Sie sind auf Seite 1von 12

Engineering Structures 29 (2007) 346357

www.elsevier.com/locate/engstruct
Active control of high rise building structures using fuzzy logic and genetic
algorithms
S. Pourzeynali

, H.H. Lavasani, A.H. Modarayi


Department of Civil Engineering, Faculty of Engineering, The University of Guilan, Rasht, Islamic Republic of Iran
Received 18 October 2004; received in revised form 15 September 2005; accepted 29 April 2006
Available online 27 June 2006
Abstract
Active tuned mass damper (ATMD) control systems for civil engineering structures have attracted considerable attention in recent years. This
paper emphasizes on the combined application of genetic algorithms and fuzzy logic (GFLC) to design and optimize the different parameters
of the ATMD control scheme for getting the best results in the reduction of the building response under earthquake excitations. Therefore, the
proposed method has the advantages of both the fuzzy logic controller (FLC) and genetic algorithms (GAs) to handle the uncertain, as well
non-linear phenomena. The building is modeled as a shear frame and the problem is solved in state space. The proposed method is applied to an
11-story realistic building, located in the city of Rasht, Islamic Republic of Iran. The results obtained from the proposed control scheme (GFLC)
are compared with those obtained from the TMD, and linear quadratic regulator (LQR) control methods. It is found that integration of the GAs
and FLC is highly effective in reduction of the seismically excited example building.
c 2006 Elsevier Ltd. All rights reserved.
Keywords: Active tuned mass damper (ATMD); Tuned mass damper (TMD); Linear quadratic regulator (LQR); Fuzzy logic controller (FLC); Genetic algorithms
(GAs); High rise buildings
1. Introduction
A critical aspect in the design of civil engineering structures
is the reduction of response quantities such as velocities,
deections and forces induced by environmental dynamic
loadings (i.e., wind and earthquake). Structural control methods
are the most recent strategies for this purpose, which can be
classied as active, semi-active, passive, and hybrid control
methods [1].
In the last three decades or so, the reduction of structural
response, caused by dynamic effects, has become a subject
of research, and many structural control concepts have been
implemented in practice.
Tuned mass dampers (TMDs) are the oldest structural
vibration control devices in existence. The concept of vibration
control using a mass damper dates back to the year 1909, when
Frahm invented a vibration control device called a dynamic

Corresponding author. Tel.: +98 911 331 0 919; fax: +98 131 6690271.
E-mail address: pourzeynali@guilan.ac.ir (S. Pourzeynali).
vibration absorber [2]. Active tuned mass damper (ATMD)
systems have also been a popular area of research in recent
decades and signicant progress has been made in this area
over these years [35]. This system has been known as an area
of research in which the motion of a structure is controlled or
modied by means of the action of a control system through
some external energy supply. An ATMD system effectively
reduces the structural response, but the required external control
forces could be extremely large in the case of massive and large
buildings. On the other hand, these systems need a continuous
power supply and a digital computer system during an
earthquake excitation, which may be difcult to provide during
strong earthquakes. As a result of these limitations, active
control systems are not used in practice as widely as passive
ones. Although passive control is widely used in practice, there
have been intensive researches in the area of active control of
structures in the recent past because of achieving higher control
of response. Further, control algorithms developed for active
control have been directly useful for developing other recent
control strategies like semi-active control. Therefore, despite
some of the obvious problems of implementation of an active
0141-0296/$ - see front matter c 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2006.04.015
S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357 347
control strategy for earthquake applications, the research in the
area of active structural control is still continuing.
Yao rst introduced formally the application of control
theory for active structural control [1]. In this system, the
control force is externally applied to the structure in order to
reduce its response. Many types of active control mechanisms
are widely reported in the literature [6]. The Kyobashi seiwa
building, the rst full-scale implementation of active control
technology, is an 11-story building with a total oor area of
423 m
2
. The control system of this building consists of two
ATMDs where the primary ATMDis used for transverse motion
and has a weight of 4 tons while the secondary ATMD has a
weight of 1 ton and is applied to reduce torsional motion [6].
Different classical and robust control algorithms have
been proposed for reducing high rise building responses [7].
The most common ones are LQR, LQG, clipped control,
bangbang control, H
2
, H

