in the endmilling operation for chatter detection.
Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture
Zhang, C.L., Yue, X., Jiang, Y.T., Zheng, W., 2010. A Hybrid Approach of ANN and HMM for Cutting
Chatter Monitoring. Advanced Materials Research 97-101, 3225-3232.
Zhao, R., Yan, R., Chen, Z., Mao, K., Wang, P., Gao, R.X., 2019. Deep learning and its applications to
machine health monitoring. Mechanical Systems and Signal Processing 115, 213-237.
Zhao, X., Patel, T.H., Zuo, M.J., 2012. Multivariate EMD and full spectrum based condition monitoring
for rotating machinery. Mechanical Systems and Signal Processing 27, 712-728.
Zheng, S., Feng, R., 2016. Feedforward compensation control of rotor imbalance for high-speed
magnetically suspended centrifugal compressors using a novel adaptive notch filter. Journal of Sound
and Vibration 366, 1-14.
210 | P a g e
Appendix A
By using certain excitation signals externally and observing the produced response, various
Suppose the shaft is additionally excited by Active magnetic bearing (AMB), the dynamics of the
overall system can be represented as follows. First of all, the auxiliary harmonic excitation
technique will be explained in brief. If i is the critical or resonance frequency of the rotor, rotor
is the rotor spin frequency or rotor velocity, the auxiliary harmonic excitation frequencies 2 can
Based on the excitation frequency (at the AMB) and the rotational speed of the rotor, the frequency
spectrum of the rotor vibration response will include different frequency components given as
The simple sine test can verify the nonlinear behaviour of the system based on the presence of
high-frequency components in a vibration spectrum. There are other methods like the correlation
coefficient-based method, coherence function method, complex stiffness method and Hilbert
transform method, which can also be used to measure the effect of nonlinearities in a given system.
However, in the present research, the vibration response of the rotor ball bearing system with a
sinusoidal excitation signal will be studied. The concept behind the multi-sine excitation method
is to use specially designed periodic excitations (also called multi-sine excitations) given as follows
(Kulesza, 2014)
i|Page
nh
QAMB (t ) Q(k ) cos( (k )t (k )) (A.3)
k 1
with Q is the vector of the amplitude of forces, ω is the vector of their harmonic components, is
the vector of their phases, and nh is the number of frequency components. The signal is called even
multi-sine if it is composed of even numbers. Similarly, the odd-multi-sine signal can be presented.
The harmonic vector in which every other odd harmonic is left can be shown as follows (Kulesza,
2014)
When a linear system is excited by the above multi-sine excitations, the response will be available
at the input frequencies only. However, multiple frequencies will appear in the response spectrum
for a nonlinear system. In this work, two different odd signals are utilized. One was developed by
and the other was developed by Schoukens et al., (2001) and contain the following harmonics:
The dynamic response of the system with the above mentioned is studied in the present work. The
crack induced nonlinearity in the overall system can be detected by analyzing the dynamic response
of the system.
Sawicki et al., (2011a, 2011b) introduced the above-mentioned technique for crack detection in
rotors. For a healthy rotor, the Fourier Transform of the dynamic response will show a single peak
corresponding to rotor spin frequency. In case of a cracked rotor, multiple frequencies (i.e. nX,
where n is an integer and X is the frequency of the revolving rotor) observed in the Fourier
Transform of the dynamic response (Figure A.1). The nonlinearity caused by parametric excitation
ii | P a g e
(i.e. time-varying stiffness of the cracked rotor) is responsible for this amazing behaviour.
