Engineering Structures 81 (2014) 208218

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Stability analysis of active vibration control of building structures using

PD/PID control
Suresh Thenozhi, Wen Yu
Departamento de Control Automtico, CINVESTAV-IPN (National Polytechnic Institute), Mexico City, Mexico

a r t i c l e

i n f o

Article history:
Received 12 August 2013
Revised 23 September 2014
Accepted 24 September 2014
Available online 17 October 2014
Keywords:
Active vibration control
Building structures
PID control
Stability

a b s t r a c t
Proportional-derivative (PD) and proportional-integral-derivative (PID) controllers are the most popular
algorithms in industrial applications. However, there are few published theory results of PD/PID controllers on the structural vibration control applications. In this paper, we analyze the stability of the active
vibration control system for both the linear and nonlinear structures. We give explicit sufcient conditions for choosing the PID gains. The theory conclusions are veried via numerical simulations and a
two-story building prototype. These results give validation of our theory analysis.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
The building structures are vulnerable to natural and manmade hazards, which may result in nancial, environmental, and
human losses. It is essential to protect these structures, including
the human occupants and non-structural components from these
threats. One approach to mitigate this undesirable behavior is to
alter the dynamic characteristics of the building with respect to a
given load, which can be achieved by adding control devices like
dampers or actuators to the building [15].
Control device and controller design are the main focus of the
traditional active vibration control systems [9,10]. Since the force
exerted by the earthquake and wind on the structures are very
huge and uncertain, these large civil structures require a large
amount of energy to control it. The structural control can be classied as passive control which does not require an external power
source [28], and active control which uses sensors and active actuators to control the unwanted vibrations [27]. There are many
active control devices designed for structural control applications
[7]. The active mass damper (AMD) is the most popular actuator,
which uses a mass without spring and dashpot [5]. In this paper,
we use AMD type actuator for the active vibration control.
In order to achieve a good performance, it is essential to design
an effective control strategy, which should be simple, robust, and
fault tolerant. Many attempts have been made to introduce
advanced controllers for the active vibration control of building
Corresponding author. Tel.: +52 55 57473734.
E-mail address: yuw@ctrl.cinvestav.mx (W. Yu).
http://dx.doi.org/10.1016/j.engstruct.2014.09.042
0141-0296/ 2014 Elsevier Ltd. All rights reserved.

structures. Instead of changing the structure stiffness, a pole-placement H1 control corresponding to a target damping ratio is proposed in [20]. In order to avoid the higher order problem in H1
control, the balanced truncation is applied in [23]. In [8], the
genetic algorithm is used to determine the feedback control. There
are several optimal control algorithms applied for the active vibration control of building structures, for example ltered linear quadratic control (LQ) [24], linear quadratic regulator (LQR) [1], and
linear quadratic Gaussian (LQG) [14]. All these controllers are
model-based, that are complex and demands the exact model of
the building structure. Some model-free controllers, such as sliding
mode control (SMC) [33], neural network control [17], and fuzzy
logic control [25] are still complex.
PID control is widely used in industrial applications. Without
model knowledge, PID control may be the best controller in realtime applications [4]. The great advantages of PID control over
the others are that they are simple and have clear physical meanings. Although theory research in PID control algorithms is well
established, it is still not well developed in structural vibration
control. In [19], a simple proportional control is applied to reduce
the building displacement due to wind excitation. In [12,13], PD
and PID controllers were used in the numerical simulations. In
[31], a Proportional-integral (PI) controller with an AMD is used
to attenuate the structural motion due to earthquake. However,
these control results are not satisfactory, because it is difcult to
tune PID gains to guarantee good performances such as rise-time,
overshoot, settling time, and steady-state error [13]. Moreover,
these works do not discuss the stability analysis of these active
control systems.

209

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

While there is no doubt about the advances in the structural

control eld, there still exist some areas which need more exploration [30] . The active devices have the ability to add force onto the
building structure. A poorly designed controller will lead to an
undesirable control performance, which can even damage the
building. So it is desired to study the stability of the closed-loop
system. Only a few structural controllers such as H1 and SMC consider the stability in their design, whereas the other control strategies do not. However, these designs have concerned only the
linear stiffness, since it represents a simple and efcient model at
least for a small operational range. In practice, these building structures possess nonlinear behavior like the hysteresis phenomenon
[6]. Also, there is a lack of experimental verication of these controllers. The practical implementation of a controller will be challenging if these issues were not addressed.
In this paper, we use standard industrial PD and PID controllers
for the active vibration control. The main contribution is that we
give theory analysis of these PD/PID controllers. Both the linear
and nonlinear cases for structural stiffness are considered in the
analysis. BoucWen model is used to model the nonlinear hysteresis phenomenon. The sufcient conditions for asymptotic stability
are derived, which are simple and explicit. The controller gains can
be decided directly from these conditions. Numerical simulations
are given to compare with SMC. An active vibration control system
for a two-story building structure equipped with an AMD is constructed for the experimental study. The experimental results via
PD and PID controllers are discussed and the effectiveness of our
theory results is demonstrated.
2. Model and active control of building structures

mx cx_ kx f e

where m is the mass, c is the damping coefcient, k is the stiffness,

_ and
f e is an external force applied to the structure, and x; x,
x are the
displacement, velocity, and acceleration, respectively.
A model for a linear multi-story structure with n-degree-offreedom (n-DOF) is shown in Fig. 1. Here it is assumed that the
mass of the structure is concentrated at each oor. Neglecting
gravity force and assuming that a horizontal force is acting on
the structure base, the equation of motion of the n-oor structure
can be expressed as [19],

mn

For unidirectional motion, the parameters can be simplied as

[19]:

m1



6
6
6
M6
6
4

0
..
.

m2
..
.


..
.

0
0
c1 c2

6
6 c2
6
6
..
C6
.
6
6
..
6
4
.
0

0
..
.
..
.

7
7
7
nn
72R ;
7
5

   mn
c2

c2 c3
..
.
..
.
0


..
.