control, sliding mode control,


pole assignment, independent model space control (IMSC), and
so on [1,6,811]. But, recently, the fuzzy controller has been
used for optimization of the active control of civil engineering
structures [3,9,10,12,13].
The main advantages of the fuzzy controller are [9]:
(a) It is one of the few mathematical model free approaches to
system identication and control which makes the system
easier to design than developing an accurate mathematical
model of the structural system needed for control system
design. This can be done by using human experience and
expertise to implement the fuzzy controller.
(b) It tolerates the uncertainties of the input data from wind
or earthquake excitations and structural vibration sensors,
consequently resulting in a controller system with a
sufcient inherent robustness.
(c) The fuzzy controller has the ability to handle the non-linear
behavior of the structure caused by large displacements or
material non-linearity and damage, although in this study
no non-linearity is considered.
(d) The fuzzy controller can be adaptive by modifying its
rules or membership functions and employing learning
techniques.
Most control design methods are based on the optimiza-
tion technique of maximizing the performance of the sys-
tem through minimizing the control energy under certain con-
straints, or minimizing the structures response quantities. The
optimization procedure can be described briey as tuning the
parameters of the controller system. Most optimization meth-
ods used in control design are traditional gradient-based search
methods. Unlike the traditional optimization method, genetic
algorithms (GAs) efciently nd an optimal solution from the
complex and possibly discontinuous solution space. Genetic al-
gorithms have been applied as effective search techniques to
many elds of optimization problems [14,15]. This method has
been successfully applied to obtain gains for the optimal con-
troller, reduce the order of the feedback controller, tune the
weights of neuro-controllers, and tune and scale parameters of
the fuzzy controller [13].
Although many studies have been made on active control
of building vibrations [16,17], in a few investigations,
combination of the genetic algorithms and fuzzy logic controller
(GFLC) are considered [13,1820]. Kumar and Garg [21] in
2004 have applied the FLC system via a neural network and
genetic algorithm to control an inverted pendulum. Therefore,
researchers are still faced with challenges when investigating
the GFLC system in reduction of the structural vibrations.
In this paper, a combined application of genetic algorithms
and fuzzy logic controller (GFLC) has been presented. In order
to ensure the structural safety, which basically depends on the
building displacement response, a fuzzy logic controller (FLC)
is designed to evaluate the active control force in an ATMD
controller system based on getting the maximum reduction in
displacement response of the buildings top story. In design
of the FLC system the buildings top story displacement
and velocity responses are considered as the feedback to the
FLC. Different parameters in the FLC system [e.g., ATMD
damping ratio, frequency ratio, mass ratio, overlap parameters
in membership functions (input and output parameters), and so
on] are optimized using the GA optimizer to get the maximum
reduction in the building response. For the numerical study,
an 11-story realistic building is chosen, which is modeled
as a shear frame, and the problem is solved in state space.
The structure is considered as a linear system. As well, for
comparison, a TMD controller, and a linear quadratic regulator
(LQR) controller system are also designed to control the
building response. It is found that integration of the fuzzy logic
and genetic algorithms results in better performance of the
ATMD system in terms of response reduction.
2. Structural model
The equation of motion of a high rise building structure
subjected to a single support seismic excitation u
g
(t ), without
any control system, can be written as
[M]{ u} + [C]{ u} + [K]{u} = [M]{r} u
g
(t ) (1)
in which the n 1 vector {u} designates the relative
displacements of each story; n is number of stories; the n 1
vector {r} is the inuence vector representing the displacement
of each degree of freedom resulting from static application of
a unit ground displacement; and the n n matrices [M], [C]
and [K] represent the structural mass, damping and stiffness
matrices, respectively. The mass matrix for a high rise building
structure, with the assumption of masses lumped at oor levels,
is a diagonal matrix in which the mass of each story is sorted
on its diagonal, as given in the following:
[M] =
_
_
_
_
_
_
m
1
0 0
.
.
. m
2
0
0
.
.
.
.
.
.
0 0 . . . m
n
_

_
(2)
where m
i
is the i th story mass.
348 S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357
Fig. 1. Example realistic building model and the ATMD mounted on its top
oor.
The structural stiffness matrix [K] is developed based on the
individual stiffness, k
i
, of each story, and is given in Eq. (3).
[K] =
_
_
_
_
_
_
_
_
k
1
+ k
2
0
k
2
.
.
.
.
.
.
.
.
. k
i j
.
.
.
k
n
0 k
n
_

_
(3)
where
K
i j
=
_

_
k
i
+ k
i +1
i = j = n
k
n
i = j = n
k
i
i j = 1
k
i +1
j i = 1
0 Else.
The structural damping matrix [C] is assumed to be
proportional to the mass and stiffness matrices as [22]
[C] = a
0
[M] + b
0
[K] (4)
a
0
=
i

2
i

j

i
+
j
, b
0
=
j

2

i
+
j
(5)
in which a
0
and b
0
are the proportional coefcients;
i
and

j
are the structural modal frequencies of modes i and j ,
respectively; and
i
and
j
are the structural damping ratios for
modes i and j .
This paper focuses on the performance of the active control
system for the reduction of the high rise building structure
responses. For this purpose, as shown in Fig. 1, an active TMD
(ATMD) system is mounted on the top oor of the building. If
required, the number of ATMDs can be increased.
The tuned mass damper (TMD) is a classical engineering
control device consisting of a mass, a spring and a viscous
damper attached to a vibrating main systemin order to attenuate
any undesirable vibration. The natural frequency of a TMD is
tuned to a frequency close to one of the natural frequencies of
the main system. Thus, there are three main parameters in a
TMD system: TMD mass, TMD stiffness coefcient and TMD
damping ratio.
The equation of motion of a multi-degree-of-freedom linear
system subjected to a single seismic excitation u
g
(t ) and an
active control actions { f } acting on m ATMD systems can be
written as
[M
T
]{ u
T
} + [C
T
]{ u
T
} + [K
T
]{u
T
} =
[M
T
]{r
T
} u
g
(t ) + [D]{ f } (6)
in which the (n + m) 1 vector {u
T
} designates the relative
displacements of each story and that of each ATMD system;
square matrices [M
T
], [C
T
] and [K
T
] of size (n +m) represent
the mass, damping and stiffness matrices of the combined
buildingATMD system, respectively; the (n + m) 1 vector
{r
T
} is the inuence vector representing the displacement of
each degree of freedom resulting from static application of a
unit ground displacement; the m 1 vector { f } contains the
externally applied control forces whose locations are identied
through the matrix [D]; and m is the number of ATMD systems
applied to control the building responses. When the number
of ATMD systems is one, then the matrix [D] is changed to
a vector {D} and the vector { f } is changed to a scalar value f .
The state equations of the combined system can be written
in the standard state-space form as follows:
{