However, in reference (Sawicki et al., 2011a, Figure 20), similar results were reported for a healthy
rotor also. Experimental results are given in that figure. Nothing has been mentioned in that
reference regarding the reason for this behaviour. Our results indicate that ball bearing
nonlinearities are the primary cause of this behaviour. This behaviour may be due to rotor/stator
rubbing action also. However, our model predicts this type of behaviour even in the absence of a
crack in the rotor. This type of behaviour has been resulted due to bearing nonlinearities. Figure
(A.1 b) depicts this behaviour. In this figure (i.e. Figure A.1, parts a-d), the bearing supports are
kept flexible. It has been observed that flexibility of bearing supports influences the system
(a) Healthy rotor with Linear Bearing with flexible supports (b) Healthy rotor with Ball Bearing with flexible supports
0 0
-2
X -2
X
2X
-4 -4
-6 -6
0 50 100 150 0 50 100 150
(c) 40% cracked rotor with Linear Bearing with flexible supports (d) 40% cracked rotor with Ball Bearing with flexible supports
0 0
log(Amplitude (m))
X X
-2 2X -2
2X
-4 -4
-6 -6
0 50 100 150 0 50 100 150
(e) Healthy rotor with Linear Bearing with rigid supports (f) Healthy rotor with Ball Bearing with rigid supports
0 0
-2 X -2 X 2X
-4 -4
-6 -6
0 50 100 150 0 50 100 150
(g) 40% cracked rotor with Linear Bearing with rigid supports (h) 40% cracked rotor with Ball Bearing with rigid supports
0 0
X X 2X
-2 2X 3X -2
-4 4X -4 3X
-6 -6
0 50 100 150 0 50 100 150
Frequency (Hz)
Figure A.1: Vertical response of the system at 27 Hz rotor spin frequency for different system
conditions
The bearing supports are considered as rigid as well as flexible (depending on requirement). By
this act, the system behaves like two Jeffcott rotors (mounted on flexible bearing supports)
connected together side by side (however, not exactly). In this situation, for a 40% cracked rotor
iii | P a g e
with linear bearing characteristics (with flexible bearing supports) only first two harmonics are
generated, the higher harmonics excited reach up to 4 i.e. 1X to 4X in presence of rigid bearing
supports. For a healthy rotor (using flexible bearing supports) in the presence of ball bearings, only
first 2 harmonics are generated. Similar is the case of a 40% cracked rotor. However, for the
cracked rotor mounted on rigid bearing supports, in the presence of ball bearings, 3 harmonics are
generated. Figure A.1 is self-explanatory in this regard. The overall effect of the ball bearing is to
To extensively observe these characteristics, the procedure discussed in ref. Sawicki et al., (2011)
was followed. For n=2, 3 and 4, in Equation (A.1), auxiliary harmonic excitation frequencies (
2 ) at 18 Hz, 45 Hz, and 72 Hz were applied one by one. In the ideal case, for a system having
nonlinearities, the following frequencies should appear in the spectrum as per Equation (A.2)
[9, 18, 27, 36, 45, 54, 63, 72, 81, 90, 99, 108, 117, 126, 135, and 144 Hz]
Figure A.2 shows multiple response frequencies. The rotor spin frequency was held at X=27 Hz.
Auxiliary excitation frequency was fixed at 18 Hz. In this figure, the abbreviation LB represents
the linear bearing (i.e. bearings with linear restoring forces) and BB represents the ball bearing (i.e.
bearings with nonlinear restoring forces). The AMB is located at disk1. For a 40% cracked rotor
using a linear bearing with flexible bearing supports, the response is shown in part (a). It contains
45 Hz harmonics only apart from input frequencies and nX harmonics. In further discussion for
the purpose of simplicity, we will write Y as a summation of input frequencies and nX harmonics.
The healthy rotor system mounted on ball bearings contained frequencies 9 Hz, 36 Hz, 45 Hz apart
from Y (Part b). No other frequencies were observed. The ball bearing nonlinearities are the basic
cause of this behaviour. Part (c) represents the response of the system for a 40% cracked rotor in
the presence of a ball bearing. The corresponding values of the magnitude with and without cracks
iv | P a g e
are shown in the same Figure. It is obvious that there is a marginal difference in response
amplitudes with such a high crack percentage. This fact reduces the significance of AHE as a crack
diagnostic technique.
Rigid bearing supports are considered for the next case (parts d, e, and f in Figure A.2). A number
of nX harmonics were generated (part d). However, in the presence of ball bearings, these
harmonics are reduced. Parts (e and f) show the response of the system with healthy and a 40%
cracked rotor respectively in the presence of ball bearings. Only 36 Hz and 45 Hz harmonics were
generated apart from Y. No other frequencies at 9, 18, 36, 45, 54, 63, 72, 81, 90, 99, 117 Hz were
observed.