0
..
.
..
.

0
..
.
..
.

xn

cn

cn



x 2 R ; f s f s;1 ; . . . ; f s;n 2 Rn is the structure stiffness force vector,
n
and f e 2 R is the external force vector applied to the structure,
such as earthquake and wind excitations.
If the relationship between the lateral force f s and the resulting
deformation x is linear, then f s is
n

k1 k2 k2
6
6 k2 k2 k3
6
6 .
..
f s Kx; where K 6
6 ..
.
6
..
6 ..
4 .
.
0
0

 
 
..
.

0
..
.
..
.

0
..
.
..
.

k2

m2

x n1

  kn1 kn
 
kn

kn

If the relationship between the lateral force f s and the resulting

deformation x is nonlinear, then the stiffness component is said
to be inelastic [6]. This happens when the structure is excited by
a very strong force, that deforms the structure beyond its limit of
linear elastic behavior. BoucWen model gives a realistic representation of the structural behavior under strong earthquake excitations. The force-displacement relationship of each of the stiffness
elements (ignoring any coupling effects) agrees the following relationship [32]:

f s;i ki xi 1  ki gui ;

i 1; . . . ; n

where the rst part is the elastic stiffness and the second part is the
inelastic stiffness, ki is the linear stiffness dened in Eq. (4),  and g
are positive numbers, and ui is the nonlinear restoring force which
satises

u_ i g1 dx_ i  bjx_ i jjui jp1 ui cx_ i jui jp

k1

x2

c2
m1

x1

c1
xg
Fig. 1. Mechanical model of a n-DOF building structure.

7
7
7
7
7 2 Rnn
7
7
7
kn 5

cn
m n1

7
7
7
7
7 2 Rnn ;
7
7
7
cn 5

   cn1 cn


Consider a simple building structure, which can be modeled by [6],

kn

C x_ f s f e
Mx

Fig. 2. Hysteresis loop of BoucWen model.

210

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

where d; b; c, and p are positive numbers. The BoucWen model has

hysteresis property. Its input displacement and the output force is
shown in Fig. 2. The dynamic properties of the BoucWen model
has been analyzed in [16].
In the case of closed-loop control systems, its input and output
variables may respond to a few nonlinearities. From the control
point of view, it is crucial to investigate the effects of the nonlinearities on the structural dynamics.
The BoucWen model represented in Eqs. (5) and (6) is said to
be bounded input-bounded output (BIBO) stable, if and only if the
set Xbw with initial conditions u0 is non-empty. The set Xbw is
dened as: u0 2 R such that f s is bounded for all C 1 input signal,
and x with xed values of parameters d; b; c, and p; ua and ub are
dened as

s
d
ua p
;
bc

s
d
ub p
cb

where u 2 R is the control signals applied to the dampers, d 2 Rn

is the damping and friction force vector of the dampers, and
C 2 Rnn is the location matrix of the dampers, dened as follows.

ijs

otherwise

m2

x2

c2
k1

m1

x1

c1
xg

f d;q md;q xs xd;q uq  dq

1 if

k2

Fig. 3. Building structure equipped with AMD.

C x_ f s f e Cu  d
Mx

x2 + xd

md
cd

For any bounded input signal x, the corresponding hysteresis

output f s is also bounded. On the other hand if u0 2 Xbw ;,
then the model output f s is unbounded. Table 1 shows how the
parameter d; b; c, affect the stability property of the BoucWen
model.
Passivity is the property stating that the system storage energy
is always lesser than its supply energy. On the other hand, the
active systems generate energy. In [16], it is shown that the
BoucWen model is passive with respect to its storage energy. Case
1 in Table 1 describes the physical system sufciently well and preserves both the BIBO stability and passivity properties.
The nonlinear differential equation Eq. (6) is continuos dependence on time. It is locally Lipschitz. For the case p > 1, we can conclude that Eq. (6) has a unique solution on a time interval 0; t0 .
This property will be used later during the stability analysis.
The main objective of structural control is to reduce the acceleration response of buildings to a comfortable level. In order to
attenuate the vibrations caused by the external force, an AMD is
installed on the structure, see Fig. 3. The closed-loop system with
the control force u is dened as

Ci;j

8i; j 2 f1; . . . ; ng; s # f1; . . . ; ng

where s are the oors on which the dampers are installed. In the


C1;1 C1;2
case of a two-story building C
, if the damper is
C2;1 C2;2


0 0
. If the damper is placed
placed on second oor, s f2g, C
0 1


1 0
.
on both rst and second oor, then s f1; 2g; C
0 1
The damper force f d , exerted by the q-th damper on the structure is

where md;q is the mass of the q-th damper, xs is the acceleration of

s-th oor on which the damper is installed,
xd;q is the acceleration of
q-th damper, uq is the control signal applied to the q-th damper, and



dq cd;q x_ d;q q md;q g tanh bh x_ d;q

Case

Conditions

Xbw

Upper bound of jutj

1
2
3
4
5
6

d > 0; b c > 0 and b  c P 0

d > 0; b  c < 0 and b P 0
d < 0; b  c > 0 and b c P 0
d < 0; b c < 0 and b P 0
d 0; b c > 0 and b  c P 0
All other conditions

R
ub ; ub 
R
ua ; ua 
R
;

max ju0j; ua
max ju0j; ua
max ju0j; ub
max ju0j; ub
ju0j
Unbounded

11

where cd;q and x_ d;q are the damping coefcient and velocity of the
q-th damper respectively and the second term is the Columb friction represented using a hyperbolic tangent dependent on a large
positive constant bh where q is the friction coefcient between
the q-th damper and the oor on which it is attached and g is the
gravity constant [22].
Since AMD controller adds force to the building structure, this
force may stabilize or destabilize the building structure. If the control algorithm generates unstable signal, the dampers will produce
a force which can make the building structure unstable. This is
more crucial for nonlinear devices, because even for a bounded
input signal, nonlinear devices may produce unstable output.
Obviously, the building structures in open-loop are asymptotically stable when there is no external force, f e 0. This is also true
in the case of inelastic stiffness, due to the BIBO stability and passivity properties. During excitation, the ideal active control force
required for cancelling out the vibration completely is Cu f e .
However, it is impossible because f e is not always measurable
and is much bigger than any control device force. Hence, the objective of the active control is to maintain the vibration as small as
possible by minimizing the relative movement between the structural oors. In the next section, we will discuss the simple PD and
PID controllers and their stability analysis.
2.1. PD control
PD control may be the simplest controller for the structural
vibration control system, which provides high robustness with
respect to uncertainties. PD control has the following form

md
m2

Table 1
Stability of BoucWen model with different d; b; c.