Z} = [A
1
]{Z} + [B
1
]{eq}
{d} = [C
1
] {Z}
(7)
in which the [2(n + m) 1] state vector {Z} is dened as
{Z} =
_
{u
T
}
{ u
T
}
_
. (8)
The state matrix [A
1
], input matrix [B
1
], output matrix [C
1
],
vector {eq} and output vector {d} are, respectively, given by
[A
1
] =
_
O
(n+1)(n+1)
I
(n+1)(n+1)
M
1
K M
1
C
_
(9)
[B
1
] =
_
O
(n+1)(n+1)
I
(n+1)(n+1)
_
(2n+2)(n+1)
(10)
{eq} = {r
T
} u
g
+ [D]{ f } (11)
[C
1
] =
_
0 1
n
0 0 0
0 0 0 1
2n+1
0
_
(12)
{d} =
_
u
n
u
n
_
(13)
in which O is a zero matrix and I is an identity matrix; 1
n
indicates that the value of column n is unity; u
n
and u
n
are the
displacement and velocity of the top story in the tall building,
respectively.
S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357 349
In order to design the TMD system, these parameters should
be optimized to get the maximum reduction in structural
dynamic response.
Active control systems utilize actuators to apply the external
control forces to the structure. In this study, the desired control
forces are determined and optimized through the following
methods:
(a) Instantaneous optimal control algorithms (or linear
quadratic regulator, LQR).
(b) Genetic algorithms and fuzzy logic controller (GFLC).
3. Linear quadratic regulator method (LQR)
The optimal linear quadratic regulator (LQR) method
requires that all the values of the state variables are to be
available. This algorithm is the classical one used for active and
semi-active control of structures. However, due to limitation
in the number of sensors that could be installed in the large
structures for measuring the state variables, the use of this
systemis restricted to some small size structures. In this system,
the active control force takes the form{ f } = [G]{Z}, where
[G] is a m 2(n + m) feedback gain matrix. Then, the control
design problem is to choose all the elements of the feedback
gain matrix [G] to yield a guaranteed desired behavior of the
closed-loop system. The selection of such entries is made by
minimizing a linear quadratic index chosen of the form
J =
_
t
d
0
[{Z}
T
[Q] {Z} + { f }
T
[R] { f }] dt (14)
where [0, t
d
] is the time interval of interest, and the symmetric
weighting matrices [Q] and [R] are the design parameters
that are selected to obtain the required performance. [Q] is a
2(n + m) 2(n + m) positive semi-denite matrix; and [R]
is a m m positive denite matrix. The control force { f } is
weighted in the performance index to allow regulation without
using excessive control energy.
4. Fuzzy logic controller (FLC)
Fuzzy set theory was developed by Lot Zadeh [23] in
1965 to deal with imprecise and uncertain phenomena often
presented in real-world applications. The primary difference
between fuzzy logic and traditional mathematics is that
the fuzzy set theory allows objects to have any degrees
of membership (between 0% and 100%) within a set,
while traditional mathematics requires objects to have either
0% or 100% membership. As a result, fuzzy set theory
involves terminology that is different from that of traditional
mathematics [10].
Fuzzy logic enables the use of linguistic directions as a
basis for control, generally very capable of handling systems. A
fuzzy logic controller (FLC) usually requires expert knowledge
in their construction. A fuzzy logic controller is incorporated
into a closed loop control system similar to conventional
controllers as shown in Fig. 2. The most widely used fuzzy
control inference R
i
is the ifthen rule, which can be written
Fig. 2. Block diagram of the fuzzy logic controller.
Fig. 3. FLC components.
as follows when two input data are used in their antecedent
parts [13]:
R
i
: if x
1
= A
i
, and x
2
= B
i
then y = C
i
. (15)
The basic structure of a typical FLC is illustrated in Fig. 3,
in which the components are dened as follows.
Fuzzier: The measured inputs in the control process, which
may be in the form of crisp values, would be converted into
fuzzy linguistic values using fuzzy reasoning mechanism.
Fuzzy rules: This is a collection of the expert control rules
needed to achieve the control goal.
Fuzzy interference engine: This unit is the fuzzy reasoning
mechanism, which performs various fuzzy logic operations
to infer the control action for a given fuzzy input.
Defuzzier: The inferred fuzzy control action is converted
into the required crisp control value in this unit.
The design of a fuzzy controller involves decisions about
a number of important design parameters that should be
determined before the actual control starts. These parameters
are the fuzzy sets in the rules, the rules themselves,
scaling factors in input and output, inference methods, and
defuzzication procedures.
5. Genetic algorithms and optimization of FLC
The basic principles of genetic algorithms (GAs) were rst
proposed by Holland [24]. GAs are general purpose search
algorithms which use principles inspired by natural genetics
to evolve solutions to problems [25]. The basic idea is to
maintain a population of chromosomes (representing candidate
solutions to the concrete problem being solved) that evolves
over time through a process of competition and controlled
variation. A GA starts with a population of randomly generated
chromosomes, and advances toward better chromosomes by
applying genetic operators modeled on the genetic processes
occurring in nature. The population undergoes evolution in a
350 S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357
form of natural selection. During successive iterations, called
generations, chromosomes in the population are rated for their
adaptation as solutions, and on the basis of these evaluations, a
new population of chromosomes is formed using a selection
mechanism and specic genetic operators such as crossover
and mutation. An evaluation or tness function must be
devised for each problem to be solved. Given a particular
chromosome, a possible solution, the tness function returns
a single numerical tness, which is supposed to be proportional
to the utility or adaptation of the solution represented by that
chromosome. Although there are many possible variants of
the basic GA, the fundamental underlying mechanism consists
of three operations: evaluation of individual tness; formation
of a gene pool (intermediate population) through a selection
mechanism; and recombination through crossover and mutation
operators. So, this algorithm includes representing designs as
individuals in a population, performing a selection (survival
of the ttest), and crossover (mating) of a generation of these
designs (from a mating pool) to create children, who in turn
become the population in the next, generally a binary string of
1s and 0s that represents the design parameter values (value
of the design variables for each individual). The operators are
applied by GAs, described in the following.
(1) Chromosome representation
Each design is represented by an n
0
-bit-long chromosome,
where n
0
is the sum of the length required to represent each
design variable, which can be determined by
2
(r1)
< (U L) 10
Pr
2
r
(16)
in which U and L = upper and lower bound of the design
variable, respectively; Pr = required decimal precision; and
r = required length in bits used to represent the design variable.
(2) Initial population
The GA starts from a population of chromosomes as a set
of initial designs. The initial population is chosen randomly.
Population size (the total number of chromosomes in the
population) is very important in order to select the global
optimum solution.
(3) Fitness function
The GA uses a function value for the selection of operator;
this function reects the objective and a penalty for constraint
violation. The tness function has been constrained in the
manner of a sequential unconstrained minimization technique.
For the present study, the tness function can be written as
F
GA
=
_
_
_
_
_
1 +
4