To further investigate the discrepancy, the AHE at disk 2 was applied instead of disk 1. The reason
for doing that lies in the fact that the crack is assumed to be lying near disk 2. It cannot be done
practically as AMB is located at disk 1. However, for theoretical investigation, excitation was
applied at Disk 2. With this act (i.e. applying excitation at disk 2), the response spectrum contains
frequencies at 9, 45, 72, 99 Hz apart from Y (Part g of Figure A.2) for a rotor with the linear
bearing. For a healthy rotor with ball bearing, the dynamic response contains frequencies at 9Hz
and 45 Hz apart from Y. Parts (h and i) show that there is very less difference in magnitude at these
frequencies when healthy and cracked rotors mounted on ball bearing are compared. Hence, it is
v|Page
(a) 40% cracked rotor with LB (d) 40% cracked rotor with LB (g) 40% cracked rotor with LB
with Flex. Supp. AHE at disk1 with Rigid Supp. AHE at disk1 with Rigid Supp.AHE at disk2
log(Amplitude (m)) 0
18Hz
0 0
18Hz
18Hz X
72Hz
X
-2
2X
-2 -2
2X 3X 99Hz
45Hz
X 45Hz
2X 4X
-4 -4
3X -4
-6 -6 -6 9Hz
0 20 40 60 80 100 0 20 40 60 80 100 0 50 100 150
(b) Healthy rotor with BB (e) Healthy rotor with BB (h) Healthy rotor with BB
with Flex. Supp. AHE at disk1 with Rigid Supp. AHE at disk1 with Rigid Supp.AHE at disk2
log(Amplitude (m))
0 0 0
18Hz 18Hz
X 45Hz
18Hz
X X 45Hz
-2 -2 -2
45Hz
2X 2X 2X
-4 -4 -4
0 0 0
18Hz 18Hz
X 18Hz
X X 45Hz 72Hz
45Hz 45Hz
-2
2X
-2 -2
2X
2X
-4 -4 -4 3X
9Hz 36Hz 36Hz
-6 -6 -6
0 20 40 60 80 100 0 20 40 60 80 100 0 50 100 150
Frequency (Hz) Frequency (Hz) Frequency (Hz)
Figure A.2: Vertical responses of the system at 27 Hz rotor spin frequency and 18 Hz auxiliary
harmonic excitation frequency
Figure A.3 shows the response of the system with AHE frequencies of 45 Hz and 72 Hz. At 45
Hz AHE, the rotor mounted on the linear bearing (Parts a), the observed frequencies were 18 Hz,
36 Hz, and 72 Hz apart from Y for the 40% cracked rotor. However, for the healthy rotor with a
ball bearing (part b), the response contained 18 Hz harmonics apart from Y. Part (c) contains the
response of the 40% cracked rotor supported on the ball bearing. 18 Hz and 72 Hz frequency
components were seen in the response apart from Y. However, the magnitude is negligible. With
Figure A.3 (parts d, e, and f) shows the response of the system with 72 Hz AHE. For the rotor with
the linear bearing (Parts d), the observed frequency was 45 Hz apart from Y for the 40% cracked
rotor. However, for the healthy rotor with a ball bearing, the response contains no frequency
components apart from Y. Similar is the case for the 40% cracked rotor on ball bearings. It can be
safely concluded that this technique has limited practical utility for the real rotor system with
multiple ball bearings (i.e. having nonlinearities) and flexible bearing supports.
vi | P a g e
(a) 40% cracked rotor with LB at AHM frequency of 45Hz (d) 40% cracked rotor with LB at AHM frequency of 72Hz
log(Amplitude (m)) 0
45Hz
0
72Hz
X X
-2 -2
18Hz 36Hz 2X 72Hz 45Hz 2X
-4 -4
-6 -6
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
(b) Healthy rotor with BB at AHM frequency of 45Hz (e) Healthy rotor with BB at AHM frequency of 72Hz
log(Amplitude (m))
0 0
X X 72Hz
45Hz
-2 18Hz -2
2X
-4 -4
-6 -6
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
(c) 40% Cracked rotor with BB at AHM frequency of 45Hz(f) 40% Cracked rotor with BB at AHM frequency of 72Hz
log(Amplitude (m))
0 0
45Hz X 72Hz
X
-2 -2
18Hz 2X 72Hz 2X
-4 -4
-6 -6
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz) Frequency (Hz)
Figure A.3: Vertical responses of the system at disk 1 at 27 Hz rotor spin frequency and 45 Hz
and 72 Hz auxiliary harmonic excitation frequency
A.3. Application of Multi-Sine Excitation (MSE)
In this method, the excitation signal contains more than one sinusoid. However, the properly
designed signal should contain a sinusoid equal to rotor spin frequency Ω. The numerical analysis
has been performed for the following two constant spin speeds i.e. Ω =27 Hz and Ω =30 Hz
For Ω =27 Hz and ω = [3, 15, 27, 39] Hz. Phase shifts of the harmonics are selected as follows: ϕ
(1) =0o, ϕ (2) = 90o, ϕ (3) =180o, and ϕ (4) =270o (Kulesza, 2014).
