EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2002; 31:235–259 (DOI: 10.1002/eqe.107)

Closed-form solution for seismic response of adjacent

buildings with linear quadratic Gaussian controllers

Y. L. Xu∗; † and W. S. Zhang

Department of Civil and Structural Engineering; The Hong Kong Polytechnic University; Hung Hom;
Kowloon; Hong Kong

SUMMARY
Closed-form solution for seismic response of adjacent buildings connected by hydraulic actuators with
linear quadratic Gaussian (LQG) controllers is presented in this paper. The equations of motion of
actively controlled adjacent buildings against earthquake are =rst established. The complex modal su-
perposition method is then used to determine dynamic characteristics, including modal damping ratio,
of actively controlled adjacent buildings. The closed-form solution for seismic response of the system
is =nally derived in terms of the complex dynamic characteristics, the pseudo-excitation method and
the residue theorem. By using the closed-form solution, extensive parametric studies can be carried
out for the system of many degrees of freedom. The bene=cial parameters of LQG controllers for
achieving the maximum response reduction of both buildings using reasonable control forces can be
identi=ed. The eAectiveness of LQG controllers for this particular application is evaluated in this study.
The results show that for the adjacent buildings of diAerent dynamic properties, if the parameters of
LQG controllers are selected appropriately, the modal damping ratios of the system can be signi=cantly
increased and the seismic responses of both buildings can be considerably reduced. Copyright ? 2001
John Wiley & Sons, Ltd.

KEY WORDS: closed-form solution; adjacent buildings; LQG controller; seismic response; parametric
study

1. INTRODUCTION

Active structural vibration control has been investigated and utilized to sustain the safety
and serviceability of civil engineering structures in the last two decades [1; 2]. A number

∗ Correspondence to: Y. L. Xu, Department of Civil and Structural Engineering, The Hong Kong Polytechnic
University, Hung Hom, Kowloon, Hong Kong.
† E-mail: ceylxu@polyu.edu.hk

Contract=grant sponsor: Hong Kong Polytechnic University.

Contract=grant sponsor: National Natural Science Foundation of China; contract=grant number: NNSF-50038010.

Received 30 September 2000

Revised 10 April 2001
Copyright ? 2001 John Wiley & Sons, Ltd. Accepted 30 April 2001
236 Y. L. XU AND W. S. ZHANG

of real implementations have also been realized in many places [3]. Correspondingly, many
control algorithms have been proposed speci=cally for civil engineering structures, such as
instantaneous optimal control [4], sliding mode control [5], linear quadratic Gaussian (LQG)
control [6; 7] and its frequency domain analog H2 control [8]. By using LQG controllers
or H2 controllers, the estimation and control design processes can be fully separated with
the design of the estimator independent of feedback to the structure [7]. Thus, much of the
stability and performance of LQG controllers can be achieved despite partial state information
and noise corruption [2]. This feature and its simplicity account for the popularity of LQG
and H2 controllers over other controllers. Accordingly, LQG controllers are selected in this
study for controlling adjacent buildings connected by hydraulic actuators against earthquake.
The use of supplemental damping devices to couple adjacent buildings for controlling
wind- or earthquake-induced responses has attracted more and more attention from engi-
neers and researchers since buildings in modern cities become more and more close to
each other. Gurley et al. [9], Kamagata et al. [10], Fukuda et al. [11], Luco et al. [12]
and Zhang and Xu [13] have studied the feasibility and eAectiveness of using passive de-
vices to connect adjacent buildings. Seto et al. [14], Yamada et al. [15] and Christenson
et al. [16] have investigated the case of connecting adjacent buildings by active control or
semi-active control devices. All these studies have come up with encouraging results that
for adjacent buildings of diAerent dynamic properties, if control devices are arranged and
designed properly, wind or earthquake-induced responses of both buildings can be reduced
signi=cantly.
Since wind or earthquake excitation is a random disturbance in nature, the statistical re-
sponses of structures with LQG controllers are usually determined in two direct ways: one
is to determine the response spectral density function through transfer function =rst and then
integrate the spectral density function to obtain the statistical response; and the other is to
solve a Lyapunov equation for the covariance matrix of the response [17]. Since this study
concerns the application of LQG controller to connect two buildings which involve many
degrees of freedom, both the aforementioned ways need great computational eAorts for =nd-
ing bene=cial parameters of LQG controllers and determining statistical responses of both
buildings. Therefore, this paper aims to =nd general yet simple closed-form solution for ac-
tively controlled adjacent buildings with LQG controllers against earthquake. The derivation
for closed-form solution is naturally ful=lled by combining the complex modal superposition
method with the pseudo excitation method and the residue theorem. The derived closed-form
solution is then used to perform parametric studies of adjacent buildings connected by LQG
controllers and to assess the eAectiveness of LQG controllers in reducing seismic responses
of both buildings. The dynamics of actuators and their interaction with structures are not
considered [18].