i =1
P
i
CR
i
UR
i
4

i =1
P
i
_
_
_
_
_
1
(17)
where CR
i
and UR
i
are the controlled and uncontrolled
responses of the building (here, the peak value of the buildings
top story displacement response), respectively; P
i
is the
weighting coefcient, which here is considered as the PGA of
the related earthquake; and summation is performed for four
considered earthquakes.
Fig. 4. Flowchart of genetic algorithms.
(4) Crossover
The crossover operator is used to produce two offspring
from the selected parents. To select the parents for crossover
from the new population, a random number between 0 and 1 is
generated.
If this random number is less than the probability of
crossover, then the chromosomes are randomly paired for
crossover. A single crossover site is selected randomly.
(5) Mutation
In order to maintain variability of population, operation
is also performed on certain individuals. The mutation is
performed on a bit-by bit basis, with a certain probability of
mutation. This operation is also performed with the help of
a random number between 0 and 1. If the random number
is less than the probability of mutation, then the bit under
consideration will be switched. The GA process is shown in
Fig. 4.
In this study, optimization of fuzzy logic controller design
for a seismically excited building has been discussed. To
minimize the peak value of top story displacement response
of a given building due to a given earthquake excitation, the
tness function f
GA
of the optimization strategy has only been
restrained on the displacement response reduction ratio of the
buildings top story.
In integration of GAs and a fuzzy logic controller (FLC),
there is no general rule method to construct the membership
S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357 351
functions. As a fuzzy concept, the membership functions are
dened by the skilled operators. This way is obviously not an
optimal solution. Fuzzy systems possess several parameters that
must be optimized. For example, tuning of the scaling functions
and fuzzy membership functions are some important tasks in
fuzzy system design. The parameterized scaling functions and
membership functions are adopted by the GAs according to a
tness function that species the design criteria in a quantitative
manner.
Scaling functions applied to the input and output variables
of a fuzzy system normalize the universes of discourse in
which the fuzzy membership functions are dened. Usually, the
scaling functions are parameterized by a single scaling factor or
a lower and upper bound in the case of linear scaling.
In the case of tuning the membership functions, an
individual represents the entire fuzzy rules as its chromosome
encodes the parameterized membership functions associated
with the linguistic terms. The most common shapes for the
membership functions (in genetic fuzzy systems) are triangular,
trapezoidal or Gaussian functions. The number of parameters
per membership function usually ranges from one to four,
where each parameter is binary or real coded.
Tuning the weighting coefcient of rule base: In the
Mamdani or Sugeno methods the entire rule bases are mixed
together by weighting coefcients, and the interval of each
weighting coefcient of rule base is [0 1]. Therefore, as in this
study the table of rule bases is symmetric, we shall optimize
half of these coefcients.
Determination of the coefcient of ATMD: In this study, the
ATMD has three parameters adopted by the GAs. The rst
parameter is the ratio of the ATMD mass to the total mass of the
structure, called as the mass ratio (M
ATMD
= mo M
Total
). The
second parameter is the damping ratio of ATMD (
ATMD
), and
the third is the factor, called the frequency ratio (K
ATMD
=
M
ATMD