For Ω=30 Hz and ω= [2, 6, 14, 22, 30] Hz. Phase shifts of the subsequent harmonics is selected in
a similar way, i.e. ϕ (1) =0o, ϕ (2) =90, ϕ (3) =180o, ϕ (4) =270o, ϕ (5) =360o (Kulesza, 2014).
Figure A.4 (a, b and c) shows vertical vibration responses of the rotor for the first multi-sine
frequency series, i.e. for ω = [3, 15, 27, 39] Hz. In this figure, LL represents the linear bearing and
BB represents the ball bearing. The rotor spin frequency is 27 Hz. The excitation was applied at
Disk 1. In the ideal case, the response spectrum should contain 12 Hz, 24 Hz, 42 Hz, 54 Hz, 66
Hz, 78 Hz and 90 Hz frequencies under the influence of crack (Kulesza, 2014). However, the rotor
vii | P a g e
with linear bearing characteristics contains 12 Hz and 66 Hz as well as Y frequencies only in the
spectrum. This happens due to the effect of flexible bearing supports. Under the influence of ball
bearing nonlinearities, 24 Hz and 78 Hz frequencies are emerged in the spectrum apart from Y
(Figure A.4 b) for rotor under healthy conditions. Part (c) shows that for a 40% cracked rotor only
12 Hz, 24 Hz, 66 Hz and 78 Hz frequencies appear in the spectrum apart from Y. However, the
amplitude at 12 Hz and 66 Hz is considerable. In case of healthy rotor, these frequencies were not
available in the spectrum. But to justify the presence of crack based on these two frequencies is
not logical.
Figure A.4 (d, e, f) shows the response of the system for second multi-sine series i.e. ω = [2, 6, 14,
22, 30] Hz. The rotor spin frequency of 30 Hz was fixed in this case. In case of a 40% cracked
rotor, mounted on linear bearings no specific frequency components apart from Y were available
in the spectrum. However, 8 Hz, 16 Hz, 32 Hz, 44 Hz, 52 Hz frequencies appear in the spectrum
apart from Y for both healthy and cracked rotor mounted on ball bearings. Unfortunately, the
magnitude of vibration at these frequencies is practically same for the healthy and cracked rotor.
viii | P a g e
(a) 40% cracked rotor with LB with odd MSE (d) 40% cracked rotor with LB with even MSE
log(Amplitude (m)) 0 0
6Hz 14Hz
15Hz 27Hz 39Hz 54Hz 2Hz 30Hz
-2 -2 60Hz
3Hz 66Hz
-4 -4
12Hz 22Hz
-6 -6
0 50 100 150 0 10 20 30 40 50 60 70 80
(b) Healthy rotor with BB with odd MSE (e) Healthy rotor with BB with even MSE
log(Amplitude (m))
0 0
15Hz 27Hz 39Hz 2Hz 6Hz 14Hz 30Hz
54Hz 60Hz
-2 3Hz -2
78Hz
-4 -4
(c) 40% cracked rotor with BB with odd MSE (f) 40% cracked rotor with BB with even MSE
log(Amplitude (m))
0 0
15Hz 27Hz 39Hz 2Hz 6Hz 14Hz 30Hz
54Hz 66Hz 60Hz
-2 3Hz -2
78Hz
-4 -4
12Hz
8Hz 22Hz 44Hz 52Hz
-6 -6
0 50 100 150 0 10 20 30 40 50 60 70 80
Frequency (Hz) Frequency (Hz)
Figure A.4: Vertical responses of the system at disk 1 at different rotor spin frequencies and multi-
sine excitation
ix | P a g e