10

m1

Amplifier

x2
x1

PD/PID
Controller

xg
Fig. 4. PD/PID control for a two-story building.

211

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

u K p x  xd  K d x_  x_ d

12

where

where K p and K d are positive-denite constant matrices, which correspond to the proportional and derivative gains, respectively and
xd is the desired position. In active vibration control of building
structures, the references are xd x_ d 0, hence Eq. (12) becomes

The building structure with the PD control in Eq. (13) can be

now written as

u K p x  K d x_



C x_ f C K p x K d x_
Mx

13

The aim of the controller design is to choose the suitable gains

K p and K d in (13), such that the closed-loop system is stable. Without loss of generality, we use a two-story building structure as
shown in Fig. 4.
When the structural parameters in Eq. (8) are completely
known, i.e. there are no uncertainties and f s is linear as in Eq. (4)
then the building structure is a linear determinant system. Many
papers have used this model for the structure control design, such
as PID control [12], H2 control [23], and optimal control [1]. However, they did not discuss the stability problem.
Assuming d 0, the closed-loop system with the PD control in
Eq. (13) is

C x_ Kx f e CK p x  K d x
_
Mx

14







m1 0
c c c2
k k2 k2
where M
> 0;C 1 2
> 0;K 1
> 0;
0 m2
c2 c2
k2 k2
 






x
m1 xg
k
0
k
0
;K p p1
> 0, and K d d1
> 0. The
x 1 ;f e
m2 xg
0 kp2
x2
0 kd2


0 0
. Now we
damper is installed on the second oor, then C
0 1

f fs fe d

20

21

Since Eq. (21) is a nonlinear system and M; C, and f are

unknown, RouthHurwitz stability criterion in Eq. (16) cannot
be applied here. The following theorem gives the stability analysis of the PD control in Eq. (13). In order to simplify the proof,
we rst consider Cnn Inn , i.e., each oor has an actuator
installed on it.
Theorem 1. Consider the structural system as Eq. (19) controlled by
the PD controller as Eq. (13), the closed-loop system as Eq. (21) is
stable, provided that the control gains satisfy1

K p > 0;

Kd > 0

22

The derivative of the regulation error x converges to the residual set

n
o
_ kx_ k2Q 6 l
f
Dx_ xj

23

 f P f T K1
where l
f f and C > Kf > 0.
Proof. We select the systems energy as the Lyapunov candidate V.

1 T
1
x_ Mx_ xT K p x
2
2

are in a position to study the system represented in (14) using linear

techniques. Eq. (14) can be written in the state-space form

The rst term of Eq. (24) represents the kinetic energy and the
second term is the virtual elastic potential energy. Since M and K p
are positive denite matrices, V P 0. The derivative of Eq. (24) is

z_ Acl z f cl

15

 


0
I22
x
 22

where z _ 2 R4 ; Acl
2 R44 ,
1
1
M K CK p M C CK d
h x
iT
4
T
and f cl 012 f e 2 R .
The stability of the closed-loop system in Eq. (15) depends on
the system matrix Acl . Its characteristic polynomial is

detsI  Acl s4 a1 s3 a2 s2 a3 s a4

16

1
1
c 1 c 2
c2 kd2
m1
m2


1
c1 kd2 c2 kd2 m1 kp2 c1 c2 k1 m2 k2 m1 k2 m2
a2
m1 m2

1 
k1 kd2 k2 kd2 c1 kp2 c2 kp2 c1 k2 c2 k1
a3
m1 m2

1 
k1 kp2 k2 kp2 k1 k2
a4
m1 m2
17

a1

Using LienardChipart criterion [21], the closed-loop system Acl is

stable if and only if

18

Now the designer can directly choose the controller gains,

which can satisfy the ve inequalities given by Eq. (18).
In practice, the parameters of the building structure are partly
known and the structure model might have nonlinearities such
as the hysteresis phenomenon. It is convenient to express Eq.
(8) as

C x_ f Cu
Mx



x_ T K p x x_ T C x_  f  K p x  K d x_ x_ T K p x
V_ x_ T M x
x_ T C K d x_  x_ T f

25

Using the matrix inequality

T

X Y Y T X 6 X T KX Y T K1 Y

26
nm

nn

which is valid for any X; Y 2 R

and any 0 < K K 2 R
can write the scalar variable x_ T f as

where

ai > 0; i 1; 2; 3; 4 and a1 a2 a3  a21 a4  a23 > 0

24

19

x_ T f

1 T
1 T
T
x_ f f x_ 6 x_ T Kf x_ f K1
f f
2
2

, we

27

where Kf is any positive denite matrix. In this paper, we select Kf

as

C > Kf > 0

28

So
T
V_ 6 x_ T C K d  Kf x_ f K1
f f

29

If we choose K d > 0, then

 f 6 km Q kx_ k2 f T K1

V_ 6 x_ T Q x_ l
f f

30

where Q K d C  Kf > 0. V is therefore an ISS-Lyapunov function.

