EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2004; 33:647–668 (DOI: 10.1002/eqe.371)

Predictive control of seismic structures with

semi-active friction dampers

Lyan-Ywan Lu∗; †
Department of Construction Engineering; National Kaohsiung First University of Science and Technology;
1 University Road; Yenchao; Kaohsiung 824; Taiwan

SUMMARY
In this paper a predictive control method especially suitable for the control of semi-active friction
dampers is proposed. By keeping the adjustable slip force of a semi-active friction damper slightly
lower than the critical friction force, the method allows the damper to remain in its slip state throughout
an earthquake of arbitrary intensity, so the energy dissipation capacity of the damper can be improved.
The proposed method is formulated in a discrete-time domain and cast in the form of direct output
feedback for easy control implementation. The control algorithm is able to produce a continuous and
smooth slip force for a friction damper and thus avoid exerting the high-frequency structural response
that usually exists in structures with conventional friction dampers. Using a numerical study, the control
performance of a multiple degrees of freedom (DOF) structural system equipped with passive friction
dampers and semi-active dampers controlled by the proposed method are compared. The numerical
case shows that by merely using a single semi-active friction damper and a few sensors, the proposed
method is able to achieve better acceleration reduction than the case using multiple passive dampers.
Copyright ? 2004 John Wiley & Sons, Ltd.

KEY WORDS: friction damper; semi-active control; predictive control; seismic structural control; direct
output feedback

1. INTRODUCTION

In mechanical engineering, dry friction has long been used for reducing the motion of moving
or rotating objects because it is the most e
ective and economic method. Recently, large-size
energy dissipation dampers using dry friction have also been developed and applied to large
civil engineering structures for seismic vibration control and have achieved successful results
[1–3]. The advantages of using friction dampers over the other types of energy dissipation
devices include: materials being less a
ected by the degradation due to aging; materials being

∗ Correspondence to: Lyan-Ywan Lu, Department of Construction Engineering, National Kaohsiung First University
of Science and Technology, 1 University Road, Yenchao, Kaohsiung 824, Taiwan.
† E-mail: [email protected]

Contract=grant sponsor: National Science Council, Taipei, Taiwan; contract=grant number: NSC-89-2625-Z-327-004

Received 25 September 2002

Copyright ? 2004 John Wiley & Sons, Ltd. Accepted 3 October 2003
648 L.-Y. LU

insensitive to the change of ambient temperature; and no material yielding and replacement
problems after a large earthquake; and no uid leaking problems.
A typical friction damper usually consists of a frictional sliding interface and a clamping
mechanism that produces normal contact force on the interface. For a passive friction damper,
the clamping force of the damper is a pre-determined xed value, as is the slip force applied
on the frictional interface. A passive damper will be activated and start to slip once the
seismically induced damper force exceeds the pre-determined slip load. A friction damper is
able to dissipate energy only if the damper is in its slip state; otherwise, the damper is no
di
erent from a regular bracing member. During an earthquake, the energy dissipated by a
friction damper is proportional to the damper slip force; therefore, a friction damper may not
be ecient if the level of the slip force is set too low. On the other hand, if the slip force is set
too high, the damper will not slip for most of the earthquake duration. In this case, the energy
dissipation capacity of the damper may not be fully used and the damper then behaves like
a bracing member rather than a true damper. Another disadvantage of using passive friction
dampers is that during an earthquake the status of a damper may be switching between slip
and stick states, which can result in unwanted high-frequency structural responses.
In order to improve the performance of friction dampers, the concept of semi-active control
is introduced to the dampers. A semi-active friction damper is able to adjust its slip force
by controlling its clamping force in real-time in response to a structure’s motion during an
earthquake [4]. Because of this adaptive nature, a semi-active friction damper is expected to be
more e
ective than a passive damper. On the other hand, just like in active structural control,
the control of semi-active friction dampers requires a feedback control algorithm and on-line
measurement of structural response in order to determine the appropriate level of adjustable
clamping forces of the dampers. Nevertheless, a semi-active control device generally has the
following advantages over an active control one [5]: (1) because the control action is carried
out by adjusting the internal mechanism (i.e., the clamping force for a semi-active friction
damper), the required control stroke and energy can be very small; and (2) because it does not
pump energy into the controlled structures, control instability can be prevented. As mentioned,
the control of semi-active dampers requires a control algorithm. Needless to say, the control
performance of the semi-active dampers signicantly relies on the control algorithm applied.
There have been many studies on the control of semi-active uid-type dampers, such as MR
or ER dampers [6–8]. However, owing to the unique features mentioned above for semi-active
friction-type dampers, the conclusions drawn from the studies on uid-type dampers may not
be applicable to friction-type dampers.
Dowdell and Cherry [9] and Kannan et al. [10] were among the rst researchers to in-
vestigate the dynamic response of structures equipped with semi-active friction-type dampers.
They adopted, respectively, on–o
and bang–bang control methods that are simple and merely
require the measurement of the velocity directions of the controlled structure. However, the
repetitive damper actions of ‘on’ and ‘o
’ may result in a discontinuous control force that
can exert a high-frequency structural response and amplify structural acceleration. Inaudi [11]
proposed a control strategy called modulated homogenous friction that produces a slip force
proportional to the prior local peak of the damper deformation. Similarly, this algorithm may
also cause discontinuous control forces. For controlling a building subjected to an impulsive
ground motion, Akbay and Aktan [12] proposed a simple control algorithm that determines
the next time-step clamping force by one pre-specied increment of the current force at a
xed time step. In order to obtain a better performance, the algorithm requires the proper

