ARTICLE IN PRESS

Mechanical Systems and Signal Processing 23 (2009) 300–315

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/jnlabr/ymssp

Semiactive nonlinear control of a building with a magnetorheological

damper system
Yeesock Kim a, Reza Langari b, Stefan Hurlebaus a,
a
Zachry Department of Civil Engineering, Texas A&M University, College Station, TX 77843-3136, USA
b
Department of Mechanical Engineering, Texas A&M University, College Station, TX 77843-3135, USA

a r t i c l e i n f o abstract

Article history: This paper proposes a linear matrix inequality (LMI)-based systematic design
Received 20 September 2007 methodology for nonlinear control of building structures equipped with a magnetor-
Received in revised form heological (MR) damper. This approach considers stability performance as well as
10 June 2008
transient characteristics in a unified framework. First, multiple Lyapunov-based
Accepted 15 June 2008
Available online 26 June 2008
controllers are designed via LMIs such that global asymptotical stability of the building
structure is guaranteed and the performance on transient responses is also satisfied.
Keywords: Such Lyapunov-based state feedback controllers are converted into output feedback
Linear matrix inequalities (LMIs) regulators using a set of Kalman estimators. Then, these Lyapunov-based controllers and
Nonlinear control
Kalman observers are integrated into a global nonlinear control system via fuzzy logic.
Multiple Lyapunov-based controllers
To demonstrate the effectiveness of the proposed approach, a three-story building
Magnetorheological (MR) damper
structure employing an MR damper is studied. The performance of the nonlinear control
system is compared with that of a traditional linear optimal controller, i.e., H2/linear
quadratic Gaussian (LQG), while the uncontrolled system response is used as the
baseline. It is demonstrated from comparison of the uncontrolled and semiactive
controlled responses that the proposed nonlinear control system design framework is
effective in reducing the vibration of a seismically excited building structure equipped
with an MR damper. Furthermore, the newly developed controller is more effective in
mitigating responses of the structure than the H2/LQG controller.
& 2008 Elsevier Ltd. All rights reserved.

1. Introduction

In recent years, fuzzy logic has attracted great attention to control system design [1–6]. A number of design
methodologies for fuzzy logic controllers have been successively applied to a variety of large-scale civil building structures.
They include trial-and-errors-based methodologies [7–12]; a self-organizing approach [13]; training using linear quadratic
Gaussian (LQG) data [14]; neural networks-based learning [15–18]; adaptive fuzzy [19]; genetic algorithms-based training
[20–24]; fuzzy sliding mode [25–27]; etc. However, no study on systematic design framework has been conducted to
design a semiactive nonlinear fuzzy controller (SNFC) for building structures equipped with a nonlinear semiactive control
device.
From a practical point of view, research related to a systematic design framework is still required for a large building
control system for mitigation of natural hazards, e.g., earthquake or strong wind. An active nonlinear fuzzy control (ANFC)
system design can be carried out in a systematic way via the so called parallel distributed compensation (PDC) approach,

 Corresponding author. Tel.: +1 979 845 9570; fax: +1 979 845 6554.
E-mail addresses: controlga@tamu.edu (Y. Kim), rlangari@tamu.edu (R. Langari), shurlebaus@civil.tamu.edu (S. Hurlebaus).

0888-3270/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2008.06.006
 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 301

which employs multiple linear controllers [1,28]. Those controllers correspond to local linear structural models with
automatic scheduling performed through fuzzy rules. Tanaka and Sano [29] proposed a theorem on the stability analysis of
an ANFC system using Lyapunov direct method. This theorem states sufficient conditions for an ANFC system to be globally
asymptotically stable by finding a common symmetric positive definite matrix such that a set of simultaneous Lyapunov
inequalities are satisfied.
However, based on the PDC approach, no systematic design framework has been investigated to design an SNFC system
for vibration control of building structures equipped with a nonlinear semiactive device subjected to destructive
environmental forces such as earthquake and strong wind loadings. First, a mathematical model of a magnetorheological
(MR) damper and its properties are described. Then, a systematic design framework of nonlinear control system is
presented, following by a numerical simulation.

2. Magnetorheological damper

In recent years, smart structures have been adopted from many engineering fields because the performance of structural
systems can be improved without either significantly increasing the structure mass or requiring high cost of control power.
They may be called intelligent structures, adaptive structures, active structures, and the related technologies adaptronics,
structronics, etc. These terminologies refer to a smart structure which is an integration of actuators, sensors, control units,
and signal processing units with a structural system. The materials that are usually used to implement the smart structure
are: piezoelectrics, shape memory alloys, electrostrictive/magnetostrictive materials, polymer gels, etc. [32].
Semiactive devices have been applied to large civil structures. Semiactive control strategies combine favorable features
of both active and passive control systems. Semiactive control devices include devices such as variable-orifice dampers,
variable-stiffness devices, variable-friction dampers, controllable-fluid dampers, shape memory alloy actuators, piezo-
electrics, etc. [31,32]. In particular, one of the controllable-fluid dampers, MR damper developed by Lord Corporation has
attracted attention in recent years because it has many attractive characteristics.
In general, an MR damper consists of a hydraulic cylinder, magnetic coils, and MR fluids that consist of micron-sized
magnetically polarizable particles floating within oil-type fluids as shown in Fig. 1. The MR damper is operated as a passive
damper; however, when a magnetic field is applied to the MR fluids, the MR fluids are changed into a semiactive device in a
few milliseconds. Its characteristics are summarized: (1) an MR damper is operated with low power sources, e.g., SD-1000
MR damper can generate a force up to 3000 N using a small battery with capacity less than 10 W; (2) it has high yield stress
level, e.g., its maximum yield stress is beyond 80 kPa; (3) the performance is stable in a broad temperature range, e.g., MR
fluids operate at the temperature between 40 and 150 1C; (4) the response time is a few milliseconds; (5) the performance
is not sensitive to contamination during manufacturing of the MR damper. Moreover, the operating point of the MR
damper, which is a current-controlled device, can be changed by a permanent magnet.
To fully use the best features of the MR damper, a mathematical model that portrays the nonlinear behavior of the MR
damper has to be developed first. However, this is challenging because the MR damper is a highly nonlinear hysteretic
device. Spencer et al. [34] proposed a modified version of the Bouc-Wen model, as shown in Fig. 2. The MR damper force

Fig. 1. Schematic of the prototype 20-ton large-scale MR damper [33].