T

Using Theorem 1 from [26], the boundedness of f K1
f f 6 lf implies
that the regulation error kx_ k is bounded. It is noted that when
2
f;
kx_ kQ > l

8t 2 0; T 

V_ < 0. Now we prove that the

nite. Let T k denotes the time

31
2
total time during which kx_ kQ >  f
2
interval during which kx_ kQ >  f .

l is

1
K > 0 means K is a positive denite matrix, i.e., with any vector x; xT Kx > 0, all of
its eigenvalues are positive.

212

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

2
 f stay outside the circle of
(1) If only nite times that kx_ kQ > l
 f (and then reenter), kx_ k2Q > l
 f will eventually stay
radius l
inside of this circle.
2
 f leave the circle innite times, since the total
(2) If kx_ kQ > l
2
_
 f leave the circle is nite, then
time kxkQ > l

1
X
T k < 1;

lim T k 0

k!1

k1

32

2
So kx_ kQ is bounded via an invariant set argument. From Eq.
2
(30) kx_ k is also bounded. Let kx_ kQ denotes the largest track-

ing error during the T k interval. Then Eq. (32) and bounded
2
kx_ kQ imply that

h
i
2
f 0
lim kx_ kQ  l

k!1

33

2
 f , hence Eq. (23) is achieved.
So kx_ kQ will converge to l

2
 f . Total time of
Since V P 0; V decreases until kx_ kQ 6 l
2
1
T
 f being nite means that V 2 x_ M x_ 12 xT K p x is bounded,
kx_ kQ > l
hence the regulation error x_ is bounded. h

The nal closed-loop equation becomes

C x_ f K p x  K d x_  n f 0
Mx

In order to analyze the stability of Eq. (40), we rst give the following properties.
P1. The positive denite matrix M satises the following
condition.


0 < km M 6 kMk 6 kM M 6 m

41

where km M and kM M are the minimum and maximum eigen > 0 is the upper bound.
values of the matrix M, respectively and m
~
P2. The term f is Lipschitz over ~
x and y

~k 6 kf k~x  y
~k
kf ~x  f y

42

Most of the uncertainties are rst-order continuous functions.

Since f s ; f e , and d are rst-order continuous functions and satisfy
Lipschitz condition, P2 can be established using Eq. (20). Now we
R
calculate the lower bound of f dx.

fdx

It is well known that the regulation error becomes smaller while

increasing the gain K d . The cost of large K d is that the transient performance becomes slow. Only when K d ! 1, the regulation error
converges to zero [18]. However, it would seem better to use a
smaller K d if the system contains high-frequency noise signals.

40

n K ix

f s dx

f e dx

d dx

43

Rt

Rt


We dene the lower bound of 0 f s dx is f s and for 0 d dx is d.
Compared with f s and d; f e is much bigger in the case of earthRt
quake. We dene the lower bound of 0 f e dx is f e . Finally, the
lower bound kf is


kf f s  f e  d

44

2.2. PID control

From the above section it is clear that any increase in the derivative gain K d can decrease the regulation error, but causes a slow
response. In the control viewpoint, the regulation error can be
removed by introducing an integral component to the PD control,
i.e., modify the PD control into PID control. The PID control law
can be expressed as

u K p x  xd  K i

x  xd ds  K d x_  x_ d

34

where K i > 0 correspond to the integration gain. For the regulation

case xd x_ d 0, Eq. (34) becomes

u K p x  K i

Theorem 2. Consider the structural system as Eq. (19) controlled by

the PID controller as Eq. (36), the closed-loop system as Eq. (40) is
T
asymptotically stable at the equilibrium n  f 0; x; x_  0, provided
that the control gains satisfy

t
0

The following theorem gives the stability analysis of PID controller (36).

xds  K d x_

35


3
km K p P kf kc 
2 
km K p
kM K i 6 /
kM M


kc
 km C
km K d P / 1
kM M
q

1
k Mkm K p .
3 m

In order to analyze the stability of PID controller, Eq. (35) is

expressed by

where /

u K p x  K d x_  n

Proof. Here, the Lyapunov function is dened as

n K i x;

n0 0

36

Now substituting Eq. (36) into Eq. (19), the closed-loop system
can be written as

C x_ f K p x  K d x_  n
Mx

1 T
1
a
a T
T
T
_
x_ Mx_ xT K p x nT K 1
x Kdx
i n x n ax M x
2
2
2
2
Z t

fdx  kf

46

where kf is dened in Eq. (46) such that V 0 0. In order to show

P
that V P 0, it is separated into three parts, such that V 3i1 V i

38

The equilibrium of Eq. (38) is n; x; x_  n ; 0; 0. Since at equilibrium point x 0 and x_ 0, the equilibrium is f 0; 0; 0. In order to
move the equilibrium to origin, we dene

n n  f 0

37

In matrix form, the closed-loop system is

3
2 3 2
Kix
n
d6 7 6
7
x_
5
4x5 4


dt
1
M C x_ f K p x K d x_ n
x_

45

39

V1

1 T
a
x K p x xT K d x
6
2

fdx  kf P 0

47

1 T
a
T
x K p x nT K 1
i nx n
6
2
11
akm K 1
i
km K p kxk2
P
knk2  kxkknk
26
2

V2

When a P k

1
m K i km K p

48

213

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

1
V2 P
2

s !2
r
km K p
3
knk P 0
k xk 
3
km K p

49

and

V3

1 T
1
x K p x x_ T Mx_ axT M x_
6
2

50

~T A~x P ky
~kkA~xk P ky
~kkAkk~xk P jkM Ajky
~kk~xk
y
when a 6

51

p
1

k Mkm K p
3 m
kM M


1 1
2
V3 P
km K p kxk2 km Mkx_ k 2akM Mkxkkx_ k
2 3
!2
r
p
1
km K p
kxk km Mkx_ k P 0