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 649

determination of the increment of the clamping force and the control time interval through
a parametric study. Later on, Xu et al. [13] adopted Akbay and Aktan’s algorithm [12] and
studied the feasibility of using semi-active friction dampers for wind-induced vibration control
of large truss structures. Some researchers proposed semi-active control laws that were mod-
ied from the optimal control [14] and the modal control [15]. It was shown in their papers
that these modied control methods were able to produce partially continuous slip forces and
a smoother structural response. Nevertheless, these modied control methods did not account
for the unique features of friction dampers mentioned above, so the damper capacity may not
be fully used.
This paper proposes a new control method named predictive control that takes into account
the features of friction dampers. By predicting a critical value called critical friction force
for a semi-active friction damper, the method is able to produce a slip force that is only
slightly lower than the critical force and constantly keeps the semi-active damper in its slip
state throughout an earthquake, so the motion of the structure can be smoothly reduced. The
organization of the paper is as follows: in Section 2, the derivation of the proposed algorithm
starts from formulating the dynamic equation of motion of a structure with friction dampers
in a discrete-time domain. Then, by using a force balance concept, the critical friction force,
which is dened as the minimum friction force in a damper required to keep the damper
in its stick state, is derived. In Section 3, by using the technique of multiple-step feedback,
the control algorithm is also reformulated in a form suitable for direct output feedback con-
trol. The characteristics of the hysteresis loop of a semi-active damper installed on a single
DOF structure controlled by the proposed algorithm are investigated in Section 4. Finally, in
Section 5 the seismic response of a multiple-story structure equipped with multiple semi-active
friction dampers is numerically studied and compared with the response of the same structure
with multiple passive dampers.

2. ANALYSIS OF STRUCTURES WITH SEMI-ACTIVE FRICTION DAMPERS

2.1. Discrete-time state–space formulation

Consider a multiple-story seismic structure equipped with semi-active friction dampers as

shown in Figure 1. The notations mi , ki and xi (t) in Figure 1 represent the mass, sti
ness and
relative-to-the-ground displacement of the i-th oor, while k b; i and Ni (t) denote the sti
ness of
the bracing and the controllable clamping force of the i-th friction damper, respectively. When
subjected to a seismic force, the motion of the structure may be formulated in a dynamic state
equation, i.e.,
ż(t) = Az(t) + Bu(t) + Ew(t) (1)
where the vector z(t) represents the state of the structure, which contains the relative-to-
ground velocity and displacement of each oor; u(t) denotes the vector of the controllable
friction forces provided by the semi-active friction dampers; w(t) is the vector of ground
accelerations; A denotes the system matrix that is composed of the structural mass, damping
and sti
ness matrices; and B and E represent the distributing matrices of the control forces and
the excitations, respectively. It must be pointed out that although Equation (1) looks similar
to that of a typical active control system, the control forces in u(t) are essentially passive

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
650 L.-Y. LU

sensor
mi
xi

Ni (t)

ki damper
kb,i
x i-1
Ni-1 (t)

ki-1 kb,i-1

sensor

Ground
motion w(t)

Controller

Figure 1. Schematic diagram of a structure with semi-active friction dampers.

friction forces and require a special treatment that will be discussed shortly. The force vector
u(t) depends not only on the damper properties (the friction coecient, time-varying clamping
forces and dampers sti
ness, etc.), but also on the structural dynamic response.
When the structure system described in Equation (1) is further discretized in the time do-
main and the excitation force is assumed to be constant within any time interval, Equation (1)
may be converted into a discrete-time form [16]:

z[k + 1] = A d z[k] + B d u[k] + E d w[k] (2)

where a variable followed by [k] or [k + 1] denotes that the variable is evaluated at the
k-th or the (k + 1)-th time step. Also, in Equation (2), A d = eA t represents the discrete-time
system matrix with t being the time interval (sampling period), while the constant coecient
matrices B d and E d are the discrete-time counterparts of the matrices B and E that may be
written explicitly as:

B d =A−1 (A d − I)B
(3)
E d =A−1 (A d − I)E

2.2. Derivation of critical friction force

In general, a semi-active friction damper is connected to a structure at its two ends. For a
building structure, dampers are usually connected to two adjacent stories as shown in Figure 1,
so the story drifts exerted by a dynamic lateral load will activate the dampers. Let y be a
vector listing all damper elongations (deformations) that are equal to the drifts of the stories

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 651

on which the dampers are installed. At any given instant in time, the relation between y and
the state of the structure z may be written as
y[k] = Dz[k] (4)
where D is a constant matrix and has a dimension of (r × 2n); n is the number of DOFs
of the structure, and r is the total number of semi-active friction dampers. Furthermore, the
damper elongations actually consist of two components, i.e.,
y[k] = yr [k] + yb [k] (5)
where yr [k] contains the slip displacements on the friction interfaces of the dampers, while
yb [k] represents the elastic deformations of the dampers, which are proportional to the axial
member forces of the dampers. Since, for friction dampers, the axial forces are equivalent to
the friction forces, therefore, by the elastic constitutive law for axial members, we have
u[k] = K b yb [k] (6)
where K b is a (r × r) diagonal matrix whose i-th diagonal element k b; i is the sti
ness of the
i-th damper (see Figure 1). By substituting y[k] from Equation (4) into Equation (5) and
yb [k] from Equation (5) into Equation (6), we obtain
u[k] = K b (Dz[k] − yr [k]) (7)
As can be seen in the last equation, the friction force vector u[k] is a function of the current
structural state z[k] as well as of the slip displacements yr [k]. A method of computing u[k]
is discussed below.
The motion of a friction damper has two states exclusively, i.e., stick and slip states, and
at any given instant in time the damper can only be in one state. Supposing that in the time
interval from the (k − 1)-th to the k-th time steps, all dampers are in their stick states, the
following condition must be satised:
yr [k] = yr [k − 1] (8)
Next, by employing the results of Equations (7) and (8), the subtraction of u[k − 1] and u[k]
leads to
u[k] = K b D(z[k] − z[k − 1]) + u[k − 1] (9)
Finally, introducing Equation (2), i.e., a dynamic force equilibrium equation, into Equation (9)
and replacing the index k by k − 1 leads to
ũ[k] = Gz z[k − 1] + Gu u[k − 1] + Gw w[k − 1] (10)
where
Gz =K b D(A d − I)
Gu =K b DB d + I (11)
Gw =K b DE d
Note that in the left-hand side of Equation (10), u[k] is replaced by the symbol ũ[k] to
signify that the damper forces computed by the equation hold under the assumption that all

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
652 L.-Y. LU

dampers are in their stick states and the actual friction forces u[k] might not be equal to ũ[k].
Nevertheless, the vector ũ[k] plays a very important role in deciding the state (stick or slip)
and actual friction force of each damper. The physical meaning of the i-th element ũi [k] in the
vector ũ[k] is the minimum friction force required by the i-th damper at the k-th time step, in
order to keep the damper in its stick state (or to prevent the dampers from entering the slip
state). In this paper, ũi [k] is referred to as the ‘critical friction force’ for the i-th damper.
Inspecting Equation (10), one notes that the vector ũ[k] can be computed immediately, once
z[k − 1], u[k − 1] and w[k − 1] have been determined at the previous time step.