 ARTICLE IN PRESS
302 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

Fig. 2. Modified Bouc-Wen model of the MR damper [34].

fMR(t) predicted by the modified Bouc-Wen model is governed by the following differential equations according to Spencer
et al. [34]
f MR ¼ c1 y_ þ k1 ðx  x0 Þ, (1)

z_ BW ¼ gjx_  y_ jzBW jjzBW jn1  bðx_  y_ ÞjzBW jn þ Aðx_  y_ Þ, (2)

1
y_ ¼ fazBW þ c0 x_ þ k0 ðx  yÞg, (3)
c0 þ c1

a ¼ aa þ ab u, (4)

c1 ¼ c1a þ c1b u, (5)

c0 ¼ c0a þ c0b u, (6)

u_ ¼ Zðu  vÞ, (7)

where zBW and a, called evolutionary variables, describe the hysteretic behavior of the MR damper; c0 is the viscous
damping parameter at high velocities; c1 is the viscous damping parameter for force roll-off at low velocities; aa, ab, c0a, c0b,
c1a, and c1b are parameters that account for the dependence of the MR damper force on the voltage applied to the current
driver; k0 controls the stiffness at large velocities; k1 represents the accumulator stiffness; x0 is the initial displacement of
spring with stiffness k1; g, b and A are adjustable shape parameters of the hysteresis loops, i.e., the linearity in the
unloading and the transition between pre-yielding and post-yielding regions; v and u are input and output voltages of a
first-order filter, respectively; and Z is the time constant of the first-order filter. Note, nonlinear phenomena occur when the
highly nonlinear MR dampers are applied to structural systems for efficient energy dissipation. Such an integrated
structure and MR damper system behaves nonlinearly, although the structure itself is usually assumed to remain linear.
In what follows, a novel methodology to design a nonlinear control law for operation of an MR damper is addressed.

3. Structural control

3.1. Design framework

Takagi–Sugeno (TS) fuzzy model, PDC, and LMIs are the backbone for the suggested SNFC system. In this section,
Takagi–Sugeno fuzzy model is briefly represented in terms of state space equations. Then, fundamentals of the PDC and
LMIs are discussed.

3.1.1. Takagi and Sugeno fuzzy model

In 1985, Takagi and Sugeno [35] suggested an effective way for modeling complex nonlinear dynamic systems by
introducing linear equations in consequent parts of a fuzzy model, which is called Takagi–Sugeno fuzzy model. It has led to
reduction of computational cost because it does not need any defuzzification procedure. However, a more important value
of the TS fuzzy model is that it provides a framework for a rigorous mathematical analysis, i.e., many modern linear system
theories can be applied to nonlinear system models in terms of IF-THEN rules. A typical fuzzy rule for a TS fuzzy model is of
 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 303

the following form

Rj : If z1FZ is p1;j and z2FZ is p2;j and . . . and ziFZ is pi;j

Then y ¼ f ðz1FZ ; :::; ziFZ Þ, (8)

where pi,j are fuzzy sets centered at the jth operating point, ziFZ
are premise variables that can be either input or output
values. The equation of the consequent part, y ¼ f ðz1FZ ; . . . ; ziFZ Þ can be any linear equation. However, it is advantageous to
represent the consequent part in terms of state space equations; a typical rule of the TS fuzzy model that is expressed in
terms of state space equations in the consequent part is of the form

Rj : If z1FZ is p1;j and . . . and znFZ is pn;j

( )
x_ ¼ Aj x þ Bj u
Then ; j ¼ 1; 2; . . . ; Nr , (9)
y ¼ C j x þ Dj u

where Nr is the number of fuzzy rules, n is the number of premise variables, x is the state vector, u is the input vector, y is
the output vector, and Aj, Bj, Cj, Dj are system matrices.
Note, Eq. (9) represents the jth local linear subsystem of a nonlinear dynamic system, i.e., a linear dynamic system
model that is operated in only a limited region. Therefore, all the local subsystems should be integrated into a global
nonlinear dynamic system by blending operating regions of each local subsystem. Such a blending job is performed
through interpolation of all the local subsystem models. Note that the rules of the local subsystem involve only linear
combinations of input and output vectors; however, the integrated subsystem is truly nonlinear. The blended TS fuzzy
model is of the form
PN r i
j¼1 wj ðzFZ Þ½Aj x þ Bj u
x_ ¼ PNr , (10)
i
j¼1 wj ðzFZ Þ

where wj ðziFZ Þ ¼ Pni¼1 mpi;j ðziFZ Þ and mpi;j ðziFZ Þ is the grade of membership of ziFZ in pi,j.
To control the responses of the blended TS fuzzy model, an effective control law associated with Eq. (10), i.e., u is
designed. In this research, multiple linear controllers associated with each local subsystem are designed and then blended
via fuzzy interpolation method.