2
3

52

If

r
3
1
1
2
2
km K 1
i km K p km M P kM M
3

53

there exists

q
1
k Mkm K p
3 m
kM M

PaP

54

km K 1
i km K p

The derivative of Eq. (46) is

T
T
x_ T K p x anT K 1
_T
_T _

V_ x_ T Mx
i n x n x n ax M x ax M x

ax_ T K d x x_ T


_T
f x_ T C x_  f  K p x  K d x_  n f 0 x_ T K p x anT K 1
i nx n


xT n ax_ T Mx_ axT C x_  f  K p x  K d x_  n f 0
axT K d x_ x_ T f

55

fn 2 Rn ; x 0 2 Rn ; x_ 0 2 Rn g

ax f 0  f  6 akf kxk

56

Using Eq. (26) we can write



axT C x_ 6 akc xT x x_ T x_

57

where kC k 6 kc .
T
T
T
Since n K i x; nT K 1
i n becomes ax n and x n becomes x K i x,
then



V_ x_ T C K d  aM  akc x_  xT aK p  K i  akf  akc x

58

Using Eqs. (41) and (58) becomes,

V_ 6 x_ T km C km K d  akM M  akc x_



 xT akm K p  kM K i  akf  akc x

59


Remark 1. Since the stiffness of the building structure has hysteresis property, the hysteresis output depends on both the instantaneous and the history of the deformation. This deformation before
applying the force (loading) and after removing the force (unloading) is not the same, i.e, the equilibrium position before the earthquake and after the vibration dies out is not the same. After the
earthquake, the stable point is moved. This corresponds to the
term f 0. So we cannot conclude that the closed-loop system is
globally stable.
It is well known that, in the absence of the uncertainties and
external force, f 0, the PD control as Eq. (13) with any positive
gains can drive the closed-loop system asymptotically stable. The
main objective of the integral action can be regarded to cancel f.
In order to decrease integral gain, an estimated f is applied to
the PID control as Eq. (36). The PID control with an approximate
force compensation ^f is

kM K i 6 3/
2

62

~ k
k
c
~
f
;k
f
kM M

 kf .

If the number of dampers installed on the buildings is less than

the number of the building oors (n), then the resulting system is
termed as under-actuated system. In that case, the location matrix
C should be included along with the gain matrices. In our experiment, there is only one damper installed (second oor) on the
structure. The PID controller becomes

Cu

( 
  
x1
kp1 0
ki1


0 kp2 x2
0 1
0

 
kd1 0
x_ 1

0 kd2 x_ 2
0 0

Cu
60

then Eq. (45) is established.

Finally, we prove the asymptotic stability of the closed-loop
system as Eq. (40). There exists a ball R of radius q > 0 centered at
the origin of the state-space on which V_  0. The origin of the
closed-loop equation as Eq. (40) is a stable equilibrium. Since the
closed-loop equation is autonomous, we use La Salles theorem.
Dene X as

n Kix

The above theorem is also applicable for the PID controllers

with an approximate f compensation as in Eq. (62). The condition
h
i

~ kc and
for PID gains in Eq. (45) becomes km K p P 32 k
f

If km C km K d P akM M kc  and km K p P a kM K i kf kc ,


kM 1K i , if
then V_  0; kxk decreases. From (54) and km K 1
i

r


1
kc
 km C
km K d P
km Mkm K p 1
3
kM M

3
km K p P kf kc 
2

61

zt to
From Eq. (55), V_ 0 if and only if x x_ 0. For a solution 
belong to X for all t P 0, it is necessary and sufcient that
0 for
x x_ 0 for all t P 0. Therefore, it must also hold that x
all t P 0. We conclude that from the closed-loop system as Eq.
(40), if 
zt 2 X for all t P 0, then f x f 0 n f 0 and n_ 0.
zt 0 is the only initial conIt implies that n 0 for all t P 0. So 
zt 2 X for all t P 0. We conclude from the
dition in X for which 
above discussions that the origin of the closed-loop system as Eq.
(40) is asymptotically stable. It establishes the stability of the proposed controller, in the sense that the domain of attraction can be
enlarged with a suitable choice of the gains. Namely, increasing
K p the basin of attraction will grow. h

u K p x  K d x_  n ^f;

From Eq. (42)

T

n
o

T
X zt xT ; x_ T ; nT 2 R3n : V_ 0

kp2 x2  ki2

0
Rt
0

0
ki2

" R t
0

x1 d s

x2 d s

Rt

#
63


x2 ds  kd2 x_ 2

64

where the scalars kp2 ; ki2 , and kd2 are the proportional, integral, and
derivative gains, respectively. In this case Eq. (45) becomes,

3
kf kc 
2
~ minfkp2 g
ki2 6 /
kM M


~ 1 kc
 km C
kd2 P /
kM M
kp2 P

65

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

q
~ 1 km M minfkp2 g.
where /
3
Remark 2. The PID tuning methods are different for the system
with and without prior knowledge. If the system parameters are
unknown, then auto-tuning techniques are employed to choose the
gains either on-line or off-line. These techniques are broadly
classied into direct and indirect methods [3]. In direct method,
the closed-loop response of the system is observed and the
controller gains are tuned directly based on the past experience
and heuristic rules. In the case of indirect method, the structure
parameters are identied rst from the measured output and
based on these identied parameters the controller is then tuned
to achieve a desired system dynamics. This paper provides a tuning
method that ensures a stable closed-loop performance. For that
purpose, the structural parameters kM M , km C; kf , and kc are
determined from the identied parameters.
Remark 3. The PID control as Eq. (34) does not need exact information about the building structure as Eq. (8). It uses only the displacements of the building and upper bound estimation of the
building parameters. If the actual control force to the building
structure satises Eq. (45), the closed-loop system is stable. And
this condition is easy to be satised from the above remark. So
we does not require the theory force (34) to match actual control
force for PID structure control. However, in many cases the actual
control force cannot reach the theory force as Eq. (34) due to the
actuator limitations, which causes saturation.

ureal sat utheory

utheory

mmax



if utheory  < mmax


if utheory  P mmax

easy to verify that the closed-loop system with PID control

(Acl 2 R55 ; kp2 350; ki2 2200, and kd2 45 is stable.
If the SMC switching gain g is greater than the system uncertainty bound, then the r z converges to zero. We consider the
switching gain of SMC g 1:3.
Here the structure is excited by a step input and the corresponding vibration response is reduced by applying the above controllers. The control objective is to bring the structural vibration as
close to zero as possible. The control signal is directly applied as
the force without applying any constraint by neglecting the damper dynamics.
Fig. 5 shows the time response of the second oor displacement
for both controlled and uncontrolled cases, the unite is centimetre.
It shows that all the three controller reduces the structure motion.
The PD controller reduces the structure oscillations but has a big
steady-state error. This error can be reduced by introducing an
integral term, hence a PID controller, which can achieve a zero
steady-state error. The control signals are shown in Figs. 68, the
unites are volt.
The performance of the SMC lies between the PD and PID controller but its control signal has many high frequency switching,
which may not be acceptable for some mechanical dampers. However in practice, the active system cannot achieve this much attenuation due to the actuator limitations.