2. EQUATIONS OF MOTION

Consider a two-dimensional system consisting of two linear elastic shear buildings connected
by active hydraulic actuators at each Moor of the same level, as shown in Figure 1. The
assumption of linear elastic buildings indicates that only small and moderate seismic events

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 237

Figure 1. Structural model of adjacent buildings with LQG controllers.

are concerned or due to the eAective active control, the buildings are able to remain elastic
and linear properties under severe seismic event. The mass of each building is concentrated
at its Moor and the stiAness is provided by its massless columns. Both buildings are as-
sumed to be subjected to the same base acceleration and any eAects due to spatial variations
of the ground motion are neglected. Both buildings and actuators are manipulated by LQG
controllers.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
238 Y. L. XU AND W. S. ZHANG

Assume that a total of degrees of freedom of two adjacent buildings are N (see Figure 1),
in which the number of degrees of freedom of the left building is L with its =rst Moor
designated as the =rst degree of freedom. N − L is then the number of degrees of freedom of
the right building with its =rst Moor designated as the L + 1 degree of freedom. The equations
of motion of the building-control device system can be expressed as
O + Cẋ(t) + Kx(t) = − ME xO g (t) + Hu(t)
M x(t) (1)
where M; C and K are the mass, damping and stiAness matrices of the adjacent buildings,
respectively, x(t) is the vector of relative displacement response with respect to the ground
with the left building’s displacements in the =rst L positions and the right building’s displace-
ments in the last N − L positions, E the index vector with all its elements equal to 1, u(t)
a r-dimensional vector consisting of r active control forces, H a N × r matrix denoting the
location of r actuators, and xO g (t) the ground acceleration.
The details of mass, damping and stiAness matrices, M; C, and K can be found in Zhang
and Xu [13]. In this study, the Kanai–Tajimi =ltered white noise spectrum is considered as
the ground acceleration spectrum.
1 + 4
g2 (!=!g ) 2
SxOg xOg (!) = S0 (2)
[1 − (!=!g ) 2 ] 2 + 4
g2 (!=!g ) 2

in which !g ;
g ; S0 may be regarded as the characteristics and the intensity of an earthquake
in a particular geological location. This spectrum can be represented in the time domain with
the following state equations:

Ṡ(t) = A c S(t) + Dc w(t) (3)

xO g (t) = C c S(t) (4)

where w(t) is zero-mean Gaussian white noise with intensity S0 and S(t) is the state of the
seismic excitation model.



0 1 2 0
Ac = ; C c = [!g 2
g !g ]; Dc = (5)
−!g2 −2
g !g 1

Writing Equation (1) as a state equation and then combining it with Equations (3) and (4)
yield the equation of motion of the system:
q̇(t) = Aq(t) + Bu(t) + Gw(t) (6)
where

       
0 I 0 0 0 x(t)
A =  −M−1 K −M−1 C −EC c  ; B =  M−1 H  ; G=  0 ; q(t) =  ẋ(t) 
0 0 Ac 0 Dc S(t)
(7)

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 239

3. LQG CONTROLLER

In reality, the structural states, that is, displacements and velocities relative to ground with each
degree of freedom, cannot be fully measured. The measurement is usually limited to absolute
accelerations [19–21]. The absolute accelerations of adjacent buildings can be related to the
relative displacements and relative velocities through the following equation:

xO a (t) = − M−1 Cẋ(t) − M−1 Kx(t) + M−1 Hu(t) (8)

If output measurements are selected as m(t) = xO a (t) − M−1 Hu(t), then the measured output
vector m(t) can be expressed as

m(t) = Cm q(t) + vm (t) (9)

where vm (t) are random signals known as measurement noises and

Cm = [−M−1 K − M−1 C 0] (10)

For practical application, sensors may not be placed at every Moor and thus only a subset
of m(t), denoted as y(t), is used.

y(t) = CL q(t) + v(t) (11)

in which the dimension of y(t) is n (n6N ). CL is a matrix obtained by eliminating the

rows related to the Moors without sensors in the matrix Cm . v(t) is the measurement noise
vector of n dimension. If the measurement noise is regarded as the white noise of the same
intensity at each measurement point and independent of each other, its covariance matrix can
be expressed as