3.1.2. Parallel distributed compensation (PDC)

Development of systematic design procedures for a nonlinear feedback control is still challenging due to its complexity.
Thus, an important question would be whether or not linear system theories can be applied to nonlinear control system
design. A clear answer can be found in the PDC. In the PDC approach, linear system theories can be applied to state
feedback controllers associated with each local linear subsystem; in particular, the PDC approach is appropriate for the
ANFC system because the subsystem in the consequent part of the fuzzy rules is described by a linear state space equation.
The control rule Rj of an ANFC system is of the form

Rj : If z1FZ is p1;j and . . . and znFZ is pn;j

Then uj ¼ Kj x. (11)

The state feedback controller in the consequent part of the jth IF-THEN rule is a local linear controller associated with a
local linear subsystem to be controlled. All the local state feedback controllers are integrated into a global nonlinear
controller using fuzzy sets
PN r i
j¼1 wj ðzFZ Þ½Kj x
u ¼ PN . (12)
i
j¼1 wj ðzFZ Þ
r

Notice that the blended state feedback controller is truly nonlinear. By substituting Eq. (12) into Eq. (10), the closed loop
nonlinear control system is derived
PNr PNr i i
j q¼1 wj ðzFZ Þwq ðzFZ Þ½Aj þ Bj Kq x
x_ ¼ PN r PN r . (13)
i i
j q¼1 wj ðzFZ Þwq ðzFZ Þ

To implement the ANFC system Eq. (13), the next step is to compute the multiple state feedback gains, Kq, q ¼ 1, y, Nr such
that they guarantee global asymptotical stability (GAS) of a structural system, while the performance on transient
responses is also satisfied. Next, they are integrated with the associated Kalman filters to convert the state feedback
control into the output feedback one and then are integrated with a clipped algorithm to convert the active mode into
semiactive one.
Although there might exist many methodologies to design the state feedback gains Kq, LMI-based control formulations
are carried out here because the LMI technique is appropriate to the formulation of multiple objectives and constraints.
Basic background of LMIs is discussed, following by the description of stability issues.
 ARTICLE IN PRESS
304 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

3.1.3. Linear matrix inequalities (LMIs)

In recent years, LMI techniques have attracted significant attention because a great variety of engineering problems can
be re-formulated as convex or pseudo-convex optimization problems in terms of LMIs. Such problems include control
system design, system identification, structural design, etc. In particular, many control problems can be re-cast in terms of
LMIs because design objectives and constraint conditions can be formulated in numerically tractable manner.
An LMI has the form of
X
m
Q ðxÞ9Q 0 þ xi Q i o0, (14)
i¼1

where Q i ¼ Q Ti 2 <nn ; i ¼ 0; . . . ; m are given as the symmetric matrices and xi 2 <m ; i ¼ 1; :::; m are the design variables to
be solved [36]. The inequality symbol of o0 represents negative definite, i.e., the largest eigenvalue of Q(x) is negative.
Finding a solution of the LMI Eq. (14) is a convex optimization problem because the LMI Eq. (14) is a convex constraint
on x. One of the main advantages of the LMI formulation is that design objectives and design constraints that can arise in
control system design can be combined in numerically tractable manner. Note that multiple LMIs that can arise in control
system design can be considered as a single LMI, i.e.,
8 9
> Q 1 ðxÞo0 >
>
< >
=
..
. is equal to Q ðxÞ :¼ diagfQ 1 ðxÞ; :::; Q n ðxÞgo0. (15)
>
> >
: Q ðxÞo0 > ;
n

A typical example that can be re-cast in terms of LMIs in the field of control engineering is the Lyapunov inequality

AT P þ PAo0, (16)
nn T
where A 2 < is the given system matrix of a dynamic system, P ¼ P is a symmetric positive definite matrix as a
design variable:

Find P ¼ PT such that AT P þ PAo0. (17)

In this research, from the Lyapunov inequality equation, a convex optimization problem to find a set of feasible solution is
formulated.

3.2. Formulations for LMI-based nonlinear control system design

This section presents an LMI-based systematic design approach for an ANFC system with multi-objective requirements
(global asymptotical stability and transient response performance).

3.2.1. Stability conditions

Stability analysis is one of the most essential issues in system and control engineering fields. It is generally assumed that
a semiactive control system is inherently stable in bounded input and bounded output (BIBO) sense because a semiactive
device does not add mechanical energy into a structural system [33]. This might be true for open loop systems; however, it
is not always necessarily true for feedback control systems because a semiactive control system can be destabilized due to
the structure–semiactive MR damper interaction, the inaccurate structure or MR damper modeling, etc. [37,38]. It should
be noted that the statement that a semiactive control system is inherently stable in the BIBO sense cannot be generalized.
Therefore, stability should be necessarily considered for a semiactive control system design. In particular, it might be much
more important issue in a semiactive nonlinear control system.
In general, it is difficult to include a stability condition for the design of a typical fuzzy logic-based controller in a
systematic way [39], e.g., Mamdani fuzzy model-based controller. The reason is that a Mamdani fuzzy system does not
provide a rigorous mathematical framework for stability analysis. However, such a drawback of the Mamdani fuzzy system
can be solved by a TS fuzzy model. Advanced system theories such as Lyapunov theorem can be applied to the TS fuzzy
model to address the stability issue because the consequent part of the TS fuzzy model is expressed in terms of linear
functions, e.g., state space equations.
Consider the following linear time-invariant (LTI) dynamic system

x_ ¼ Ax, (18)
where x is the states and A is a system matrix. The stability of this LTI dynamic system can be checked via eigenvalue
analysis. The asymptotic stability can be also investigated using Lyapunov theorem:

Theorem 1. [40]. The equilibrium point of the LTI dynamic system is asymptotically stable if there exist a symmetric positive
definite matrix P and a positive definite matrix QLy such that

AT P þ PA ¼ Q Ly . (19)
 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 305

Based on this Lyapunov equation, Tanaka and Sugeno [41] suggested a stability condition for a nonlinear Takagi–Sugeno
(TS) fuzzy model.

Theorem 2. [41]. The equilibrium point at the origin of the continuous nonlinear TS fuzzy model is globally asymptotically stable
if there exists a common symmetric positive definite matrix P such that

ATj P þ PAj o0 for all j ¼ 1; 2; . . . ; Nr , (20)

where Nr is the number of the fuzzy rules. This stability condition for an open loop system can be extended into a stability
condition for a closed loop dynamic system.

Theorem 3. [42]. The equilibrium point at the origin of the continuous closed loop TS fuzzy model is globally asymptotically
stable if there exists a common symmetric positive definite matrix P such that

ðAj þ Bj Kq ÞT P þ PðAj þ Bj Kq Þo0 for all j; q ¼ 1; 2; . . . ; Nr . (21)

Note, N 2r LMI formulations should be solved to determine the common symmetric positive definite matrix P; however, the
computational cost can be reduced by approximately a half by grouping terms.