2
).
6. Numerical study
In order to investigate the performance of the proposed
control strategy in reducing the structural responses under
earthquake loadings, an 11-story shear frame building, located
in city of Rasht in the north of Iran, is chosen as an example
problem (Fig. 1). The structure represents a typical medium-
size multistory building. The structural properties of this
building are provided in Table 1.
As shown in Fig. 1, the ATMD system is considered to be
mounted on the top oor of the building, and the active control
force is exerted utilizing a hydraulic actuator. However, in real
applications, some electronic sensors have to be installed on
this oor in order to measure the building responses needed as
the feedback into the closed-loop control system.
To investigate the effectiveness of the control system
for different disturbances, four different seismic motions are
used in the numerical simulations. These ground acceleration
records are: El Centro 1940, Takochi-Oki (Hachinohe) 1968,
Northridge 1994 and Kobe 1995 earthquakes. The absolute
peak ground accelerations (PGAs) of these earthquake records
are 0.3417, 0.2250, 0.8267 and 0.8178 g, respectively.
Table 1
Example building structural data
Stories Mass (kg) Stiffness (N/m)
1 2.15e5 4.68e8
2 2.01e5 4.76e8
3 2.01e5 4.68e8
4 2.00e5 4.5e8
5 2.01e5 4.5e8
6 2.01e5 4.5e8
7 2.01e5 4.5e8
8 2.03e5 4.37e8
9 2.03e5 4.37e8
10 2.03e5 4.37e8
11 1.76e5 3.12e8
The results of example building responses controlled by the
genetic algorithms and fuzzy logic controller (GFLC) method
are compared with those controlled by tuned mass damper
(TMD) and active LQR control methods. A brief description
of each method is given in the following.
The TMD frequency was tuned close to the rst modal
frequency of the building with a frequency ratio of . The mass
of the TMD system was chosen to be mo per cent of total mass
of the building and its damping ratio was considered to be
per cent of the critical value. These parameters are calculated
by the GA optimizer as: mo =
TMD mass
building total mass
= 3%; = 1.2;
and = 7%.
In the design of the LQR controller, the full state
feedback closed-loop system is used. Twelve displacements
and velocities of oors 111 and that of the moving mass
(TMD mass) are used in the simulation. An integrator is used
to convert the acceleration measurements to the velocity and
displacement of all oors and the moving mass. The 24 24
weighting matrix Q adopted here is a diagonal matrix with the
following structure [26]:
[Q] =
_
Q
11
0
0 Q
22
_
(18)
where the diagonal matrices Q
11
and Q
22
contain the weights
associated with the relative displacements and the relative
velocities, respectively. Any change in [Q] elements results into
a different sliding surface [26]. Therefore, referring to Ref. [26],
these matrices are chosen as
Q
11
=
1
diag(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0.15, 0.005) (19)
Q
22
=
2
diag(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0.15, 0.005) (20)
in which
1
,
2
and the weighting coefcient R (in Eq. (14))
are calculated as 0.1, 0.001, and 0.05, respectively; and the gain
matrix [G] is also obtained as
[G] = [3.02, 5.64, 8.43, 11.20, 14.00,
16.67, 19.21, 21.90, 24.00, 25.80, 24.00,
8.20, 0.20, 0.28, 0.41, 0.53, 0.64, 0.73,
0.80, 0.877, 0.93, 0.98, 0.93, 3.41].
In this study, the design of the fuzzy controller uses crisp
data directly from the model of the building. These data are
352 S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357
Fig. 5. (a) Membership functions of input variables (displacement and
velocity). (b) Membership functions of output variable (Active control force).
then converted into linguistic or fuzzy membership functions
through the fuzzication process. The controller is designed
based on two input variables (displacement and velocity of the
buildings top story), each having ve trapezoidal membership
functions (Fig. 5a), and one output variable (active external
control force) with seven triangular membership functions
(Fig. 5b). The fuzzy input and output variables membership
function abbreviations used to dene the fuzzy space are: LP =
Large and Positive; P = Positive; Z = Zero; N = Negative; LN
= Large and Negative (for input variable); and PL = Positive
and Large; PM = Positive and Medium; PS = Positive and
Small; ZR = Zero; NL = Negative and Large; NM = Negative
and Medium; NS = Negative and Small (for output variable).
Further, in order to map the FLC input variables onto
its output variable, a fuzzy associative memory (FAM) is
developed, which is shown in Table 2.
The fuzzy controller will couple the point-valued MAX
MIN fuzzy inference engine product rule to combine the mem-
bership values for each rule (Mamdani type), and the center of
area (COA) defuzzier scheme to obtain the crisp value.
In order to generalize the trapezoidal membership functions
used for input variables, the parameters a
i
, b
i
, c
i
and d
i
(shown
Table 2
Fuzzy associative memory (FAM) of the fuzzy controller
u
11
u
11
(Velocity)
(Displacement) LN N Z P LP