2.3. Determination of damper states and internal friction forces

In this subsection, the method of determining the internal friction forces of friction dampers
is discussed. It is assumed that the friction material in the damper obeys Coulomb’s friction
law and has equal static and dynamic friction coecients. Now, consider a structure with
a single semi-active friction damper. In this case, the actual friction force vector u[k] and
critical force vector ũ[k] shall be reduced to scalars u[k] and ũ[k]. The state (either stick or
slip) of the damper may be judged to be
(a) in the stick state, if
|ũ[k]|¡umax [k] =
N [k] (12a)
(b) in the slip state, if
|ũ[k]|¿umax [k] =
N [k] (12b)
where
is the frictional coecient and N [k] is the time-varying clamping force of the damper.
Once the state of the damper is determined by the last two equations, its frictional force can
be readily determined by


ũ[k] (for stick state)
u[k] = (13)
sgn(ũ[k])
N [k] (for slip state)

where sgn(x) means taking the sign of the variable x. Because ũ[k] represents a resisting
force, the term sgn(ũ[k]) is employed to denote the direction of the resisting slip force. Once
u[k] is obtained from Equation (13) and substituted back into Equation (2), the structural
response z[k + 1] can be determined, and then the simulation process proceeds to compute
the response of the next time step.
Now, consider a more realistic case, i.e., a structure equipped with multiple semi-active
friction dampers. The process of determining the damper forces becomes more complicated
because the status of dampers as a whole can have many possibilities. In order to explain more
clearly, let s and (r − s) denote the numbers of the elements in the vector ũ[k] that satisfy
Equations (12a) and (12b), respectively. Now, the status of the dampers can be classied into
three cases:
(a) Complete slip state: when s = 0, all dampers are in the slip state. In this case, all
elements of ũ[k] satisfy the inequality of Equation (12b). In other words, the critical
friction forces of all dampers exceed their slip forces; therefore, the actual friction force

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 653

of each damper shall be equal to the slip force, i.e.,

ui [k] = sgn(ũi [k])
Ni [k]; i = 1; : : : ; r (14)

(b) Complete stick state: when s = r (recall r is the total number of dampers), all dampers
are in the stick state. In this case, since all dampers satisfy the assumption of Equation
(8), the actual friction force ui [k] of each damper is equal to its critical friction force
ũi [k], i.e.,

ui [k] = ũi [k]; i = 1; : : : ; r (15)

(c) Partial slip state: when 0¡s¡r, some dampers are in the slip state, while the others
are in the stick state. In this case, the actual damper friction forces cannot be obtained
immediately because the assumption of Equation (8) is not completely satised. A
recursive computational procedure is required to compute the damper force and to
make sure all dampers are consistent with the assumptions made. The concept of the
procedure for computing the internal friction forces of multiple friction devices was
elaborated by Wang and Liao [17].
Although, as discussed above, the analysis of semi-active friction dampers at each time
step may involve the above three cases, for the semi-active dampers controlled by the method
proposed in the next section, cases (b) and (c) will never occur. By properly computing the
clamping force Ni [k] for each damper, the proposed method makes all elements of ũ[k] satisfy
the inequality of Equation (12b) at any time step, and thus ensures the dampers stay in case
(a) (i.e., the complete slip state) all the time. This will be discussed in detail in the next
section.

3. PREDICTIVE CONTROL FOR SEMI-ACTIVE FRICTION DAMPERS

3.1. Control with full-state feedback

The energy dissipation rate of a friction damper is the product of the slip force and the slip
rate of the frictional interfaces; therefore, the energy dissipation capacity of a friction damper
may not be best used if the level of the slip force is set too low or the damper does not slip
most of the time. On the other hand, as mentioned in the last section, by using the information
measured at the previous time step, Equation (10) is able to predict the damper critical friction
force (the minimum force required to keep a damper in its slip state) at the current time step.
Because of this discovery, one may use Equation (10) to compute the possible clamping force
that ensures the damper slips all the time. In other words, if one applies a clamping force
Ni [k] for a certain damper such that the resulting slip force is always slightly less than the
value ũi [k] predicted by Equation (10), the damper will remain in the slip state for the entire
time history. Based on this concept, the control rule for determining the clamping force of a
semi-active friction damper is proposed as:

|ũi [k]|
Ni [k] =  ; 06 ¡ 1 (for i = 1–r) (16)

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
654 L.-Y. LU

where  is a selectable constant parameter and ũi [k] is obtained from Equation (10). Sub-
stituting Ni [k] from Equation (16) into the inequality of Equation (12b), one should be able
to verify that the inequality is always true for i = 1–r. This implies that all dampers are al-
ways kept in their slip states. This is a condition for the complete slip-state case discussed
above; therefore, the damper friction forces can be computed by substituting Equation (16)
into Equation (14). After the substitution, one may rewrite Equation (16) in a vector form
and obtain
u[k] = ũ[k] (17)
Basically, Equation (17) states that the damper friction force vector u[k] (also known as the
control forces) is proportional to and also less than ũ[k], because 06¡1. Recall that ũ[k]
is the predicted critical friction forces. For this reason, the aforementioned control law is also
referred to as ‘predictive control’ in this paper. Finally, using Equation (10) in Equation (17),
one obtains an explicit formula for computing the control force vector
u[k] = (Gz z[k − 1] + Gu u[k − 1] + Gw w[k − 1]) (18)
where Gz , Gu and Gw are given in Equation (11). After being multiplied by the factor ,
these matrices may also be treated as the control gains. It is obvious in Equation (18) that
the parameter  plays an important role in the proposed control law. A larger value of  will
lead to a higher control force, but this does not necessarily guarantee better energy dissipation
capacity, as will be seen in the numerical results shown in the latter section.