2
No control

66

where utheory is the theory force, ureal is the actual control force, mmax
is the maximum torque of the AMD actuator. Now the linear PID
controller becomes nonlinear PID. The asymptotic stability of Theorem 2 becomes stable as Theorem 1, see [2].

Displacement

214

PD
PID

1.5

SMC

1
0.5
0
-0.5

10

15

20

3. Simulations and experimental results

30

35

40

45

50

Fig. 5. The displacements of the second oor using PD, PID, and SMC control.

Consider the system described by Eq. (2) with linear stiffness, has
the following set of parameters: the matrix M is m1 3:3 kg and
m2 6:1 kg; C is given by c1 2:5 N s=m and c2 1:4 N s=m, and
K is given by k1 4080 N=m and k2 4260 N=m. These parameters
are obtained by identifying the two-story lab prototype [11].
We compare the performances of PD, PID, and SMC. Like PID
controller, SMC is a popular robust controller which is often seen
in the structural vibration control applications [30]. A switching
control law is used to drive the state trajectory onto a pre-specied
surface. In the case of structural vibration control, this surface corresponds to a desired system dynamics.
A general class of discontinuous structural control is dened by
the following relationships [34].

Control signal

3.1. Numerical simulations

-0.5

-1

-1.5

10

15

20

25

30

35

40

45

50

Time (s)
Fig. 6. Control signal of PD control for the simulation model.

g>0

67

where the linear term ueq is the equivalent control force, r is the
T
sliding surface and sign r signr1 ; . . . ; signr2n  . The sliding
surface
can
be a function of the regulation error, then


r xT ; x_ T T z 2 R2n . The equivalent control can be estimated
using a low pass lter or neglected, if the system parameters are
unknown.
If the PD control as Eq. (13) has kp2 350 and kd2 45, they satisfy the condition as Eq. (18), hence the closed-loop is stable. It is

Control signal

8
>
< g if r > 0
u ueq  gsignr
0 if r 0 ;
>
:
g if r < 0

25

Time (s)

-0.5

-1

-1.5

10

15

20

25

30

35

40

45

Time (s)
Fig. 7. Control signal of PID control for the simulation model.

50

215

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

3.2. Experimental results

x 10-4

1.5
1
0.5
0
-0.5
-1
-1.5

Estimated
Measured

Magnitude

1.5

0.5

10

15

20

25

30

35

40

45

50

10

15

Fig. 10. Comparison of the measured and estimated position data using Fourier
spectra.

AMD

Accelerometer

Data
Acquisition
Unit
Hydraulic
Shaker
Fig. 11. Two-story building prototype with the shaking table.

3
Uncontrolled
PD

2
1
0
-1
-2
-3
0

Frequency (Hz)

Displacement (cm)

Control signal

To illustrate the theory analysis results, a two-story building

prototype is constructed which is mounted on a shaking table,
see Fig. 11. The building structure is constructed of aluminum.
The shaking table is actuated using a hydraulic control system
(FEEDBACK EHS 160), which is used to generate earthquake signals. The AMD is a linear servo actuator (STB1108, Copley Controls
Corp.), which is mounted on the second oor. The moving mass of
the damper weights 5% (0:45 kg) of the total building mass. The
linear servo mechanism is driven by a digital servo drive (Accelnet
Micro Panel, Copley Controls Corp). ServoToGo II I/O board is used
for the data acquisition purpose.
The PD/PID control needs the structure position and velocity
data. During the seismic excitation, the reference where the displacement and velocity sensors are attached will also move, as a
result the absolute value of the above parameters cannot be
sensed. Alternatively, accelerometers can provide inexpensive
and reliable measurement of the acceleration at strategic points
on the structure. Three accelerometers (Summit Instruments
13203B) were used to measure the absolute accelerations on the
ground and each oor. The ground acceleration is then subtracted
from the each oor accelerations to get the relative oor movement. The relative velocity and position data are then estimated
using the numerical integrator proposed in [29].
The position estimation with respect to the Loma Prieta EastWest earthquake signal is shown in Fig. 9. The effect of the proposed numerical integrator on frequency characteristics is studied
by plotting Fourier spectra. A sinusoidal signal composed with
6 Hz; 7 Hz, and 8 Hz is used here to excite the linear actuator.
The linear actuator has a position and acceleration sensor. The
acceleration of the actuator is measured, which is then integrated
twice to obtain the position estimation. The FFT diagram of the
measured and estimated position is generated, see Fig. 10. As can
be seen from the gure that the frequency information is not
affected except in the low frequency range. This low-frequency
error is caused due to the presence of bias and noise in the accelerometer output.
The control programs were operated in Windows XP with
Matlab 6.5/Simulink. All the control actions were employed at a

10

15

20

25

30

Time (s)

Time (s)
Fig. 12. The displacements of the rst oor using PD control.
Fig. 8. Control signal of SMC control for the simulation model.

Displacement (mm)

15

Measured
Estimated

10

0
-2

24.8

25

25.2 25.4 25.6 25.8

26

0
-5
-10
-15

10

15

20

25

Time (sec)
Fig. 9. Comparison of the measured and estimated position data.