Corollary 1. [42]. The equilibrium point at the origin of the continuous closed loop TS fuzzy model is globally asymptotically
stable if there exists a common symmetric positive definite matrix P such that

ðAj þ Bj Kj ÞT P þ PðAj þ Bj Kj Þo0 for all j ¼ 1; 2; . . . ; Nr , (22)

GTjq P þ PGjq o0 for all j; q ¼ 1; 2; . . . ; Nr , (23)

where
ðAj þ Bj Kq Þ þ ðAq þ Bq Kj Þ
Gjq ¼ ; joqpN r . (24)
2

Using Eqs. (22)–(24), the number of LMIs to be solved has been reduced from N 2r of Eq. (21) to Nr(Nr+1)/2.
Note, from Eqs. (22)–(24), control gains Kj are not automatically obtained; the equations are only used to check if the
designed control systems are stable, i.e., the system matrices Aj and Bj are given, the control matrix Kj should be designed
first. Then, they are checked to determine whether the controllers Kj, j ¼ 1, 2, y, Nr stabilize the closed loop systems. This
procedure generally requires many trial-and-errors; therefore, it is desirable to integrate the stability conditions with the
control system design procedure. In other words, it is necessary to formulate Kj and P as design matrix variables at the
same time.

3.2.2. LMI formulation of stabilizing control

In this section, the control system design procedure is integrated with the checking process of the stability condition
[43]. By substituting Eq. (24) into Eq. (23), the set of equations can be written as

ðAj þ Bj Kj ÞT P þ PðAj þ Bj Kj Þo0; j ¼ 1; 2; . . . ; N r , (25)

ðAj þ Bj Kq ÞT P þ ðAq þ Bq Kj ÞT P þ PðAj þ Bj Kq Þ þ PðAq þ Bq Kj Þo0; joqpNr . (26)

1
Pre- and post-multiplying Eqs. (25) and (26) by P 40 leads to

P1 ðAj þ Bj Kj ÞT PP1 þ P1 PðAj þ Bj Kj ÞP1 o0; joqpr ¼ 1; 2; . . . ; N r , (27)

P1 ðAj þ Bj Kq ÞT PP1 þ P1 ðAq þ Bq Kj ÞT PP1 þ P1 PðAj þ Bj Kq ÞP1 þ P1 PðAq þ Bq Kj ÞP1 o0,
joqpr ¼ 1; 2; . . . ; Nr . (28)
1
Defining Q ¼ P and KjQ ¼ Mj leads to

QATj þ Aj Q þ MTj BTj þ Bj Mj o0; j ¼ 1; 2; . . . ; Nr , (29)

QATj þ Aj Q þ QATq þ Aq Q þ MTq BTj þ Bj Mq þ MTj BTq þ Bq Mj o0; joqpN r . (30)

Eqs. (29) and (30) are used to design stabilizing feedback control gains which can be solved using standard interior point
optimization-based LMI solver. However, these stabilizing control formulations do not directly address the transient
performance. The performance-based design can be achieved through integration of a pole-placement algorithm with the
stabilizing controllers.
 ARTICLE IN PRESS
306 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

3.2.3. LMI formulation of pole-assignment control

In structural engineering, the performance on transient responses is an important issue; however, the stability LMI
formulation does not directly deal with that issue. Therefore, in this section, the pole-assignment concept is recast by LMI
formulation. The formulation of the pole-placement in terms of a LMI is motivated by Chilali and Gahinet [44]. A D-stable
region is defined first.

Definition 1. Let D be a subset of the complex plane that represents behavior of a dynamic system x_ ¼ Ax. If all the poles
(or eigenvalues) of the dynamic system are located in the subset region D, the dynamic system x_ ¼ Ax (or the system
matrix A) is D-stable.

Definition 2. LMI stability region

The closed loop poles (or eigenvalues) of a dynamic system are located in the LMI stability region
T
D ¼ fs 2 C : f D ðsÞ :¼ a þ bs þ b s̄o0g, (31)

if there exist a symmetric positive definite matrix a ¼ ½akl  2 <mm and a matrix b ¼ ½bkl  2 <mm . The characteristic
function fD(s) is a m  m Hermitian matrix.

Hong and Langari [1] applied for a circular LMI region D that is convex and symmetric with respect to real axis such that
it is associated with Chilali and Gahinet theorem [44]. A circular LMI region that has a center at (qc, 0) and a radius rc40
is defined

D ¼ fx þ jy 2 C : ðx þ qc Þ2 þ y2 or 2c g. (32)

The associated characteristic function fD(s) is given by

!
r c s̄ þ qc
f D ðsÞ ¼ . (33)
s þ qc r c

Fig. 3 shows a schematic of the circular LMI region. This circular LMI region can be related to a LMI stability region
described in terms of an m  m block matrix.

Theorem 4. [44]. The system dynamics of x_ ¼ Ax is D-stable if and only if there exists a symmetric matrix Q such that
T
M D ðA; XÞ :¼ a  Q þ b  ðAQ Þ þ b  ðAQ ÞT ¼ ½akl Q þ bkl AQ þ blk QAT 1pk;lpm o0; Q 40. (34)

Fig. 3. Circular region D for pole location [1].

 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 307

It should be noted that MD(A, Q) and fD(s) have close relationship, i.e., replacing ð1; s; s̄Þ of fD(s) by (Q, AQ, QAT) of
MD(A, Q) yields
!
r c Q qc Q þ QAT
o0; Q 40. (35)
qc Q þ AQ r c Q

From Eq. (35), a LMI for the pole-placement controller is derived.

Theorem 5. [1]. The continuous closed loop TS fuzzy control system is D-stable if and only if there exists a positive symmetric
matrix Q such that
!
r c Q qc Q þ Q ðAj þ Bj Kq ÞT
o0. (36)
qc Q þ ðAj þ Bj Kq ÞQ r c Q

Remark. It should be noted that the inequality Eq. (36) is not a LMI because the matrices Q and Kj are coupled. This
nonlinear matrix inequality can be transformed into a LMI by defining a new matrix variable, i.e., Mj ¼ KjQ.