LP NS NS NM NL NL
P NS NM NM NM NL
Z PS ZR ZR ZR NS
N PL PM PM PM PS
LN PL PL PM PS PS
Table 3
Weighting coefcients of the FAM obtained by the GA
u
11
u
11
(Velocity)
(Displacement) LN N Z P LP
LP
NS
0.1
NS
0.9
NM
0.9
NL
0.9
NL
0.1
P
NS
0.9
NM
0.8
NM
0.6
NM
0.8
NL
0.9
Z
PS
0.7
Z R
1
Z R
1
Z R
1
NS
0.7
N
PL
0.9
PM
0.8
PM
0.6
PM
0.8
PS
0.9
LN
PL
0.1
PL
0.9
PM
0.9
PS
0.9
PS
0.1
in Fig. 5a) are considered, in which using the GA optimizer,
the performance of the designed FLC system can be optimized.
The corresponding parameters for the output variable, as shown
in Fig. 5b, are considered as a
0
and b
0
.
The maximum displacement response of the top story of
the example building due to earthquake excitation is taken
as objective of the optimization problem, which should be
minimized. Therefore, this maximum displacement response is
used in the tness function f
GA
given by Eq. (17).
The parameters that should be specied prior to the solution
of the optimization problem are the properties of the structure
and the earthquake excitation. The design variables of the
FLC are the parameters of the input and output membership
functions described earlier. Due to symmetry, parameters of
only half of the input and output membership functions have
been considered as design variables.
The weighting coefcients of fuzzy associative memory
rules are the other parameters that must also be optimized. The
interval of the weighting coefcient of each rule is [0 1]. Here
also due to symmetry, only half of the weighting coefcients of
fuzzy associative memory rules have been considered as design
variables. The parameters of input and output membership
functions; and the scale factors of the building responses are
calculated by the GA optimizer as: ai = 0.125, bi = 0.333,
ci = 0.33, ao = 0.125, bo = 0.35; and scale factors of
displacement, velocity and force are 0.08, 0.7 and 5% w
T
;
respectively, where w
T
is the building total weight. Moreover,
the values of weighting coefcients of the FAM obtained by the
GA are given in Table 3.
In the GA tournament some GA operator parameters
are very important in improving the GA tournament. These
parameters are the number of initial population, crossover
probability and mutation probability. In this study, the number
S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357 353
Fig. 6. Comparison of controlled displacement of the example building
calculated by the TMD, LQR, and GFLC systems with the uncontrolled ones
for the El Centro earthquake.
of initial population is taken to be about twenty. The crossover
probability and mutation probability are taken as 80% and 85%,
respectively.
The constraints of convergence can be written in this GA
tournament as follows. All parameters of the best chromosome
in the end course equate with all parameters of the best
chromosome before the end course.
The mass ratio (mo), damping ratio () and frequency ratio
() of the ATMD are obtained by the GA as: 3%, 7% and 1.0,
respectively.
Table 4
Comparison of the effectiveness of the different controller systems used in this
study (for the El Centro earthquake)
Building
oor
Maximum uncontrolled
response (m)
Controlled to
uncontrolled response
ratio (reduction ratio)
[5]/[4]
TMD LQR GFLC
[1] [2] [3] [4] [5] [6]
1 0.019 0.68 0.49 0.45 0.92
2 0.039 0.64 0.46 0.4 0.87
3 0.057 0.65 0.47 0.4 0.85
4 0.074 0.65 0.47 0.38 0.81
5 0.09 0.64 0.48 0.38 0.79
6 0.10 0.67 0.50 0.39 0.78
7 0.12 0.62 0.48 0.36 0.75
8 0.13 0.64 0.46 0.36 0.78
9 0.14 0.67 0.48 0.35 0.73
10 0.14 0.67 0.50 0.36 0.72
11 0.147 0.673 0.49 0.35 0.71
Fig. 7. Comparison between the active control forces needed in the LQR and
GFLC systems for the El Centro earthquake.
The results of controlled displacement response of the
example building top story due to the El Centro earthquake
calculated by the TMD, LQR, and GFLC systems are compared
with the corresponding uncontrolled ones in Fig. 6, and Table 4.
The corresponding active control force is also shown in Fig. 7.
354 S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357
Table 5
Comparison of the effectiveness of the different controller systems used in this
study (for the Hachinohe earthquake)
Building
oor
Maximum uncontrolled
response (m)
Controlled to
uncontrolled response
ratio (reduction ratio)
[5]/[4]
TMD LQR GFLC
[1] [2] [3] [4] [5] [6]
1 0.014 0.86 0.79 0.57 0.72
2 0.028 0.86 0.75 0.61 0.81
3 0.04 0.88 0.80 0.60 0.75
4 0.053 0.87 0.77 0.57 0.74
5 0.064 0.86 0.78 0.56 0.72
6 0.074 0.86 0.78 0.54 0.69
7 0.085 0.86 0.76 0.54 0.71
8 0.094 0.86 0.755 0.53 0.70
9 0.10 0.89 0.76 0.53 0.70
10 0.11 0.86 0.72 0.50 0.69
11 0.11 0.90 0.75 0.52 0.69
As can be seen from the gures and table, the response
reduction ratio (ratio of the controlled to uncontrolled response)
for maximum displacement of the top oor of the 11-story
example building is about 67%, 49%, and 35% for the TMD,
LQR, and GFLC systems, respectively. Therefore, it is seen that
the designed GFLC system is more effective than the two other
systems in view of reducing the displacement response of the
example building.