30

sampling frequency of 1:0 kHz. The control signal generated by

the control algorithm is fed as voltage input to the amplier. The
current control loop is used to control the AMD operation. The
amplier converts its voltage input to a respective current output
with a gain of 0.5. The AMD have a force constant of 6:26 N=A or
3:13 N=V.
Now, we show the procedure for selecting the gains for a stable
operation. The theorems in this paper give sufcient conditions for
the minimal values of the proportional and derivative gains and
maximal values of the integral gains. In order to do a fair comparison, both the PD and PID controller uses the same proportional
and derivative gains. We rst design the PID controller based
on the identied parameters of the two-story lab prototype.

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

Uncontrolled
PID

2
1
0
-1
-2
-3

10

15

20

25

30

Time (s)

Displacement (cm)

Fig. 13. The displacements of the rst oor using PID control.

Uncontrolled
PD

3
1
-1
-3
-5

10

15

20

25

30

Time (s)
Fig. 14. The displacements of the second oor using PD control.

Displacement (cm)

5
Uncontrolled
PID

3
1

achieved using the controllers. Figs. 1215 show the time response
of the rst and second oor displacements for both controlled and
uncontrolled cases. The control algorithm outputs are shown in
Figs. 16 and 17.
From Table 2 one can observe that the controllers effectively
decrease the vibration. The controlled response using the PD controller is reduced signicantly by applying a damping provided
by the derivative gain. Figs. 13 and 15 show the vibration attenuation achieved by adding an integral action to the above PD controller. The results demonstrate that PID controller performs
better than PD controller.
Remark 4. It is worth to note the frequency characteristics of an
integrator. An ideal integrator acts like a low-pass lter. The bode
magnitude plot of an ideal integrator is shown in Fig. 18. At 1.6 Hz
the integrator attenuates the input power by 20 dB and at 16 Hz it
reaches to 40 dB. During earthquake the structure oscillates at its
natural frequencies. If the natural frequency is very small then the
integrator produces a larger output. The structure prototype we
used for the experiments have natural frequencies 2:1 Hz and
8:9 Hz. Since these frequencies have an attenuation more than
20 dB a larger value can be used for K i . On the other hand, if the
building has a natural frequency less than 1:6 Hz, then the integral
gain should be reduced accordingly. The error input to the
integrator is the position data. From Figs. 1215 we can see that
the position data for the most part takes successive positive and
negative values. Hence, the integrator output for high frequency
input signal is small due to the rapid cancellation between these
positive and negative values.

-1
-3
-5

Table 2
Comparison of vibration attenuation obtained using PD and PID controller.

10

15

20

25

30

Time (s)
Fig. 15. The displacements of the second oor using PID control.

The following set of parameters were used for the control design:
kM M 6:1; km C 0:6; kf 365, and kc 5:8. Applying these
values in Theorem 2 we get


km K p P 556;

kM K i  3066;

km K d P 65

68

In order to evaluate the performance, these controllers are

implemented to control the vibration on the excited lab prototype.
The control performance is evaluated in terms of their ability to
reduce the relative displacement of each oor of the building.
The proportional, derivative, and integral gains are further
adjusted to obtain a higher attenuation. Finally, the PID controller
gains are chosen to be

kp 635;

ki 3000;

kd 65

Control action

PD control

PID control

No control

Floor-1 displacement
Floor-2 displacement

0.1699
0.5141

0.1281
0.3386

1.0688
3.3051

1.5

Control signal (V)

Displacement (cm)

216

1
0.5
0
-0.5
-1
-1.5

10

15

20

25

30

Time (s)
Fig. 16. Control signal of PD control for the prototype.

69

and the PD controller gains are

kd 65

70

Theorem 1 requires that the PD controller gains need to be positive. In the experiments, the negative gains resulted in an unstable
closed-loop operation, which satises the conditions in Theorem 1.
Since Theorem 2 provides sufcient conditions, violating it does
not mean instability. We have found that, when kM K i is more
than 4200, the system becomes unstable. This satises the condition as Eq. (68).
P
Table-2 shows the mean squared error, MSE N1 Ni1 e2i of the
displacement with proposed controllers, here N is the number of
data samples and e xd  x x, where x is the position

Control signal (V)

kp 635;

1
0.5
0
-0.5
-1
0

10

15

20

25

Time (s)
Fig. 17. Control signal of PID control for the prototype.

30

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

4. Conclusion

20

Magnitude (dB)

217

10
0
-10
-20
-30
-40
0

10

12

14

16

18

20

Frequency (Hz)
Fig. 18. Bode magnitude plot of an ideal integrator.

In this paper, the model of building structures with an active

vibration control is analyzed. The theory contribution of this paper
is that the stability of the AMD PD/PID control for building structures is proven. By using Lyapunov theory, sufcient conditions
of stability are derived to tune the PD/PID gains. The technical
advance of this paper is that a systematic tuning method of PID
is proposed based on the stability analysis. The theory results are
successfully applied to a numerical example and a two-story building prototype. The results show that even though the chosen gains
are not optimal, the controllers guarantee stable control
performances.
Acknowledgments
The authors are grateful to Dr. Antonio Concha for his help in
identifying the building parameters. The authors would like to
thank Mr. Jess Meza and Mr. Gerardo Castro for their assistance
to complete the experiments. The rst author would like to thank
Consejo Nacional de Ciencia y Tecnologa (CONACyT) of Mexico for
the nancial support. This work was partially supported by State
Key Laboratory of Synthetical Automation for Process Industries
and the Project 111 (No. B08015) of China.

Displacement (cm)

Fig. 19. Fourier spectrums of PD and PID control signals.

References

Uncontrolled
SMC

3
1
-1
-3
-5

10

15

20

25

30

Time (s)

Control signal (V)

Fig. 20. The displacements of the second oor using SMC control.

1
0.5
0
-0.5
-1
0

10

15

20

25

30

Time (s)
Fig. 21. Control signal of SMC control for the prototype.