Corollary 2. The continuous closed loop TS fuzzy control system is D-stable if and only if there exists a positive symmetric matrix
Q such that
!
r c Q qc Q þ QATj þ MTj BTj
o0. (37)
qc Q þ Aj Q þ Bj Mj r c Q

This LMI (37) directly deals with the transient response performance of the dynamic system.

In summary, three LMIs Eqs. (29), (30) and (37) are solved simultaneously to obtain Q and Mj. Then the common
symmetric positive definite matrix P and state feedback control gains Kj are determined

P ¼ Q 1 ,
Kj ¼ Mj Q 1 ¼ Mj P; j ¼ 1; 2; . . . ; Nr . (38)
These state feedback control gains are integrated with state estimators to construct output feedback controllers.

3.3. Output feedback-based semiactive nonlinear fuzzy control

In this section, two more design units, which are a state estimator and a semiactive converting algorithm, are introduced
to convert the full state feedback-based active nonlinear control system into an output feedback-based semiactive
nonlinear control system.

3.3.1. State estimator

From practical point of view, it is challenging to implement the full state feedback strategies because all the states are
not readily available; however, it can be solved by the aid of estimation theories.
Consider the following state space equation
x_ ¼ Aj x þ Bj u, (39)

y ¼ Cj x þ Dj u. (40)
This system model can be expanded to an observer model by adding and subtracting a term Ljy into Eq. (39),
and applying Eq. (40)

x_^ ¼ ðAj þ Lj Cj Þx^ þ ðBj þ Lj Dj Þu  Lj y. (41)

The estimated states are used for feedback control
^
u ¼ Kj x. (42)
In this paper, the optimal observer gains Lj are obtained by Kalman filter estimation procedure using the Matlab subroutine
lqew in the Control System Toolbox [45].

3.3.2. Clipped algorithm

Once the output feedback-based active nonlinear fuzzy controller (ANFC) is designed, a semiactive converter and an MR
damper are integrated with the output ANFC to develop an SNFC system. In general, an MR damper cannot be directly
controlled by a control algorithm. The reason is that a controller generates force signals while an MR damper requires
voltage or current signals to be operated. Therefore, a unit that converts a control force signal into a voltage signal should
be integrated with the ANFC system to construct an SNFC system. Such a unit would be either to use an inverse MR damper
model or to implement a converting algorithm. Candidates for the inverse MR damper models may include Bingham,
 ARTICLE IN PRESS
308 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

polynomial, Bouc-Wen, and modified Bouc-Wen models. Another good candidate for the conversion is a clipped
algorithm [46]

v ¼ V a Hðff ANFC  f m gf m Þ, (43)

where v is the applied voltage level, Va is the arbitrary voltage level, H is a Heaviside step function, fm is a measured MR
damper force, and fANFC is a control force signal generated by an ANFC system
P N r Qn
j¼1 i¼1mi;j ðziFZ Þ½Kj x
^
f ANFC ¼ PNr Qn . (44)
i
j¼1 i¼1 mi;j ðzFZ Þ

By substitution of Eq. (44) into Eq. (43), the final voltage equation [47] can be written as
(PNr Qn i
) !
j¼1 i¼1 mi;j ðzFZ Þ½Kj x
^
v ¼ V aH PNr Qn  fm fm . (45)
i
j¼1 i¼1 mi;j ðzFZ Þ

This SNFC system is applied to a three-story shear type building structure equipped with an MR damper to demonstrate its
performance.

4. Numerical example

In this section, a three-story shear type building structure employing an MR damper is presented. The goal is to evaluate
the performance of the semiactive nonlinear fuzzy control system. A typical example of a building structure employing an
MR damper is depicted in Fig. 4. Note, the MR damper can be applied to arbitrary locations within this building structure.
The reason to install the MR damper on the 1st floor is that previous researchers [48] have demonstrated the effectiveness
of this approach. In addition, Fig. 5 shows how an MR damper is implemented into a building structure.
The associated equation of motion is given by

Mx€ þ Cx_ þ Kx ¼ Cf MR ðt; x1 ; x_ 1 ; v1 Þ  MKw

€ g, (46)

where
2 3
m1 0 0
6 0 7
M¼4 0 m2 5 (47)
0 0 m3

Fig. 4. A schematic of a building–MR damper system.

 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 309

Fig. 5. Integrated building structure–MR damper system.

is the mass matrix,

2 3
c1 þ c2 c2 0
6 c c2 þ c3 c3 7
C¼4 2 5 (48)
0 c3 c3

is the damping matrix,

2 3
k1 þ k2 k2 0
6 k2 þ k3 k3 7
K ¼ 4 k2 5 (49)
0 k3 k3

is the stiffness matrix,

2 3
f MR ðt; x1 ; x_ 1 ; v1 Þ
6 7
f MR ðt; x1 ; x_ 1 ; v1 Þ ¼ 4 0 5 (50)
0

is MR damper force input matrix; w € g denotes the ground acceleration; mi are the mass of the ith floor; ki are the stiffness of
the ith floor columns; ci are the damping of the ith floor columns; the vector x and x_ are the displacement and velocity
relative to the ground; x€ is the acceleration; x1 and x_ 1 are the relative displacement and velocity at the 1st floor level to the
ground of the three-story building structure, respectively; v1 is the voltage level to be applied; and C and K are location
vectors of control forces and disturbance signals, respectively. This second-order differential equations can be converted
into state space

z_ ¼ Az þ Bf MR ðt; z1 ; z4 ; v1 Þ  Ew
€g
y ¼ Cz þ Df MR ðt; z1 ; z4 ; v1 Þ þ n, (51)

where
 
0 I
A¼ 1 1 (52)
M K M C

is the state matrix,

 
0
B¼ (53)
M1 F

is the input matrix,

2 3
I 0
6 7
C ¼ 40 I 5 (54)
1 1
M K M C

is the output matrix,

2 3
0
6 7
D ¼ 40 5 (55)
M1 F
 ARTICLE IN PRESS
310 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