E[v(t)vT (t + )] = ISv () (12)

where E is the expectation operator, I the identity matrix, Sv the intensity of noise and ()
is the Dirac delta function.
The LQG cost function is


Tf
1 T T
J = lim E (q(t) Qq(t) + u(t) Ru(t)) dt (13)
Tf →∞ 2Tf −Tf

where Q and R are positive-semi-de=nite and positive-de=nite weighting matrices, respec-

tively.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
240 Y. L. XU AND W. S. ZHANG

A reasonable controller design for the LQG control problem can be obtained by using the
linear quadratic regulator feedback gain matrix Kc which operates on the state estimate q̂
generated by the Kalman =lter:

u(t) = −Kc q̂(t) (14)

Kc = R−1 BT X2 (15)

in which, X2 is the solution of the following algebraic Riccati equation:

X2 A + AT X2 − X2 BR−1 BT X2 + Q = 0 (16)

The state equation for the Kalman =lter [22] is given by

˙ = Aq̂(t) + Bu(t) + Kf (y(t) − CL q̂(t))

q̂(t) (17)

where Kf is the estimator gain matrix given by

Kf = Y2 CTL Sv−1 (18)

Y2 is the solution of the following algebraic Riccati equation:

AY2 + Y2 AT − Y2 CTL CL Y2 Sv−1 + GGT S0 = 0 (19)

4. CLOSED-FORM SOLUTIONS

4.1. Solution for dynamic characteristics

Equations (6), (11), (14) and (17) can be rearranged as

ṗ(t) = Âp(t) + f(t) (20)

where




A − BKc BKc G 0 w(t)
 = ; f= ;
0 A − Kf CL G −Kf v(t)

 (21)
q(t)
p= ; e(t) = q(t) − q̂(t)
e(t)

The solution of the homogeneous form of Equation (20) can be then taken as

p(t) = et (22)

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 241

The associated complex eigenvalue problem of Equation (20) becomes

 =  (23)

where  is the eigenvalue and  is the associated eigenvector. The solution of Equation
(23) comprises a set of 2N  (N  = 2N + 2) eigenvalues and eigenvectors that exist in either
complex conjugate pairs (underdamped mode) or real pairs (overdamped mode). For complex
conjugate pairs,

∗ ∗ 
j = j+N
and j = j+N
(j = 1; 2; : : : ; N ) (24)

The eigenvalue is usually written under the form

∗ 
j = j+N
= −!j
j + i!dj (j = 1; 2; : : : ; N ) (25)

in which

!j = |j |;
j = −Re(j )= |j | and !dj = !j 1 −
j2 (26)

For real pairs, it is convenient to express real pairs j in the following form analogous to
Equation (25)

j = −!j
j + !dj ; j+N
= −!j
j − !dj (27)

!j ; !dj and
j are determined by

 
!j = j j+N
;
j = −(j + j+N
)=(2!j ) and !dj = !j
j2 − 1 = (j −j+N
)=2 (28)

where !j ; !dj , and

j are the modal frequency, the damped modal frequency and the modal
damping ratio, respectively, associated with mode j. The superscript ∗ means the conjugation
and i is the imaginary unit.

4.2. Solution for seismic response

To =nd the closed-form solution for seismic responses of adjacent buildings with LQG con-
trollers, the pseudo-excitation method [13] is used in conjunction with the complex modal
superposition method. The seismic input w(t) and measurement noise vector v(t) are assumed
to be independent in this study. The spectral density matrix Swv of both the ground excitation

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
242 Y. L. XU AND W. S. ZHANG

and the measurement noise is thus given by



S0 0 1 0
Swv = = S0 (29)
0 Sv I 0 2 I

in which

Sv
2 = (30)
S0

Note that



1 0 1 0
Swv = S0
0 I 0 I

Thus, letting


1 0
L=
0 I

the pseudo-excitation vectors for the system expressed by Equation (20) can be constituted
as [13; 23]


w(t)
= Lk S0 exp(i!t) (k = 1; 2; : : : ; n + 1) (31)
v(t) k

where L k is the kth column of L.

Considering the kth pseudo-excitation vector


w(t)
v(t) k

Equation (20) becomes

ṗk (t) = Âpk (t) + Fk S0 exp(i!t) (32)

where


G 0
Fk = Lk (33)
G −Kf

To decouple Equation (32), the following co-ordinate transformation is adopted:

pk (t) = Uzk (t) (34)

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 243

where zk is the 2N  -dimensional generalized co-ordinate vector and U is the 2N  × 2N  right

modal matrix, that is,
U = [1 ; 2 ; : : : ; 2N
] (35)
By using this transformation, Equation (32) can be reduced to 2N  decoupled modal equa-
tions with the jth modal equation being

ż kj (t) = j zkj (t) + rkj S0 exp(i!t) (36)
where
T
rkj = j Fk (37)

j is the jth column of V, and V = U−T , the left modal matrix.