Furthermore, from Fig. 7, it can be seen that the active
control force needed for obtaining the above reductions in the
GFLC system is more than that of the LQR system. In this
case, the maximum control force for the GFLC system is about
67 tons, while that of the LQR is about 42 tons. However,
29% more displacement reduction in the GFLC system in
comparison with the LQR system results in some more
reduction in member size of the designed building, which in
turn is signicant from an economical standpoint. On the other
hand, the LQR system needs the values of all state variables
to be available, which practically means that electronic sensors
must be installed in all stories of the building. Therefore,
selection of the control strategy among the above three systems
is a compromise between the economical advantages which can
be achieved from the reduction of structure member sizes as
a result of the application of control system, versus the extra
expenses associated with installing the corresponding control
systems.
The effectiveness of these control systems in reducing the
response of the example building due to other three earthquakes
is also shown for comparison in Figs. 810 and Tables 57.
Almost the same behavior as for the El Centro earthquake can
be observed for these earthquakes too.
Moreover, it is seen from Tables 57 that the Kobe
earthquake causes the maximum displacement (about 50 cm) at
the top oor of the example building, which is a very high value
in comparison with the other earthquakes. This is expected due
to fact that the PGA of the Kobe earthquake is very high (about
0.8178 g) and almost remains constant for a long period of time
in comparison with other three earthquakes.
Fig. 8. Comparison of controlled displacement of the example building
calculated by the TMD, LQR, and GFLC systems with the uncontrolled ones
for the Hachinohe earthquake.
In making the nal judgment on the effectiveness of the
three proposed control systems, the average response reduction
ratios (controlled to uncontrolled displacement ratio) for all
four earthquakes on all 11 stories of the example building are
shown in Table 8. It is seen from the table that in general the
three TMD, LQR, and GFLC systems are capable of reducing
the maximum displacement of the building in each story to
about 82.3%, 71.6% and 59.4% of the uncontrolled response,
respectively. Therefore, on average the designed GFLC system
S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357 355
Fig. 9. Comparison of controlled displacement of the example building
calculated by the TMD, LQR, and GFLC systems with the uncontrolled ones
for the Northridge earthquake.
is more effective than the others, but at the expense of requiring
larger control forces.
7. Conclusions
This paper focuses on the combined application of genetic
algorithms and a fuzzy logic controller (GFLC) for the
reduction of the high rise building responses subjected to
earthquake excitations using the ATMD control system. A
fuzzy logic controller (FLC) is designed for evaluating the
Fig. 10. Comparison of controlled displacement of the example building
calculated by the TMD, LQR, and GFLC systems with the uncontrolled ones
for the Kobe earthquake.
active control force in an ATMD controller system based on
getting the maximum reduction in displacement response of
the buildings top story. In the design of the FLC system the
buildings top story displacement and velocity responses are
considered as the feedback to the FLC. Different parameters
in the FLC system are optimized using the GA optimizer
to obtain the maximum reduction in the building response.
For the numerical study, an 11-story realistic shear building
is chosen and the problem is solved in state space. As well,
for comparison, a TMD controller, and a linear quadratic
356 S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357
Table 6
Comparison of the effectiveness of the different controller systems used in this
study (for the Northridge earthquake)
Building
oor
Maximum uncontrolled
response (m)
Controlled to
uncontrolled response
ratio (reduction ratio)
[5]/[4]
TMD LQR GFLC
[1] [2] [3] [4] [5] [6]
1 0.046 0.86 0.72 0.67 0.93
2 0.088 0.91 0.72 0.66 0.91
3 0.123 0.89 0.89 0.65 0.73
4 0.15 0.93 0.73 0.66 0.90
5 0.18 0.89 0.72 0.66 0.91
6 0.194 0.92 0.77 0.67 0.87
7 0.204 0.93 0.83 0.68 0.82
8 0.210 0.95 0.86 0.68 0.79
9 0.22 1.00 0.86 0.71 0.82
10 0.23 1.00 0.91 0.73 0.80
11 0.23 1.00 0.95 0.74 0.78
Table 7
Comparison of the effectiveness of the different controller systems used in this
study (for the Kobe earthquake)
Building
oor
Maximum uncontrolled
response (m)
Controlled to
uncontrolled response
ratio (reduction ratio)
[5]/[4]
TMD LQR GFLC
[1] [2] [3] [4] [5] [6]
1 0.06 0.82 0.83 0.77 0.93
2 0.12 0.82 0.84 0.77 0.91
3 0.18 0.83 0.80 0.73 0.91
4 0.24 0.83 0.80 0.75 0.94
5 0.29 0.82 0.83 0.79 0.95
6 0.34 0.85 0.79 0.76 0.96
7 0.39 0.85 0.79 0.75 0.95
8 0.43 0.84 0.80 0.78 0.97
9 0.46 0.85 0.79 0.77 0.97
10 0.48 0.85 0.78 0.75 0.96
11 0.50 0.84 0.78 0.74 0.95
Table 8