Table 1
Parameters for SD-1000 MR damper model [34]

Parameter Value

1
c0a (N s cm ) 21.0
c0b (N s cm1 V1) 3.50
k0 (N cm1) 46.9
c1a (N s cm1) 283
c1b (N s cm1 V1) 2.95
k1 (N cm1) 5.00
x0 (cm) 14.3
aa (N cm1) 140
ab (N cm1 V1) 695
g (cm2) 363
b (cm2) 363
A 301
N 2
Z (s1) 190

is the feedthrough matrix,

 
0
E¼ (56)
F
is the disturbance location matrix,
2 3
1 1 0
6 7
F ¼ 4 0 1 1 5 (57)
0 0 1
is the location matrix that a Chevron brace is located within a building structure; n is the noise; z is the state vector; and z1
and z4 are the displacement and the velocity at the 1st floor level of the three-story building structure, respectively. In this
building structure, a SD-1000 MR damper has been applied whose parameters are given in Table 1.
Properties of the three-story building structure are adopted from a scaled model [48] of a prototype building structure
that was developed by Chung et al. [49]. The mass of each floor m1 ¼ m2 ¼ m3 ¼ 98:3 kg; the stiffness of each story
k1 ¼ 516,000 N m1, k2 ¼ 684,000 N m1, and k3 ¼ 684,000 N m1; and the damping coefficients of each floor c1 ¼ 125 N s m1,
c2 ¼ 50 N s m1, and c3 ¼ 50 N s m1. In addition, the SD-1000 MR damper whose approximate capacity is 1500 N is installed
on the 1st floor using a Chevron brace, which leads to a nonlinear dynamic model, i.e., a building–MR damper system.
Using Eqs. (29), (30) and (37), one can design fuzzy state feedback controllers that guarantee GAS of the closed loop
control system and provides the desired transient responses by constraining the closed loop poles in a region D such that
(qc, rc) ¼ (50, 45). This region puts a lower bound on both exponential decay rate and damping ratio of the closed loop
response.

4.1. Control performance evaluation: normal case

The 1940 El-Centro earthquake input is applied as a ground motion shown in Fig. 6. The time history responses that are
controlled by the SNFC system at the entire floor are compared with the performance of a traditional optimal controller,
i.e., H2/LQG, while the uncontrolled system response is used as the baseline. The parameters of the linear quadratic
regulator (LQR) and the Kalman filter are adopted from Dyke et al. [48,50]. Figs. 7–9 compare time history responses at the
1st, 2nd, and 3rd floor of an uncontrolled, a H2/LQG controlled, and an SNFC controlled systems, respectively. The upper
graph at each figure shows the time history of displacement responses; the lower graph depicts the time history of
acceleration responses. According to the time history responses, both displacement and acceleration responses are
significantly reduced when either H2/LQG or SNFC systems are applied. Fig. 10 compares maximum values of the interstory
responses of the uncontrolled, the H2/LQG controlled, and the SNFC controlled systems. The graph (a) is about the
comparison of the maximum displacement responses, (b) compares maximum drift responses, and (c) compares maximum
acceleration responses. In terms of the maximum values of the displacement and drift responses at the 2nd and the 3rd
floors, the SNFC system is better than the H2/LQG controller; while the slightly larger responses at the 1st floor within the
SNFC system are observed to accompany the response reduction of the 2nd and the 3rd floor levels. However, the SNFC
system yields significant improvement for the acceleration response reduction on the 1st and 2nd floors over the H2/LQG
control system. Furthermore, the normalized maximum value of the applied MR damper force, which is a value normalized
by the seismic weight of the building structure [30] within the SNFC system, is evaluated
 
f MR ðtÞ
J 1 ¼ max , (58)
t Ws
 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 311

Fig. 6. 1940 El Centro earthquake record.

Fig. 7. Time history responses of an uncontrolled, a H2/LQG control and an SNFC systems at the 1st floor.

where fMR(t) is the applied MR damper force and Ws is the seismic weight of the building structure. The result is: the use of
MR damper force by the proposed SNFC system is 0.2563, while the H2/LQG control system uses the MR damper force of
0.3490, i.e., the SNFC system is more efficient than the H2/LQG controller. Therefore, it can be summarized: (1) the SNFC
system is slightly more effective in reducing the displacement and drift responses of the building than the H2/LQG
controller while it is much more effective in mitigating the acceleration responses than the H2/LQG control system; and (2)
the SNFC system uses less power than the H2/LQG controller.

4.2. Control performance evaluation: severe noise case

In addition, the robustness of the SNFC system is compared with that of the H2/LQG control system with 30% zero-mean
Gaussian white noise. Figs. 11–13 compare the time history responses at the 1st, 2nd, and 3rd floor of an uncontrolled,
a H2/LQG controlled, and an SNFC controlled systems, respectively. The upper graph in each figure shows the time history
of displacement responses and the lower graph shows the time history of acceleration responses. Fig. 14 compares
maximum values of the interstory responses of the uncontrolled, the H2/LQG controlled, and the SNFC controlled systems.
The graph (a) is about the comparison of maximum displacement responses, (b) compares maximum drift responses,
 ARTICLE IN PRESS
312 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

Fig. 8. Time history responses of an uncontrolled, a H2/LQG control and an SNFC systems at the 2nd floor.

Fig. 9. Time history responses of an uncontrolled, a H2/LQG control and an SNFC systems at the 3rd floor.

Fig. 10. Interstory responses of an uncontrolled, a H2/LQG control and an SNFC systems.
 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 313

Fig. 11. Time history responses at the 1st floor (disturbed by 30% sensing noise from a zero-mean Gaussian white noise generator).

Fig. 12. Time history responses at the 2nd floor (disturbed by 30% sensing noise from a zero-mean Gaussian white noise generator).