The solution of the =rst-order jth Equation (36) to the kth pseudo excitation vector is
rkj
zkj (!; t) = S0 exp(i!t) (j = 1; 2; : : : ; 2N  ) (38)
i! − j
Denoting the mth components of j as mj and the mth component of pk as pk m , then the
pseudo response pk m is given by
rkj




2N 
2N
pk m (!; t) = mj zkj (!; t) = mj S0 exp(i!t) (39)
j=1 j=1 i! − j
Since the eigenvectors are in pairs for either underdamped mode or overdamped mode,
Equation (39) can be reduced to

N


pk m = Hj (!)(i! k mj + !k mj ) S0 exp(i!t) (40)
j=1

in which Hj (!) is the frequency response function for the jth mode.
1
Hj (!) = (41)
!j2 − !2 + i2
j !j !
The pseudo response pk m can be the pseudo displacement response, pseudo velocity re-
sponse and pseudo acceleration response. The proper use of the pseudo displacement responses
of adjacent buildings can result in the pseudo shear force responses of both buildings. Also,
the proper use of the pseudo state estimator responses in conjunction with Equation (14)
can lead to the pseudo control forces. For instance, the mth (m6N ) pseudo displacement,
velocity, or acceleration response can be obtained from Equation (40) if the coeWcients k mj
and !k mj are calculated by the following equations:
When the jth mode is an underdamped mode

 2Re(mj rkj ) for displacement


k mj = 2Re(j mj rkj ) for velocity (42)



2Re(j2 mj rkj ) for acceleration

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
244 Y. L. XU AND W. S. ZHANG



 −2Re(mj rkj j∗ ) for displacement


!k mj = −2!j2 Re(mj rkj ) for velocity (43)




−2!j2 Re(j mj rkj ) for acceleration

When the jth mode is an overdamped mode


 mj rkj + mj+N
rkj+N
for displacement


k mj = j mj rkj + j+N
mj+N
rkj+N
for velocity (44)



j2 mj rkj + j+N
2

mj+N
rkj+N
for acceleration


 −(mj rkj j+N
+ mj+N
rkj+N
j ) for displacement


!k mj = −!j2 (mj rkj + mj+N
rkj+N
) for velocity (45)



 −!2 (  r + 

j j mj kj j+N mj+N
rkj+N
) for acceleration

According to the principle of the pseudo-excitation method, the response spectral density
of pk m can be then obtained by

Spk m pk m (!) = pk m (!)pk∗m (!)


N 
N
= Hi (!)Hj∗ (!)(!k mi !k mj + i!( k mi !k mj − k mj !k mi ) + !2 k mi k mj )S0
i=1 j=1
(46)

The variance response of pki under the kth pseudo-excitation can be evaluated as
 +∞
2
#pk m = Spk m pk m (!) d! (47)
−∞

The above integration in the complex plane can be accomplished using the residue theorem
to have the closed-form solution as








N 
N 
N 
N 
N 
N
#p2k m = $0; ij uk mi uk mj + $01; ij uk mi vk mj + $1; ij vk mi vk mj (48)
i=1 j=1 i=1 j=1 i=1 j=1

where
 √  √
& !k mi S0 & !k mj S0
uk mi = ; uk mj =
2 !i
i !i 2 !j
j !j
 √  √ (49)
& k mi S0 & k mj S0
vk mi = ; vk mj =
2
i !i 2
j !j

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 245

Figure 2. Top Moor displacement response of left building vs parameter '.

8
i
j (
i + (
j )(3=2
$0; ij = (50)
(1 − (2 )2 + 4
i
j ((1 + (2 ) + 4(
2i +
j2 )(2

8
i
j ((2 − 1)(3=2
$01; ij = (51)
(1 − (2 )2 + 4
i
j ((1 + (2 ) + 4(
2i +
j2 )(2

8
i
j (
j + (
i )(3=2
$1; ij = (52)
(1 − (2 )2 + 4
i
j ((1 + (2 ) + 4(
2i +
j2 )(2

!j
(= (53)
!i

The =nal variance response of pm can be determined by a summation with respect to the
n + 1 pseudo excitation vectors.


n+1
#p2m = #p2k m (54)
k=1

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
246 Y. L. XU AND W. S. ZHANG

Figure 3. Top Moor displacement response of right building vs parameter '.