3.2. Control with direct output feedback

In the control law described by Equation (18), the computation of the vector u[k] not only
requires the value of vector u[k − 1] but also the measurement of excitation vector w[k − 1]
as well as the full-state vector z[k − 1]. The measurement of the ground excitation w[k − 1] is
generally not a problem, since there are merely three components for any given earthquake.
Furthermore, the vector u[k − 1] is computed at the previous time step, so it can be stored in an
electrical memory for immediate use. Nevertheless, the requirement for the full-state feedback
z[k − 1] in Equation (18) makes the control method impractical for structures involving a
large number of DOFs, because the control implementation may require a large number of
sensors. For this reason, in this sub-section a direct output feedback method that requires the
feedback of only partial state variables is derived from Equation (18). This method employs
the technique of multiple-step feedback. First of all, let us dene a sensor measurement vector
ys [k − 1] that is related to the state vector by
ys [k − 1] = Cz[k − 1] (19)
where C is the sensor placement matrix. Let q be the number of sensors deployed in the
structure, so we have the dimensions ys [k − 1] ∈ Rq×1 and C ∈ Rq×2n . Note that usually q is
much less than 2n. Next, by replacing the index k by (k − 2) in Equation (2) and also replacing
the index (k − 1) by (k − 2) in Equation (19), one may solve for the sensor measurement
ys [k − 2] and obtain
ys [k − 2] = CA−1 −1 −1
d z[k − 1] − CA d B d u[k − 2] − CA d E d w[k − 2] (20)

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 655

where the discrete-time system matrix A d is usually square and non-singular [16]. Similarly,
by using Equations (2) and (19) repeatedly, the sensor measurement vector ys [k − m] taken
at the m-th preceding time step may be expressed by

m−1
ys [k − m] = CA−(m−1)
d z[k − 1] − CA−(m−j)
d B d u[k − j − 1]
j=1


m−1
− CA−(m−j)
d E d w[k − j − 1] (for m¿2) (21)
j=1

where the matrix operation A−a −1 a

d = (A d ) (for a¿0). Finally, putting Equations (19) and
Equation (21) (representing m − 1 equations, for m¿2) together, one can combine them into
an expanded matrix equation, i.e.,

 − 1] = A
y[k  d u [k − 1] + E
 d z[k − 1] + B  d w[k
 − 1] (22)

where
   

 ys [k − 1] 
 
 u[k − 1] 


 
 
 


   
 ys [k − 2] 
 
 u[k − 2] 

y[k
 − 1] = .. ∈ Rqm×1 ; u [k − 1] = .. ∈ Rmr×1 (23)

 
 
 


 . 
 
 . 


 
 
 

   
ys [k − m] u[k − m]
   
 w[k − 1]  C

 
  

 
  CA−1 

 w[k − 2] 
  d 
w[k
 − 1] = ∈ Rmp×1 ;  d =
A 

 ∈ Rqm×2n (24)
 ..   .. 

 . 
  . 

 
  

 

w[k − m] CA−(m−1)
d
 
0 0 0 ··· 0
 
0
 −CA−1
d Bd 0 ··· 0 

d =
B  ∈ Rqm×mr (25)
 .. .. .. .. .. 
. . . . . 
 
0 −CA−(m−1)
d Bd −CA−(m−2)
d Bd ··· −CA−1
d Bd

 
0 0 0 ··· 0
 
0
 −CA−1
d Ed 0 ··· 0 

E d =   ∈ Rqm×mp (26)
 .. .. .. .. .. 
. . . . . 
 
0 −CA−(m−1)
d Ed −CA−(m−2)
d Ed ··· −CA−1
d Ed

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
656 L.-Y. LU

where p denotes the number of independent excitation components. In the above equations,
a vector or a matrix with an over-bar indicates an augmented vector or matrix. The vector
y[k
 − 1] contains the sensor measurements taken at the previous m time steps, namely, the
k-th, (k − 1)-th; : : : ; (k − m)-th time steps, etc. Similarly, u [k − 1] and w[k
 − 1] represent,
respectively, the augmented vectors for the damper forces and excitations recorded from the
preceding m time steps. Furthermore, from Equation (22) one can easily solve for z[k − 1]
and obtain
−1
z[k − 1] = A
 d (y[k
 − 1] − B
 d u [k − 1] − E
 d w[k
 − 1]) (27)
In order to ensure that the matrix A d can be inversed in the last equation, the number of
sensors q and the number of feedback steps m must satisfy the condition qm = 2n, and also
A d must be non-singular. Equation (27) may also be treated as a state estimator. Finally,
substituting Equation (27) into Equation (10) results in a control law of direct output feed-
back, i.e.,
ũ[k] = G y y[k
 − 1] + G u u [k − 1] + G w w[k
 − 1] (28)
where the augmented coecient matrices may be written explicitly
 −1
G y =Gz A d

 −1
G u = −Gz A d B d + G u D1
 (29)
 −1
G w = −Gz A d E d + G w D2


where D1 = [I1 ; 0; : : : ; 0] ∈ Rr×mr , D2 = [I2 ; 0; : : : ; 0] ∈ Rp×mp , I1 ∈ Rr×r and I1 ∈ Rp×p . After ũ[k]
is determined by Equation (28), the controllable clamping force Ni [k] (i = 1–r) for each semi-
active damper can be computed by substituting each element ũi [k] of ũ[k] into Equation (16).
Finally, the resulting control force vector can be evaluated by using Equation (28) in Equa-
tion (17), i.e.,
u[k] = (G y y[k
 − 1] + G u u [k − 1] + G w w[k
 − 1]) (30)

After being multiplied by the factor , the matrices G y , G u and G w may be treated as
the control gains. Through Equations (28) and (30), the proposed control method has been
transformed into a direct output feedback control, since it merely requires the multiple-step
feedback of the sensor outputs y[k
 − 1].