Fig. 13. Time history responses at the 3rd floor (disturbed by 30% sensing noise from a zero-mean Gaussian white noise generator).

and (c) compares maximum acceleration responses. The figures show that both SNFC and H2/LQG control system are
effective in reducing all the interstory, displacement, drift, and acceleration responses of the entire floor levels of the three-
story building structure with severe sensing noises. However, the SNFC system yields much smaller acceleration responses
 ARTICLE IN PRESS
314 Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315

Fig. 14. Interstory responses of an uncontrolled, a LQG control and an SNFC systems (disturbed by 30% sensing noise from a zero-mean Gaussian white
noise generator).

than the H2/LQG control system. Furthermore, the normalized maximum value of the MR damper force that is applied to
the SNFC system is 0.260, while that of the H2/LQG control system is 0.3743, i.e., the SNFC system is more efficient than the
H2/LQG controller. That is, compared with the H2/LQG control system, the SNFC system effectively reduces the structural
responses, in particular, acceleration responses with less power sources, which eventually may lead to the reduced
distribution of the associated shear forces within the structure. Therefore, it is demonstrated that the SNFC system is
cost-effective in mitigating the seismic responses of a low-rise building structure subjected to El-Centro earthquake
disturbance.

5. Conclusion

In this paper an LMI-based systematic design methodology for the semiactive nonlinear fuzzy control (SNFC) of a class
of nonlinear building–magnetorheological damper systems with multi-objective requirements (i.e., stability conditions and
the performance on transient responses) in a unified framework was proposed. The framework is based on PDC that forms
a fuzzy gain scheduling scheme. It employs multiple Lyapunov-based linear controllers that correspond to local linear
dynamic models with automatic scheduling performed via fuzzy rules. Each local linear controller is formulated via linear
matrix inequalities (LMIs) such that global asymptotical stability is guaranteed and the performance on transient responses
is also satisfied. Then, the associated Kalman observers are designed for output feedback control systems. Lastly, all of the
local linear controllers and the estimators are blended via fuzzy interpolation method.
It is demonstrated from simulation that the proposed SNFC system effectively reduces the vibration of the seismically
excited building structure equipped with an MR damper. In addition, the proposed SNFC system is close or better than the
previously designed control system, H2/LQG controller. In particular, the SNFC system is more effective in mitigating
acceleration responses of the structure than H2/LQG control system while it uses less control power than H2/LQG control
system.
Further research is recommended to apply for the proposed SNFC system to the larger-scale building structures
equipped with many MR dampers. It is also recommended that an experimental study to verify the effectiveness of the
SNFC system be investigated. This might include a small-scale and hybrid full-scale testing.

References

[1] S.K. Hong, R. Langari, An LMI-based HN fuzzy control system design with TS framework, Information Sciences 123 (2000) 163–179.
[2] R. Langari, Synthesis of nonlinear control strategies via fuzzy logic, in: Proceedings of the American Control Conference, San Francisco, CA, 1993,
pp. 1855–1859.
[3] R. Langari, Past, present and future of fuzzy control: a case for application of fuzzy logic in hierarchical control, in: Proceedings of Annual Conference
of the North American Fuzzy Information Processing Society, New York City, NY, 1999, pp. 760–765.
[4] S. Lei, R. Langari, Hierarchical fuzzy logic control of a double inverted pendulum, in: Proceedings of IEEE International Conference on Fuzzy Systems,
San Antonio, TX, 2000, pp. 1074–1077.
[5] J. Yen, R. Langari, Fuzzy Logic—Intelligence, Control, and Information, Prentice-Hall, Upper Saddle River, NJ, USA, 1998.
[6] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338–353.
[7] M. Battaini, F. Casciati, L. Faravelli, Fuzzy control of structural vibration. An active mass system driven by a fuzzy controller, Earthquake Engineering &
Structural Dynamics 27 (1998) 1267–1276.
 ARTICLE IN PRESS
Y. Kim et al. / Mechanical Systems and Signal Processing 23 (2009) 300–315 315