The closed-form solution derived above makes it possible to carry out extensive parametric
studies and to evaluate the performance of both buildings with LQG controllers.

5. APPLICATION

For application, two 20-storey buildings having the same Moor elevations with actuators con-
necting two neighbouring Moors are used. The mass, shear stiAness, and internal and ex-
ternal damping coeWcients of the left building are uniform for all storeys with the mass
of 1:29 × 106 kg, the shear stiAness of 4:0 × 109 N=m, the internal damping coeWcient of
3:0 × 106 N s=m, and the external damping coeWcient of 8:0 × 104 N s=m. For the right build-
ing, the mass, shear stiAness, internal damping coeWcient and external damping coeWcient are
also uniform for all storeys with the same mass, internal and external damping coeWcients as
the left building but with the shear stiAness of 2:0 × 109 N=m only. Hence, the two buildings
have the same height but the right building is slender than the left building. Such a selection
is made because if the dynamic properties of both buildings are the same or close to each
other, the use of control devices linking the two buildings will not function or not function
properly [24]. The damping coeWcients selected ensure the =rst modal damping ratio in either
building is around 1 per cent. The parameters for the ground motion (see Equation (2)) are

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 247

Figure 4. Base shear force response of left building vs parameter '.

selected as !g = 15:0 rad=s,

g = 0:65 and S0 = 4:65 × 10−4 m2 =rad s3 , respectively, to reMect
=rm-soil conditions [25]. The intensity ratio of the measurement noise to the ground motion,
, in Equation (30) is selected as 1=7. The active control devices and sensors are arranged
=rst at every Moor and then at selected locations for a comparison.

5.1. Selection of weighting matrices

The performance of LQG controllers depends on the weighting matrices R and Q. To achieve
the bene=cial performance of LQG for the maximum reduction of key structural responses of
both buildings with reasonable control forces, several potential weighting matrices are selected.
The key structural responses are then computed in terms of the closed-form solution, from
which the best weighting matrices can be identi=ed. The basic con=gurations of the weighting
matrices in this study are taken as
 
QdL 0 0 0 0
 0 QdR 0 0 0
 
R = 10−8 I; Q = '   0 0 Q vL 0 0 
 (55)
 0 0 0 QvR 0 
0 0 0 0 0

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
248 Y. L. XU AND W. S. ZHANG

Figure 5. Base shear force response of right buildings vs parameter '.

where the submatrices QdL and QdR are assigned to the displacement responses of the left
and right buildings, respectively, the submatrices QvL and QvR are assigned to the velocity
responses of the left and right buildings, respectively, the matrix R is allocated to the control
forces and ' is a proportional coeWcient. Clearly, by varying the coeWcient ', a proper
trade oA between control eAectiveness and control energy consumption can be achieved. In
consideration that the present study involves two buildings, the four cases are selected for
=nding the balanced submatrices QdL , QdR , QvL and QvR . These four cases are de=ned as
follows:

Case A: QdL = QdR = QvL = QvR = I,

Case B: QdL = QdR = I; QvL = QvR = 0,
Case C: QdL = QdR = 0; QvL = QvR = I;
Case D: QdL = QdR = I; QvL = QvR = 0:1 I.

In Case A, the four submatrices are equally assigned; in Case B, the assignment to the
two displacement submatrices only indicates that the displacement response reduction is max-
imized regardless of the velocity response reduction; Case C indicates that the velocity re-
sponse reduction is maximized regardless of the displacement response reduction; and Case D

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 249

Figure 6. Control force of adjacent buildings vs parameter '.

implies that the displacement response reduction is given priority over the velocity response
control.
For each case mentioned above, the key responses of actively controlled adjacent buildings
are computed against the parameter '. Then, by comparing the results among all the cases
the bene=cial case and parameter ' can be found for achieving the maximum or bene=cial
response reduction of both buildings with reasonable control forces. Figures 2 and 3 depict the
variations of the top Moor displacement responses of the left and right buildings, respectively,
with the parameter '. Figures 4 and 5 display the variations of the base shear force responses
of the left and right buildings, respectively. Figure 6 shows how the control force at the top of
the building varies with the parameter '. It is seen that for all cases the top Moor displacement
and base shear force responses of both buildings are rapidly reduced until ' reaches a value
of about 2 × 105 . After that, the gradients of response reduction become small in Cases A, B
and D, but not in Case C, where the further increase of ' makes the displacement and shear
force responses to become larger with larger control force required. Thus, Case C, where the
velocity response reduction is maximized regardless of the displacement response reduction,
is disregarded in the subsequent computation. Furthermore, if the weighting matrices in Case
A are adopted, one may bene=t from the response reduction of the left building but not the
right building. The situation is reversed if the weighting matrices in Case B is used. As a
result, a compromise is made in this study to select the weighting matrices of Case D.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
250 Y. L. XU AND W. S. ZHANG

Figure 7. Variations of modal frequencies with parameter '.