4. CONTROL OF A SINGLE DOF STRUCTURAL SYSTEM

4.1. Hysteretic behavior of a semi-active friction damper

The energy dissipation characteristic of a damper is usually understood by its hysteresis loop.
On the other hand, the hysteresis loop of a semi-active damper greatly depends on the control
algorithm applied. In other words, the same semi-active damper may have di
erent hysteretic
behaviors when using di
erent control algorithms. In order to observe how the proposed
control method alters the hysteretic behavior of a friction damper, in this sub-section a single

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 657

DOF structural system equipped with a semi-active friction damper under a harmonic load is
investigated. The natural frequency and damping ratio of the structure are taken to be 1 Hz
and 5% respectively, and a sinusoidal function w(t) = 0:5g sin(2t) (1g = 9:8 m= s2 ) is chosen
as the ground excitation. The sti
ness of the damper bracing is assumed to be equal to three
times the structural sti
ness.
The hysteresis loops of a semi-active friction damper using the proposed control method
with di
erent values of  are shown in Figures 2(a)–(d). For comparison, the hysteresis
loop of a passive damper is also plotted in Figure 2(e) with a pre-set slip force equal to
30% of the structure weight. In the gures, the x-axis and y-axis represent, respectively, the
damper elongation and the damper axial force that is equivalent to the internal friction force.
Note that in Figure 2(e), the stick state of the passive friction damper can be recognized by
an inclined line segment with a slope equal to the damper sti
ness, while the slip state is
indicated by a horizontal line segment. From Figures 2(a)–(d), it can be observed that the
proposed control method leads to an inclined elliptical hysteresis loop. There are no straight-
line segments shown in the loops, therefore, it can be concluded that the damper remains
in its slip state all the time. The geometric shapes and areas of the loops depend on the
value of . The hysteresis loop becomes a at ellipse when  decreases. On the other hand,
when  increases, the loop becomes slender. The reason is that a larger  results in a larger
damper slip force and thus decreases the slip displacement of the damper. When  gradually
approaches one, the slip force of the semi-active damper approaches the critical friction force.
If  exceeds one, theoretically the damper does not slip at all and becomes a simple bracing,
which does not have the energy dissipation ability. Moreover, if the value of  is too close
to one, owing to the presence of measurement noise or modeling error, it is possible that
the damper may enter the stick state for certain instants in time. This may impair the energy
dissipation capacity of the friction dampers.

4.2. Energy dissipation ratios for di
erent earthquakes

In order to investigate the amount of energy dissipated by the proposed method with di
erent
values of , four di
erent ground accelerations are imposed on the above single DOF sys-
tem. These four ground motions are suggested by the International Association on Structural
Control (IASC) for benchmark control problems [18]. These four ground motions are, re-
spectively, Hachinohe (PGA = 0:250g), El Centro (PGA = 0:348g), Kobe (PGA = 0:834g) and
Northridge (PGA = 0:843g). The rst one represents a modest earthquake, the second one a
strong earthquake, and the third and fourth ones severe earthquakes. In order to facilitate the
comparison, let us dene an energy dissipation ratio Rd as below
EI
Rd = (31)
ED
where EI and ED indicate, respectively, the seismic input energy and the energy dissipated by
the damper [1]. The value of Rd represents the percentage of the seismic energy dissipated
by the damper.
Table I lists the energy dissipation ratio Rd for the four earthquakes with di
erent settings
of the control parameter , and also listed is the energy ratio due to the passive control for
comparison. As expected, in the passive control case, Rd increases as the earthquake intensity
increases from modest to severe. On the other hand, the proposed predictive control with a

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
658 L.-Y. LU

5 hysteretic loop 5 hysteretic loop

x 10 x 10
1.5 1.5

1 1

0.5 0.5
force (NT)

force (NT)
0 0

-0.5 -0.5

-1 -1

-1.5
-1.5
-0.5 0 0.5 -0.5 0 0.5
(a) displacement (m) (b) displacement (m)

5 hysteretic loop 5 hysteretic loop

x 10 x 10
1.5 1.5

1 1

0.5 0.5
force (NT)
force (NT)

0 0

-0.5 -0.5

-1 -1

-1.5 -1.5
-0.5 0 0.5 -0.6 -0.4 -0.2 0 0.2 0.4 0.6
(c) displacement (m) (d) displacement (m)

5 hysteretic Loop
x 10
1

0.5
force (NT)

-0.5

-1
-0.5 0 0.5
(e) displacement (m)

Figure 2. Hysteresis loops of friction damper for di
erent control parameters: (a) predictive control
( = 0:99); (b) predictive control ( = 0:95); (c) predictive control ( = 0:90); (d) predictive control
( = 0:85); and (e) passive control.

xed value of  yields a constant Rd , even though the intensities and waveforms of the four
input ground motions are very di
erent. For example, with  = 0:95, for the four selected
earthquakes, Rd is roughly equal to 78%. This is a potential advantage for the aspect of

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 659

Table I. Comparison of energy dissipation ratios.

Control Energy Hachinohe El Centro Kobe Northridge
method index (PGA = 0:250g) (PGA = 0:348g) (PGA = 0:834g) (PGA = 0:843g)

Passive EI (kN-m) 6.10 27.89 143.79 47.75

ED (kN-m) 0.00 15.81 115.39 35.76
Rd 0.00% 56.70% 80.25% 75.37%

Predictive EI (kN-m) 9.92 21.82 129.25 53.95

 = 0:99 ED (kN-m) 8.90 19.52 115.89 47.86
Rd 89.71% 89.48% 89.66% 88.72%

Predictive EI (kN-m) 13.80 22.11 140.64 42.08

 = 0:95 ED (kN-m) 10.70 17.17 109.29 32.67
Rd 77.53% 77.66% 77.71% 77.63%

Predictive EI (kN-m) 17.07 22.07 142.23 38.10

 = 0:90 ED (kN-m) 10.67 13.85 89.30 23.91
Rd 62.53% 62.77% 62.70% 62.76%

structural control design. Moreover, it is also observed that the energy ratio Rd increases as
the value of  increases. In the cases studied, Rd increased from around 63% to 90% as 
increased from 0.90 to 0.99.