[8] M. Battaini, F. Casciati, L. Faravelli, Controlling wind response through a fuzzy controller, Journal of Engineering Mechanics (ASCE) 130 (2004)
486–491.
[9] C.H. Loh, L.Y. Wu, P.Y. Lin, Displacement control of isolated structures with semi-active control devices, Journal of Structural Control 10 (2003)
77–100.
[10] K.S. Park, H.M. Koh, C.W. Seo, Independent modal space fuzzy control of earthquake-excited structures, Engineering Structures 26 (2003) 279–289.
[11] R.S. Subramaniam, A.M. Reinhorn, M.A. Riley, S. Nagarajaiah, Hybrid control of structures using fuzzy logic, Microcomputers in Civil Engineering 11
(1996) 1–17.
[12] M. Symans, S.W. Kelly, Fuzzy logic control of bridge structures using intelligent semi-active seismic isolation systems, Earthquake Engineering &
Structural Dynamics 28 (1999) 37–60.
[13] M. Al-Dawod, B. Samali, F. Naghdy, K.C.S. Kwok, F. Naghdy, Fuzzy controller for seismically excited nonlinear buildings, Journal of Engineering
Mechanics (ASCE) 130 (2004) 407–415.
[14] M. Al-Dawod, B. Samali, F. Naghdy, K.C.S. Kwok, Active control of along wind response of tall building using a fuzzy controller, Engineering Structures
23 (2001) 1512–1522.
[15] L. Faravelli, R. Rossi, Adaptive fuzzy control: theory versus implementation, Journal of Structural Control 9 (2002) 59–73.
[16] L. Faravelli, T. Yao, Use of adaptive networks in fuzzy control of civil structures, Microcomputers in Civil Engineering 11 (1996) 67–76.
[17] K.C. Schurter, P.N. Roschke, Neuro-fuzzy control of structures using magnetorheological dampers, in: Proceedings of the American Control
Conference, Arlington, VA, 2001, pp. 1097–1102.
[18] A. Tani, H. Kawamura, S. Ryu, Intelligent fuzzy optimal control of building structures, Engineering Structures 20 (1998) 184–192.
[19] L. Zhou, C.C. Chang, L.X. Wang, Adaptive fuzzy control for nonlinear building–magnetorheological damper system, Journal of Structural Engineering
(ASCE) 129 (2003) 905–913.
[20] A.S. Ahlawat, A. Ramaswamy, Multi-objective optimal design of FLC driven hybrid mass damper for seismically excited structures, Earthquake
Engineering & Structural Dynamics 31 (2000) 1459–1479.
[21] A.S. Ahlawat, A. Ramaswamy, Multiobjective optimal FLC driven hybrid mass damper system for torsionally coupled, Seismically Excited Structures,
Earthquake Engineering & Structural Dynamics 31 (2002) 2121–2139.
[22] A.S. Ahlawat, A. Ramaswamy, Multiobjective optimal fuzzy logic control system for response control of wind-excited tall buildings, Journal of
Engineering Mechanics (ASCE) 130 (2004) 524–530.
[23] H.S. Kim, P.N. Roschke, Design of fuzzy logic controller for smart base isolation system using genetic algorithm, Engineering Structures 28 (2006)
84–96.
[24] G. Yan, L.L. Zhou, Integrated fuzzy logic and genetic algorithms for multi-objective control of structures using MR dampers, Journal of Sound and
Vibration 296 (2006) 368–382.
[25] H. Alli, O. Yakut, Fuzzy sliding-mode control of structures, Engineering Structures 27 (2005) 277–284.
[26] S.B. Kim, C.B. Yun, B.F. Spencer, Vibration control of wind-excited tall buildings using sliding mode fuzzy control, Journal of Engineering Mechanics
(ASCE) 130 (2004) 505–510.
[27] A.P. Wang, C.D. Lee, Fuzzy sliding mode control for a building structure based on genetic algorithms, Earthquake Engineering & Structural Dynamics
31 (2002) 881–895.
[28] J. Joh, R. Langari, F.T. Jeung, W.J. Chung, A new design method for continuous Takagi–Sugeno fuzzy controller with pole placement constraints: an LMI
approach, in: Proceedings of the 1997 IEEE International Conference on Systems, Man, and Cybernetics, Orlando, FL, 1997, pp. 2969–2974.
[29] K. Tanaka, M. Sano, A robust stabilization problem of fuzzy control systems and its application to backing up control of a truck-trailer, IEEE
Transactions on Fuzzy Systems 2 (1994) 119–134.
[30] L.M. Jansen, S.J. Dyke, Semiactive control strategies for MR dampers: comparative study, Journal of Engineering Mechanics (ASCE) 126 (2000)
795–803.
[31] L. Gaul, S. Hurlebaus, H. Albrecht, J. Wirnitzer, Enhanced damping of lightweight structures by semi-active joints, Acta Mechanica 195 (2008)
249–261.
[32] S. Hurlebaus, L. Gaul, Smart structure dynamics, Mechanical Systems and Signal Processing 20 (2006) 255–281.
[33] G. Yang, B.F. Spencer, J.D. Carlson, M.K. Sain, Large-scale MR fluid dampers: modeling and dynamic performance considerations, Engineering
Structures 24 (2002) 309–323.
[34] B.F. Spencer, S.J. Dyke, M.K. Sain, J.D. Carlson, Phenomenological model for magnetorheological dampers, Journal of Engineering Mechanics (ASCE)
123 (1997) 230–238.
[35] T. Takagi, M. Sugeno, Fuzzy identification of systems and its applications to modeling and control, IEEE Transactions on Systems, Man, and
Cybernetics 15 (1985) 116–132.
[36] S. Boyd, L.E. Ghaoui, E. Feron, V. Balakrishnan, Linear Matrix Inequalities in System and Control Theory, Society for Industrial and Applied
Mathematics, Philadelphia, USA, 1994.
[37] J.L. Kuehn, H.L. Stalford, Stability of a Lyapunov controller for a semi-active structural control system with nonlinear actuator dynamics, Journal of
Mathematical Analysis and Applications 251 (2000) 940–957.
[38] G. Jin, M.K. Sain, B.F. Spencer, Nonlinear blackbox modeling of MR-dampers for civil structural control, IEEE Transactions on Control System
Technology 13 (2005) 345–355.
[39] F. Casciati, Checking the stability of a fuzzy controller for nonlinear structures, Microcomputers in Civil Engineering 12 (1997) 205–215.
[40] H.K. Khalil, Nonlinear Systems, Prentice-Hall, Upper Saddle River, NJ, USA, 2002.
[41] K. Tanaka, M. Sugeno, Stability analysis and design of fuzzy control systems, Fuzzy Sets and Systems 45 (1992) 135–156.
[42] H.O. Wang, K. Tanaka, M. Griffin, An analytical framework of fuzzy modeling and control of nonlinear systems: stability and design issues,
in: Proceedings of the American Control Conference, Seattle, WA, 1995, pp. 2272–2276.
[43] S.S. Farinwata, D. Filev, R. Langari, Fuzzy Control—Synthesis and Analysis, Wiley, New York, 2000.
[44] M. Chilali, P. Gahinet, HN design with pole placement constraints: an LMI approach, IEEE Transactions on Automatic Control 41 (1996) 358–367.
[45] MATLAB, Control System Toolbox, The Math Works, Inc., Natick, MA, USA, 2001.
[46] O. Yoshida, S.J. Dyke, Seismic control of a nonlinear benchmark building using smart dampers, Journal of Engineering Mechanics (ASCE) 130 (2004)
386–392.
[47] Y. Kim, R. Langari, Nonlinear identification and control of a building structure with a magnetorheological damper, in: Proceedings of the American
Control Conference, New York City, NY, 2007, pp. 3353–3358.
[48] S.J. Dyke, B.F. Spencer, M.K. Sain, J.D. Carlson, Modeling and control of magnetorheological dampers for seismic response reduction, Smart Materials
and Structures 5 (1996) 565–575.
[49] L.L. Chung, R.C. Lin, T.T. Soong, A.M. Reinhorn, Experiments on active control for MDOF seismic structures, Journal of Engineering Mechanics (ASCE)
115 (1989) 1609–1627.
[50] S.J. Dyke, B.F. Spencer, M.K. Sain, J.D. Carlson, An experimental study of MR dampers for seismic protection, Smart Materials and Structures 7 (1998)
693–703.