5.2. Modal properties

By using the weighting matrix Q in Case D, the =rst =ve natural frequencies and modal
damping ratios of the actively controlled adjacent buildings are computed against the parameter
'. The results are depicted in Figures 7 and 8 for the natural frequencies and modal damping
ratios, respectively. It is seen from Figure 7 that the =rst =ve natural frequencies do not vary
with the parameter ' when ' is in the range from 10 to 2 × 105 . Further analysis shows
that the =rst, third and =fth natural frequencies of the actively controlled adjacent buildings
are almost the same as the =rst three natural frequencies of the uncontrolled right building
of 3.02, 9.03 and 14:99 rad=s, respectively. The second and fourth natural frequencies of the
actively controlled adjacent buildings are almost the same as the =rst two natural frequencies
of the uncontrolled left building of 4.27 and 12:77 rad=s, respectively.
The =rst =ve modal damping ratios of the actively controlled adjacent buildings increase
with the increasing value of ' when ' is in the range from 10 to 6 × 105 , as shown in Figures
8(a) and 8(b). Since the =rst, third and =fth modes of vibration of the system are dominated
by the right building and the second and fourth modes of vibration are dominated by the left
building, the pattern of the curves in Figure 8(a) is diAerent from that in Figure 8(b). When
' is further increased from 2 × 105 , the second modal damping ratio starts to decrease but the
fourth modal damping ratio still increases until ' reaches 8 × 106 . The =rst, third and =fth

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 251

Figure 8. Variations of modal damping ratios with parameter '.

modal damping ratios which are dominated by the right building are always increased with
increasing value of ', and eventually the modes of vibration are overdamped. The selection of
parameter ' could have been compromised by the information on the key structural response
reductions and the modal properties as well as control forces. This study selects ' of 2 × 105
as a bene=cial value for the subsequent computation of seismic response using the derived
closed-form solution.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
252 Y. L. XU AND W. S. ZHANG

Figure 9. Variations of displacement response of adjacent buildings with height.

The further modal analysis of the actively controlled adjacent buildings with the bene=cial
weighting matrix Q and bene=cial parameter ' selected above shows that the real parts of all
eigenvalues of  are negative. Therefore, according to Lyapunov’s criterion about the stability
of linear systems, the present actively controlled adjacent buildings form a stable system.

5.3. Seismic response

To demonstrate the overall performance of the LQG controllers, the standard deviations of
displacement, shear force, acceleration responses at each Moor for each building with and
without active control devices are computed. Figures 9 and 10 shows the variations of the
standard deviation of displacement and shear force response, respectively, with the height of
the buildings. The reduction of the responses from the LQG controllers is signi=cant for all
Moors in either building. In particular, the top Moor displacement standard deviation of the
unlinked left building is 44:7 mm but with the active control devices installed, it is reduced
to 17:8 mm, leading to a 60 per cent reduction of the response. For the right building, the
top Moor displacement standard deviation is 61:1 mm for the unlinked building and 22:1 mm
for the linked building, resulting in a 64 per cent reduction. For base shear force response,
without control the bottom shear force standard deviation is 1:41 × 107 N in the left building

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 253

Figure 10. Variations of shear force response of adjacent buildings with height.

and 9:68 × 106 N in the right building. With the LQG controllers, the base shear force standard
deviation is reduced to 5:65 × 106 N in the left building and 3:57 × 106 N in the right building,
leading to a 60 per cent and a 63 per cent reduction, respectively. It may be worthwhile
to point out that the =rst modal damping ratio of either building without control is around
1 per cent. If this value becomes larger, the vibration reduction may become smaller.
The variations of acceleration response with the building height, as shown in Figure 11,
are diAerent from displacement and shear force response pro=les shown in Figures 9 and 10.
The acceleration response for each unlinked building does not vary monotonically with the
building height. This is due to the contributions from higher modes of vibration. The top Moor
acceleration standard deviation of the unlinked left building is 1:11 m=s2 but with the LQG
controllers installed, it is reduced to 0:519 m=s2 , leading to a 53 per cent reduction of the
response. For the right building, the top Moor acceleration standard deviation is 0:807 m=s2
for the unlinked building and 0:427 m=s2 for the linked building, resulting in a 47 per cent
reduction.
To further enhance the understanding of seismic response of the adjacent buildings with and
without control devices, the spectral density functions of the top Moor displacement response,
base shear force response, and top Moor acceleration response of both buildings with and
without control devices are computed and plotted in Figures 12, 13 and 14, respectively. It