Sometimes, the integral control results in an actuator saturation. But as discussed in Remark 2, the output of the integrator is
small in our case. Fig. 19 shows the magnitude spectrum of control
signals of the PD and PID controllers. As the building structure is
excited mainly in its natural frequency (2:1 Hz), the major control
action occurs in this zone. Even though the K i gain is large, PID controller produces less control effort than the PD controller, but still
achieves a better vibration attenuation.
Remark 5. From our experience, the classic SMC performs poor
while x starts damping from a large to a small value for the shaking
table. In Fig. 20, after 22 s we can see that the vibration level
increases. This is due to the fact that SMC switches aggressively
with a gain of g, even though the actual vibration is considerably
small, see Fig. 21.

[1] Alavinasab A, Moharrami H. Active control of structures using energy-based

LQR method. Comput-Aided Civ Inf Eng 2006;21:60511.
[2] Alvarez-Ramirez J, Kelly R, Cervantes I. Semiglobal stability of saturated linear
PID control for robot manipulators. Automatica 2003;39:98995.
[3] strm KJ, Hgglund T, Hang CC, Ho WK. Automatic tuning and adaptation for
PID controllers-a survey. Control Eng Pract 1993;1:699714.
[4] strm KJ, Hgglund T. Revisiting the ZieglerNichols step response method
for PID control. J Process Control 2004;14:63550.
[5] Chang JCH, Soong TT. Structural control using active tuned mass damper. J Eng
Mech ASCE 1980;106:10918.
[6] Chopra AK. Dynamics of structures: theory and application to earthquake
engineering. 2nd ed. Prentice Hall; 2001.
[7] Datta TK. A state-of-the-art review on active control of structures. ISET J Earthq
Technol 2003;40:117.
[8] Du H, Zhang N. H1 control for buildings with time delay in control via linear
matrix inequalities and genetic algorithms. Eng Struct 2008;30:8192.
[9] Fisco NR, Adeli H. Smart structures: part Iactive and semi-active control.
Scientia Iran 2011;18:27584.
[10] Fisco NR, Adeli H. Smart structures: part IIhybrid control systems and control
strategies. Scientia Iran 2011;18:28595.
[11] Garrido-Moctezuma RA, Concha SA. Estimation of the parameters of structures
using acceleration measurements. In: 16th IFAC symposium on system
identication, vol. 16. Brussels, Belgium; 2012. p. 179196.
[12] Guclu R, Yazici H. Vibration control of a structure with ATMD against
earthquake using fuzzy logic controllers. J Sound Vib 2008;318:3649.
[13] Guclu R. Sliding mode and PID control of a structural system against
earthquake. Math Comput Modell 2006;44:2107.
[14] Ho CC, Ma CK. Active vibration control of structural systems by a combination
of the linear quadratic Gaussian and input estimation approaches. J Sound Vib
2007;301:42949.
[15] Housner GW et al. Present and future. J Eng Mech 1997;123:897974.
[16] Ikhouane F, Maosa V, Rodellar J. Dynamic properties of the hysteretic Bouc
Wen model. Syst Control Lett 2007;56:197205.
[17] Kim JT, Jung HJ, Lee IW. Optimal structural control using neural networks. J
Eng Mech 2000;126:2015.
[18] Lewis FL, Dawson DM, Abdallah CT. Robot manipulator control: theory and
practice. 2nd ed. Marcel Dekker, Inc; 2004.
[19] Nerves AC, Krishnan R. Active control strategies for tall civil structures. Proc
IEEE Int Conf Ind Electron Control Instrum 1995;2:9627.
[20] Park W, Park KS, Koh HM. Active control of large structures using a bilinear
pole-shifting transform with H1 control method. Eng Struct 2008;30:
333644.
[21] Poznyak AS. Advanced mathematical tools for automatic control engineers.
Deterministic systems, Vol. I. Springer; 2009.
[22] Roldn C, Campa FJ, Altuzarra O, Amezua E. Automatic identication of the
inertia and friction of an electromechanical actuator. New advances in
mechanisms, transmissions and applications, vol. 17. Netherlands: Springer;
2014. p. 40916.
[23] Saragih R. Designing active vibration control with minimum order for exible
structures. IEEE Int Conf Control Autom 2010:4503.

218

S. Thenozhi, W. Yu / Engineering Structures 81 (2014) 208218

[24] Seto K. A structural control method of the vibration of exible buildings in

response to large earthquake and strong winds. In: Proceedings of the 35th
conference on decision and control, vol. 1. 1996. p. 65863.
[25] Shook DA, Roschke PN, Lin PY, Loh CH. GA-optimized fuzzy logic control of a
large-scale building for seismic loads. Eng Struct 2008;30:43649.
[26] Sontag ED, Wang Y. On characterizations of the input-to-state stability
property. Syst Control Lett 1995;24:3519.
[27] Spencer BF, Sain MK. Controlling buildings: a new frontier in feedback. IEEE
Control Syst Mag Emerging Technol 1997;17:1935.
[28] Symans MD, Constantinou MC. Semi-active control systems for seismic
protection of structures: a state-of-the-art review. Eng Struct
1999;21:46987.
[29] Thenozhi S, Yu W, Garrido R. A novel numerical integrator for velocity and
position estimation. Trans Inst Meas Control 2013;35:82433.

[30] Thenozhi S, Yu W. Advances in modeling and vibration control of building

structures. Annu Rev Control 2013;37:34664.
[31] Tinkir M, Kalyoncu M, S
ahin Y. Deection control of two-oors structure
against northridge earthquake by using PI controlled active mass damping.
Appl Mech Mater 2013;307:12630.
[32] Wen JK. Method for random vibration of hysteretic systems. J Eng Mech
Division (ASCE) 1976;102:24963.
[33] Yang JN, Wu JC, Agrawal AK, Hsu SY. Sliding mode control with compensator
for wind and seismic response control. Earthquake Eng Struct Dynam
1997;26:113756.
[34] Yu X, Kaynak O. Sliding-mode control with soft computing: a survey. IEEE
Trans Industr Electron 2009;56:327585.