5. CONTROL OF A MULTIPLE DOF STRUCTURAL SYSTEM

5.1. Numerical model

In this section, a multiple DOF structure namely a 3-story building structure, is taken as
an example for the numerical study (see Figure 3). The structural matrices for this example
structure are shown in Table II and were obtained from a system identication test performed
on a real steel frame [19]. The frame was erected on the shaking table in the earthquake sim-
ulation laboratory of the National Center for Research on Earthquake Engineering (NCREE),
Taipei, Taiwan. The frame is primarily devoted to the testing of various structural control
devices. When modeled as a shear building, the three natural frequencies of the frame were
identied to be 1.07, 3.48 and 5:98 Hz. Whenever a damper or a bracing is added to a story
of the structure, the sti
ness of the damper or bracing is assumed to be three times the hori-
zontal story sti
ness. The El Centro earthquake record was used as the input ground motion;
however, the peak ground acceleration (PGA) of the record might be adjusted in order to
investigate the e
ect of control. In the following sub-sections, we rstly investigate the control
performance of the structure equipped with a single friction damper and then with multiple
dampers. Also, for comparison, both controls with semi-active dampers as well as passive
dampers are considered in the study. For the passive control (passive dampers), we dene a
variable called the slip load ratio r as the control parameter:

N0
r= (32)
W

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
660 L.-Y. LU

450

3rd Floor

Steel wavy plate

Concrete blocks with concrete
300

2nd Floor
400 25

30 170 170 30
900 300
50
1st Floor

Semi-active
300
Friction Damper
bracing
Z Ground
Floor 30 Shaking Table
Surface

Y Unit : cm
X

Figure 3. The three-story steel testing frame at NCREE (Taipei, Taiwan).

Table II. Structural matrices of the three-story steel frame [19].

 
10:914 0 0
 
Mass matrix M= 0 11:213 0  ton
0 0 11:213
 
2:124 −1:131 0:093
 
Damping matrix C =  −1:131 2:364 −0:209  kN=(m=s)
0:093 −0:209 2:325
 
3673:845 −4555:417 1172:960
 
Sti
ness matrix K =  −4556:112 8753:602 −5451:810  kN=m
1172:190 −5452:178 9186:159

where N0 is the constant clamping force in the passive damper and W is the total structural
weight. Also, note that in the semi-active control case, only the sensor measurements of the
relative-to-the-ground displacement(s) and velocity(-ies) of the oor(s) that are equipped with
semi-active friction damper(s) are taken to be feedback signals for the control.

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 661

3F Max. Acc. with Elce (scaled)

20
α = 0.99
α = 0.95
α = 0.90
15 passive

acceleration (m/s2)
10

0
0.1 0.2 0.3 0.4 0.5 0.6
(a) PGA (g)

1F Max. Acc. with Elce (scaled)

12
α = 0.99
α = 0.95
10 α = 0.90
passive
acceleration (m/s2)

0
0.1 0.2 0.3 0.4 0.5 0.6
(b) PGA (g)

Figure 4. Comparison of maximum structural accelerations due to the El Centro earthquake:

(a) the 3rd oor; and (b) the 1st oor.

5.2. Control using a single friction damper

Consider a single semi-active friction damper deployed between the ground oor and the
rst oor of the structure as shown in Figure 3. When the structure is subjected to the
El Centro earthquake with di
erent PGA levels, Figure 4 compares the maximum structural
accelerations of the rst and third oors for the predictive control cases with  = 0:9, 0.95
and 0.99. Figure 5 compares the peak values of the required control forces (damper slip
forces). Also shown in Figures 4 and 5 are the peak acceleration and control force of the
passive control case with r = 0:3. From these two gures it can be observed that: (1) in the
predictive control cases, no matter what value of  is chosen, both the structural response
and the control force are linearly proportional to the excitation level, even though the friction
behavior itself is generally non-linear. This implies that the semi-active friction damper using
the proposed control method becomes a linear damper. On the other hand, the passive damper
still behaves non-linearly in Figure 4. The passive damper was activated only when the PGA
level reached a level higher than 0:3g. (2) When compared with the passive control case, the

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
662 L.-Y. LU

Max. Control Force with Elce (scaled)

30
α = 0.99
α = 0.95
25 α = 0.90
passive
20

N
15

4
10
10

0
0.1 0.2 0.3 0.4 0.5 0.6
PGA (g)

Figure 5. Comparison of maximum control forces due to the El Centro earthquake.

proposed method with  = 0:95 or 0.90 provides a better reduction on acceleration response;
at the same time, it demands a lower damper slip force for the earthquake with PGA less
than 0:4g (see Figure 5). (3) A larger value of , which produces a larger damper slip force,
does not necessarily result in a better reduction on the structural responses, so using a very
large  may not be desirable. (4) In Figure 4(b), the peak acceleration of the rst oor of the
passive control case is much higher than the values obtained in the predictive control cases.
Note that the damper was installed on the rst-oor panel. The reason is explained in the
next paragraph.
When the structure is subjected to the El Centro earthquake with a PGA level of 0:4g and
controlled by the proposed method with  = 0:95, Figure 6 compares the acceleration time
histories of the rst and the third oors of the structure. In Figure 6(a) it is shown that
the acceleration signal of the rst oor in the passive control case contains high-frequency
components and its peak acceleration can be as high as that measured at the third oor (see
Figure 6(b)). This is primarily due to frequent interchanges between slip and stick states in
the passive damper. The interchanges produce a discontinuous and higher-frequency motion
of the structure, especially at the oor where the damper is installed. However, the proposed
control method eliminates this discontinuous motion so the resulting structural response is
smoother and the oor acceleration can be e
ectively reduced.