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
254 Y. L. XU AND W. S. ZHANG

Figure 11. Variations of acceleration response of adjacent buildings with height.

is clearly seen that all the peaks in the response spectra of both buildings are signi=cantly
reduced when the LQG controllers are used. The eAects of higher modes of vibration on
the acceleration response are much larger than on the top displacement and base shear force
responses. From Figure 13, it is also seen that there are only very small changes of the natural
frequencies of each building after the installation of the active control devices.
The values of control forces required for the achievement of signi=cant vibration reduction
of the adjacent buildings are important for the design of the hydraulic actuators and adjacent
buildings. In terms of the closed-form solution, the control forces can be easily calculated.
Figure 15 shows the variations of control force with the building height. The maximum
control force occurs at the top of the building as expected. The fact that very small control
forces occur at the Moors near the ground indicates that the corresponding actuators may be
removed. To this end and also to demonstrate the capacity of the present formulation and
computer program, the actuators and sensors at the bottom ten Moors are removed (Case II)
and the seismic responses of both buildings are computed using the same weighting matrices.
Some major seismic responses of both buildings and the control force at the top Moor are
listed in Table I and compared with the case where the actuators and sensors are installed
at all the Moors (Case I). It is seen from Table I that while the major seismic responses
of both buildings remain almost the same for the two cases, the control force at the top

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 255

Figure 12. Spectral density of top Moor displacement with and without active control devices.

Moor in Case II, however, increases about 31 per cent compared with Case I. Thus, a proper
trade oA between control eAectiveness and control cost should be considered in the design of
controllers.

6. CONCLUSIONS

The closed-form solution for seismic response of adjacent buildings connected by hydraulic ac-
tuators with LQG controllers has been derived in this paper by combining the complex modal

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
256 Y. L. XU AND W. S. ZHANG

Figure 13. Spectral density of base shear force with and without active control devices.

superposition method with the pseudo-excitation method and the residue theorem. With the
derived closed-form solution, extensive parametric studies have been performed on the system
of many degrees of freedom to seek bene=cial weighting matrices, modal characteristics, and
maximum seismic response reductions with reasonable control forces. It was found from the
two example buildings of diAerent dynamic properties that if weighting matrices are selected
appropriately, the modal damping ratios of the system could be signi=cantly increased and
the earthquake-induced dynamic responses of both buildings could be considerably reduced.
Some other issues related to this study, such as the eAects of the non-stationary ground
motion and the three-dimensional vibration mitigation analysis including torsional eAects, need
further investigation. The dynamics of actuators and sensors and their eAects on the stability

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 257

Figure 14. Spectral density of top Moor acceleration with and without active control devices.

of the coupled buildings and the direct measurements of relative velocity and displacement
responses of the building should be addressed in further investigations towards real applica-
tions.

ACKNOWLEDGEMENTS

The writers are grateful for the =nancial support from the Hong Kong Polytechnic University through
its Area of Strategic Development Programme in Structural Vibration Control and the National Natural
Science Foundation of China under Grant NNSF-50038010.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
258 Y. L. XU AND W. S. ZHANG

Figure 15. Variations of control force of adjacent buildings with height.

Table I. Comparison of major seismic responses of both buildings and control force.
Standard deviation Installed at all Installed at top 10
Moors (Case I) Moors only (Case II)
Top Moor displacement of left building (mm) 17.8 18.1
Top Moor displacement of right building (mm) 22.1 22.2
Bottom shear force of left building (N) 5:65 × 106 5:71 × 106
Bottom shear force of right building (N) 3:57 × 106 3:74 × 106
Top Moor acceleration of left building (m=s2 ) 0.519 0.554
Top Moor acceleration of right building (m=s2 ) 0.427 0.453
Control force at top Moor (N) 8:79 × 104 1:15 × 105

REFERENCES

1. Soong TT. Active Structural Control: Theory and Practice. Longman: London, U.K. and Wiley: New York,
NY, 1990.
2. Housner GW, Bergman LA, Caughey TK, Chassiakos AG, Claus RO, Masri SF, Skelton RE, Soong TT,
Spencer BF, Yao JTP. Structural control: past, present and future. Journal of Engineering Mechanics; ASCE
1997; 123(9):897– 971.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259
 CLOSED-FORM SOLUTION FOR CONTROLLED ADJACENT BUILDINGS 259