The average reduction (for 11 stories) in maximum displacement response of
the example building
Earthquake excitation Average response reduction
TMD LQR GFLC
El Centro earthquake 65% 48% 38%
Hachinohe earthquake 87% 76.5% 55%
Northridge earthquake 93.5% 81.5% 68%
Kobe earthquake 84% 80% 76%
Total average 82.3% 71.6% 59.4%
regulator (LQR) controller system are also designed to control
the building response.
From the numerical results of the study, it is found that:
1. In viewof the building response reduction, the GFLCsystem
is more effective than the traditional controller systems.
2. The external control force needed to reduce the building
response in the GFLC system is more than that of the LQR
system, while the obtained reduction by GFLC is more than
that of the LQR.
3. The optimum values of the ATMD mass, damping, and
frequency ratios are obtained (by the GA optimizer) to be
about 3%, 7%, and 1.0, respectively; while those of the TMD
control system are obtained to be about 3%, 7%, and 1.2,
respectively.
4. The GA optimizer is a very powerful tool to optimize
the FLC system by considering simultaneously as many
parameters as desired.
References
[1] Datta TK. Control of dynamic response of structures. In: IndoUS
symposium on emerging trends in vibration and noise engg, 1996. p.
1820.
[2] Randa R, Soong TT. Parametric study and simplied design of mass
dampers. Eng Struct 1998;20(3):193204.
[3] Aldawod M, Samali B, Naghady F, Kwok Kenny CS. Active control of
along wind response of tall building using a fuzzy controller. Eng Struct
2001;23:151222.
[4] Soong TT, Hanson R. Recent developments in active and hybrid control
research in US. In: Int. workshop on structural control. 1993. p. 48390.
[5] Sarbjeet S, Datt TK. Open closed loop linear control of building frames
under seismic excitation. J Struct Eng, ASCE 1998;124(1):4351.
[6] Yang JN. Recent advances in active control of civil engineering structures.
Probab Eng Mech 1988;3(4).
[7] Sponcer BF, Soong TT. New applications and development of active,
semi-active and hybrid control techniques for seismic and non- seismic
vibration in the USA. In: Proceeding of international Post-SMiRT
conference seminar on seismic isolation. 1999. p. 235.
[8] Samali B, Al-Dawod M. Performance of a ve-storey benchmark model
using an active tuned mass damper and a fuzzy controller. Int J Eng Struct
2003;25:1597610.
[9] Ha QP. Active structural control using dynamic out put feed back sliding
mode. In: Proc. 2001 Australian conference on robotics and automation
Sydney. 2001. p. 145.
[10] Symans MD, Kelly W. Fuzzy logic control of bridge structures using
intelligent semi-active seismic isolation systems. Earthq Eng Struct Dyn
1999;28:3760.
[11] Sarbjeet S, Datta TK. Nonlinear sliding mode control of seismic response
of building frames. J Eng Mech, ASCE 2000;126(6):3407.
[12] Lin Y, Cheng C, Lee C. A tuned mass damper for suppressing the coupled
exural and torsional buffeting response of long-span bridges. Eng Struct
2000;22:1195204.
[13] Ahlawat AS, Ramaswamy A. Multi objective optimal structural vibration
control using fuzzy logic control system. Struct Eng November 2001;
127(11):13307.
[14] Goldbereg DE. Genetic algorithm in search, optimization and machine
learning. Reading (MA): Addition-Wesley; 1989.
[15] Kim Y-J, Ghaboussi J. A new method of reduced order feedback control
using genetic algorithms. Earthq Eng Struct Dyn 1999;28:193212.
[16] Agrawal AK, Yang JN. Effect of xed time delay on stability and
performance of actively controlled civil engineering structures. Earthq
Eng Struct Dyn 1997;26:116985.
[17] Soon TT. Active structural control: Theory and practice. Essex (England):
Long-man Scientic and Technical; 1990.
[18] Rojas I, Bernier JL, Rojas FJ, Puntonel CG. The synergy between
multideme genetic algorithms and fuzzy systems. In: ESANN2001
proceeding European symposium on articial neural networks bruges.
2001. p. 199204.
[19] Fatehi A, Abe K, Lucas C. Challenges on the design of the fuzzy logic
controller by the genetic algorithms. In: 2nd Int. workshop on intelligent
systems of the 4th joint conf. on information systems, 1998.
S. Pourzeynali et al. / Engineering Structures 29 (2007) 346357 357
[20] Cordon O, Herrera F, Villar P. Generating the knowledge base of a fuzzy
rule-based system by the genetic learning of the data base. IEEE Trans
Fuzzy Systems 2001;9(4).
[21] Kumar M, Garg DP. Inteligent learning of fuzzy logic controllers via
neural network and genetic algorithm. In: Proceeding of 2004 JUSFA,
JapanUSA symposium on exible automation. p. 1921.
[22] Clough RW, Penzien J. Dynamics of structures. 2nd ed. New York: Mc
Graw-Hill; 1993.
[23] Zadeh LA. Fuzzy sets. Information Control 1965;8:33853.
[24] Holland JH. Adaptation in natural and articial systems. MIT press; 1975.
[25] G omez-Skarmeta F, Jim enez F. Fuzzy modeling with hybrid systems.
Fuzzy Sets and Syst 1999;104:199208.
[26] Singh MP, Matheu EE. Active and semi-active control of structures under
seismic excitation. Earthq Eng Struct Dyn 1997;26:193213.

Das könnte Ihnen auch gefallen