5.3. Control using multiple friction dampers

When passive friction dampers are used for seismic structural control, it is usually required
that the dampers be installed in most of the stories (in order to enhance the energy dissipation
of the whole structural system). The cost of constructing a structure with passive dampers
grows as the number of stories increases. On the other hand, since a semi-active damper can
provide a better energy dissipation capacity, it is expected that the number of semi-active
dampers required for ecient seismic control may be reduced. If this is demonstrated to be
true, the use of semi-active dampers will be more cost e
ective. This subsection investigates
this possibility.

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 663

1F Acc. with Elce (0.4g)

10
passive
pred. α = 0.95

Acc. (m/s2)
0

-5

-1 0
0 5 10 15 20 25 30 35
(a) time (sec)

3F Acc. with Elce (0.4g)

10
passive
pred. α = 0.95

5
Acc. (m/s2)

-5

-1 0
0 5 10 15 20 25 30 35
(b) time (sec)

Figure 6. Comparison of acceleration time histories due to the El Centro earthquake: (a) acceleration
of the 1st oor; and (b) acceleration of the 3rd oor.

For this end, let us consider several possible deployments of dampers and sensors as shown
in Figure 7. The four plots in Figure 7 represent, respectively, the installation of (A) three con-
ventional bracings, (B) three passive friction dampers, (C) three semi-active friction dampers
with full-state feedback, and (D) a single semi-active friction damper at the rst oor with par-
tial state (direct-output) feedback. In cases C and D the proposed control method is adopted
to regulate the clamping force of the semi-active dampers. Note that case D is the most
economically desirable case since it requires fewer dampers and sensors.
Table III summarizes the maximum oor displacements, the maximum oor accelerations as
well as the maximum damper slip forces (control forces) resulting from the di
erent damper
arrangements mentioned above. In order to provide a basis for comparison, the responses
of the bare frame are also given in the second row of Table III. The term ‘bare frame’
means the original structural system without adding any bracing or damper. The El Centro
earthquake with the original PGA (i.e., 0:348g) was used as the input ground motion. Note
that in the table, the terms ‘displacement’ and ‘acceleration’ represent the relative-to-the-
ground displacement and the absolute acceleration, respectively. In the third row of Table III,

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
664 L.-Y. LU

(a) (b)

(c) (d)

Relative velocity measurement Passive damper

Relative Displacement measurement
Semi-active damper

Figure 7. Placement of friction dampers and sensors for di
erent control cases.

it is observed that although the structure with the conventional bracing system may reduce
the top-oor displacement down to 13% of the bare frame displacement, it also increases
the oor acceleration by 33%. This is a shortcoming and a trade-o
of using a bracing
system. Next, from the fourth, fth and sixth rows of Table III, which represent the passive
control case (case B) with three di
erent pre-set slip forces, it is shown that as r increases
from 0.1 to 0.3 the maximum structural displacements are reduced but at the same time
the maximum structural accelerations are increased (although they are still lower than the
maximum responses of the bare frame). This implies that the use of passive dampers with a
high level of slip loads may not be good for acceleration control.
As for semi-active control case C, as shown in the seventh row of Table III, the predictive
control achieves an acceleration reduction ratio that is much superior to both cases A and B
(for all values of r); at the same time, it maintains a displacement reduction as good as that
of case A. Also, as observed in the seventh row of the table, the required maximum control
force for each semi-active friction damper in case C depends on the oor level on which the
damper is installed. The maximum control force has an average value equal to 58:6 kN, which
is lower than the slip load of case B with r = 0:2.

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 Table III. Comparison of structural responses and control forces for di
erent control cases.

Max. oor displacement a Max. oor Max. damper

(relative to the ground) (m) acceleration a (m=s2 ) slip force a (kN)
Control Number of Parameter
Case devices dampers setting 3F 2F 1F 3F 2F 1F 3F 2F 1F

Copyright ? 2004 John Wiley & Sons, Ltd.

Bare frame None 0 none 0.221 0.168 0.076 12.379 9.543 7.104 none none none

A Conventional 0 none 0.029 0.023 0.013 16.437 14.683 9.431 none none none
bracings (0.13) (0.14) (0.17) (1.33) (1.54) (1.33)

B Passive friction 3 r = 0:1 0.059 0.048 0.022 6.129 4.993 4.449 32.7 32.7 32.7
dampers (0.27) (0.29) (0.29) (0.50) (0.52) (0.63)
r = 0:2 0.018 0.017 0.011 6.643 5.023 6.830 65.4 65.4 65.4
(0.08) (0.10) (0.14) (0.54) (0.53) (0.96)
r = 0:3 0.016 0.014 0.009 9.514 7.165 7.137 98.1 98.1 98.1
(0.07) (0.08) (0.12) (0.77) (0.75) (1.00)

C Semi-active friction 3  = 0:95 0.030 0.024 0.013 3.768 3.478 3.314 30.9 68.2 76.6
dampers (full state) (0.14) (0.14) (0.17) (0.30) (0.36) (0.47)
PREDICTIVE CONTROL OF SEISMIC STRUCTURES

D Semi-active friction 1  = 0:95 0.023 0.022 0.019 4.406 4.224 3.884 none none 96.2
damper (direct output) (0.10) (0.13) (0.25) (0.36) (0.44) (0.55)
a Numbers in parentheses denote the ratio to the responses of the bare frame (i.e., the numbers shown in the second row).

Earthquake Engng Struct. Dyn. 2004; 33:647–668

665
666 L.-Y. LU

Floor level
1
Case D (predictive)
Case B (passive, r = 0.1)
Case B (passive, r = 0.2)
Case B (passive, r = 0.3)
Ground 0
0 3 6 9 12
(a) Max. acceleration (m/s^2)

2
Floor level

1
Case D (predictive)
Case B (passive, r = 0.1)
Case B (passive, r = 0.2)
Case B (passive, r = 0.3)
Ground 0
0 0.02 0.04 0.06 0.08
(b) Max.displacement (m)

Figure 8. Comparison of maximum responses at di
erent oors: (a) oor

accelerations; and (b) oor displacements.