3. Spencer BF, Sain MK. Controlling buildings: a new frontier in feedback. IEEE Control Systems Magazine
1997; 17:19 –35.
4. Yang JN, Akbarpour A, Ghaemmaghami P. New optimal control algorithms for structural control. Journal of
Engineering Mechanics; ASCE 1987; 113(9):1369 –1386.
5. Slotine JJE, Li W. Applied Nonlinear Control. Prentice-Hall: Englewood CliAs, NJ, 1991.
6. Skelton RE. Control of state and input covariance for dynamic systems. Proceedings of the 27th IEEE
Conference on Decision and Control, vol. 3, 1988; 1271–1278.
7. Chen G, Hsu S-H. Linear Stochastic Control Systems. CRC Press: Boca, Raton, FL, 1995.
8. Doyle JC, Glover K, Khargonekar P, Francis B. State-space solutions to standard H2 and H∞ control problems.
IEEE Transactions on Automatic Control 1989; 34:831– 847.
9. Gurley K, Kareem A, Bergman LA, Johnson EA, Klein RE. Coupling tall buildings for control of response
to wind. In Structural Safety & Reliability, Schueller, Shinozuko & Yao (eds). Balkema: Rotterdam, 1994;
1553 –1560.
10. Kamagata K, Miyajima K, Seto K. Optimal design of damping devices for vibration control of parallel structures.
Proceedings of the 3rd International Conference on Motion and Vibration Control, Chiba, vol. 2, 1996;
334 – 339.
11. Fukuda Y, Matsumoto Y, Seto K. Bending and torsional vibration control of Mexible structures arranged in
parallel. Proceedings of the 3rd International Conference on Motion and Vibration Control, Chiba, vol. 3,
1996; 12–17.
12. Luco JE, De Barros FCP. Optimal damping between two adjacent elastic structures. Earthquake Engineering
and Structural Dynamics 1998; 27(7):649 – 659.
13. Zhang WS, Xu YL. Dynamic characteristics and seismic response of adjacent buildings linked by discrete
dampers. Earthquake Engineering and Structural Dynamics 1999; 28(10):1163 –1185.
14. Seto K, Matsumoto Y. A structural vibration control method of Mexible buildings in response to
large earthquakes and strong winds. Proceedings of the 2nd International Workshop on Struct. Control,
Hong Kong, 1996; 490 – 496.
15. Yamada Y, Ikawa N, Yokoyama H, Tachibana E. Active control of structures using the joint member with
negative stiAness. Proceedings of the 1st World Conference on Struct. Control, Los Angeles, California,
U.S.A., vol. 2, TP2, 1994; 41– 49.
16. Christenson RE, Spencer BF, Johnson EA. Coupled building control using active and smart damping strategies.
Proceedings of the 5th International Conference on Application of Arti4cial Intelligence to Civil and Structural
Engineering. Oxford, England, September 1999.
17. Suhardjo J, Spencer BF, Sain MK. Feedback-feedforward control of structures under seismic excitation.
Structural Safety 1990; 8:69 –89.
18. Dyke SJ, Spencer BF, Quast P, Sain MK, Kaspari DC, Soong TT. Acceleration feedback control of MDOF
structures. Journal of Engineering Mechanics; ASCE 1996; 122(9):907– 918.
19. Yang JN, Li Z. Control of hysteretic system using velocity and acceleration feedbacks. Journal of Engineering
Mechanics; ASCE 1994; 120(2):418 – 418.
20. Spencer BF, Suhardjo J, Sain MK. Frequency domain optimal control strategies for aseismic protection. Journal
of Engineering Mechanics; ASCE 1994; 120(1):135–159.
21. Dyke SJ, Spencer BF, Quast P, Sain MK. Role of control-structure interaction in protective system design.
Journal of Engineering Mechanics; ASCE 1995; 121(2):322 – 338.
22. Kwakernaak H, Sivan R. Linear Optimal Control Systems. Wiley: New York, 1972.
23. Lin JH, Zhang WS, Li JJ. Structural response to arbitrarily coherent stationary random excitations. Computers
and Structures 1999; 50:629 – 633.
24. Xu YL, He Q, Ko JM. Dynamic response of damper-connected adjacent buildings under earthquake excitation.
Engineering Structures 1999; 21:135 – 148.
25. Heredia-Zavoni E, Vanmarcke EH. Seismic random-vibration analysis of multisupport-structural systems. Journal
of Engineering Mechanics; ASCE 1994; 120(5):1107–1128.

Copyright ? 2001 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:235–259