The last row of Table III shows the control result of the most economically desirable case,
i.e., case D. Comparing the performance of cases D and C, one can conclude that the former
is almost as e
ective as the latter, although the former requires a larger damper slip force.
Note that case D uses only one semi-active friction damper and measures only the rst-oor
response, while case C requires three semi-active dampers and full-state feedback. This implies
that if the semi-active dampers are able to supply suciently large slip forces, it is possible to
use very few semi-active friction dampers with the proposed control method for the vibration
mitigation of multiple DOF seismic structural systems.
Moreover, it is also interesting to compare the performance of case D (single semi-active
damper case) and case B (multiple passive damper case). Figures 8(a) and (b) plot the
maximum oor accelerations and oor displacements at di
erent stories, respectively. The
following observations can be made from the gures: (1) for suppressing the acceleration
response, the single semi-active damper with predictive control and direct-output feedback is
superior to the case of using multiple passive dampers (for r = 0:1, 0.2 and 0.3). (2) In case

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
 PREDICTIVE CONTROL OF SEISMIC STRUCTURES 667

B, although the passive dampers with a larger value of r = 0:3 e
ectively reduce the maximum
displacement of the top oor down to 7% of the bare frame response, they only reduce the
maximum acceleration to 77%. In case D, the maximum displacement and acceleration of the
top oor are reduced to 10% and 36%, respectively. This implies that the semi-active control
is more e
ective for the simultaneous control of displacement and acceleration responses.

6. CONCLUSIONS

In this paper a predictive feedback control method has been proposed for controlling semi-
active friction dampers that are installed on a seismic structure. By assuming that a friction
damper is in its stick state at the next time step, the proposed method predicts the minimum
friction force required to keep the damper in its stick state and adjust its slip force to be
slightly less than this value. By doing so, the semi-active friction damper is able to remain
in its slip state and continuously dissipate energy throughout an earthquake of any intensity
and waveform. When under a sinusoidal excitation, a semi-active damper controlled by the
proposed method has an elliptical hysteresis loop. The area of the hysteretic ellipse and energy
dissipation ratio of a semi-active damper depend on a single parameter  that can be pre-
selected by the control designer. From the numerical simulation of a multiple DOF structure,
it was observed that semi-active control with the proposed control method is e
ective for both
displacement and acceleration controls. By using a single semi-active friction damper and a
few sensors, the proposed method is able to achieve better results on structural acceleration
control than the case in which all stories are equipped with passive friction dampers. This
demonstrates that semi-active friction dampers with the proposed control method can form an
e
ective and economical semi-active control system.

ACKNOWLEDGEMENTS

This research was sponsored in part by the National Science Council, Taipei, Taiwan, through Grant
No. NSC-89-2625-Z-327-004. The author is grateful to his formal graduate student Mr G. L. Lin for
preparing the tables and plots.

REFERENCES
1. Soong TT, Dargush GF. Passive Energy Dissipation Systems in Structural Engineering. Wiley: New York,
1997.
2. Fitzgerald TF, Anagnos T, Goodson M, Zsutty T. Slotted bolted connections in aseismic design for concentrically
braced connections. Earthquake Spectra 1992; 5(2):383 – 391.
3. Hale TH, Pall R. Seismic upgrade of the Freeport water reservoir, Sacramento. Proceedings of the 12th World
Conference on Earthquake Engineering, Auckland, New Zealand, 2000, paper 0269.
4. Dupont P, Kasturi P, Stokes A. Semi-active control of friction dampers. Journal of Sound and Vibration 1997;
202(2):203 – 218.
5. Symans MD, Constantinou MC. Semi-active control systems for seismic protection of structures: a state-of-the-
art review. Engineering Structures 1999; 21:469 – 487.
6. Spencer BF Jr, Dyke SJ, Sain MK, Carlson JD. Phenomenological model for magnetorheological dampers.
Journal of Engineering Mechanics (ASCE) 1997; 123(3):230 – 238.
7. Symans MD, Constantinou MC. Seismic testing of a building structure with a semi-active uid damper control
system. Earthquake Engineering and Structural Dynamics 1997; 26:759 –777.

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668
668 L.-Y. LU

8. Proceedings of the 3rd World Conference on Structural Control, 2002, Como, Italy. John Wiley & Sons, Ltd,
Chichester, UK.
9. Dowdell DJ, Cherry S. Structural control using semi-active friction dampers. Proceedings of the First World
Conference on Structural Control, Los Angeles, 1994; Vol. 3:FA1-59 – 68.
10. Kannan S, Uras HM, Aktan HM. Active control of building seismic response by energy dissipation. Earthquake
Engineering and Structural Dynamics 1995; 24:747–759.
11. Inaudi JA. Modulated homogeneous friction: a semi-active damping strategy. Earthquake Engineering and
Structural Dynamics 1997; 26:361– 376.
12. Akbay Z, Aktan HM. Abating earthquake e
ects on buildings by active slip brace devices. Shock and Vibration
1995; 2:133 –142.
13. Xu YL, Qu WL, Chen ZH. Control of wind-excited truss tower using semi-active friction damper. Journal of
Structural Engineering (ASCE) 2001; 127:861– 868.
14. Dowdell DJ, Cherry S. On passive and semi-active friction damping for seismic response control of structures.
Proceedings of the 11th World Conference on Earthquake Engineering, Acapulco, Mexico, 23 – 28 June, 1996;
paper 957.
15. Lu LY, Lin GL. Seismic protection of structures using semi-active friction dampers. Proceedings of the 9th
International Conference on Computing in Civil and Building Engineering, Taipei, Taiwan, 3 – 5 April, 2002;
537– 542.
16. Meirovitch L. Dynamics and Control of Structures. Wiley: New York, 1990.
17. Wang YP, Liao WH. Dynamic analysis of sliding structures with unsynchronized support motions. Earthquake
Engineering and Structural Dynamics 2000; 29:297– 313.
18. Ohtori Y, Christenson RE, Spencer BF Jr, Dyke SJ. Benchmark control problems for seismically excited
nonlinear buildings. CRIEPI report U00055, Central Research Institute of Electric Power Industry: Tokyo,
2001.
19. Lu LY, Chung LL. Modal control of seismic structures using augmented state matrix. Earthquake Engineering
and Structural Dynamics 2001; 30:237– 256.

Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2004; 33:647–668