EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2002; 31:955–976 (DOI: 10.1002/eqe.132)

Optimal placement of dampers for passive response control

Mahendra P. Singh1; ∗; † and Luis M. Moreschi 2; ‡

1Department of Engineering Science and Mechanics; Virginia Polytechnic Institute and State University;
Blacksburg; VA 24061-0219; U.S.A.
2 Bechtel Power Corp.; Frederick; Maryland; U.S.A.

SUMMARY
The e6ectiveness of viscous and viscoelastic dampers for seismic response reduction of structures is
quite well known in the earthquake engineering community. This paper deals with the optimal utilization
of these dampers in a structure to achieve a desired performance under earthquake-induced ground
excitations. Frequency-dependent and -independent viscous dampers and viscoelastic dampers have been
considered as the devices of choice. To determine the optimal size and location of these dampers in the
structure, a genetic algorithm is used. The desired performance is de@ned in terms of several di6erent
forms of performance functions. The use of the genetic approach is not limited to any particular form of
performance function as long as it can be calculated numerically. For illustration, numerical examples
for di6erent building structures are presented showing the distribution and size of di6erent dampers
required to achieve a desired level of reduction in the response or a performance index. Copyright
? 2002 John Wiley & Sons, Ltd.

KEY WORDS: passive response control; dampers

INTRODUCTION

It has been a common practice to design structural systems for the forces and deforma-
tions that are imposed by a design level earthquake. In the current design practice, structural
elements are permitted to yield under the design forces to dissipate the earthquake energy
without collapse. To accommodate large deformations associated with yielding, structural el-
ements are detailed to provide adequate ductility. This design philosophy has served well
but because yielding is associated with permanent deformations and damage, the methods are
being sought and used to reduce or eliminate yielding and the associated damage.

∗ Correspondence to: Mahendra P. Singh, Department of Engineering Science and Mechanics, Virginia Polytechnic
Institute and State University, Blacksburg, VA 24061-0219, U.S.A.
† E-mail: [email protected]
‡ Formerly a graduate student at the Department of Engineering Science and Mechanics, Virginia Tech.
Contract=grant sponsor: National Science Foundation; contract=grant numbers: CMS-9626850, CMS-9987469.

Received 10 April 2001

Revised 1 August 2001
Copyright ? 2002 John Wiley & Sons, Ltd. Accepted 7 August 2001
956 M. P. SINGH AND L. M. MORESCHI

Passive and active structural response control devices and schemes are being considered to
control the structural response. For some structures, base isolation is appropriate for reducing
the response, and has been used in many buildings. For others, energy dissipation devices
such as viscous dampers, viscoelastic dampers, friction dampers and yielding devices have
been considered and installed as energy sinks to reduce the structural response and energy
dissipation demand on the structural members. Although the e6ectiveness of these devices
for reducing the seismic and wind response is con@rmed through analytical and experimental
studies, the development of the methods for optimal utilization of these devices is still an
important research issue. It is well known that the installation of an energy dissipation device
or a damper will cause a reduction in the response. Some response quantities will be reduced
more than others. The damping capacity of a device (rated damping force of the device)
and where and how many are placed on a structure will have a signi@cant e6ect on the
ability to reduce the response and to achieve the desired design objectives. In absence of
a better rationale, it may be convenient to distribute the dampers uniformly in all storeys
of a building. However, such a distribution may not be the most e6ective arrangement. There-
fore, the optimal design studies for supplementary damping devices have been of continued
interest in civil engineering for some time. In this study, we use the viscous and viscoelastic
dampers as the energy dissipation devices of choice, and obtain their optimal placement and
size to achieve a desired reduction in the response.
In the literature, researchers have suggested di6erent practical schemes for optimal design
of dampers. Ashour and Hanson [1] suggested to place dampers at the locations that will
maximize the damping ratio of the fundamental mode, as this mode is often the most dominant
mode in multi-storey building structures. Zhang and Soong [2] and Shukla and Datta [3] have
used a sequential optimization procedure to determine the optimal location of viscoelastic
dampers in multi-storey building structures. Wu et al. [4] extended the sequential approach
applications to three-dimensional building structures. Natke and Soong [5] used the topological
optimization, and Milman and Chu [6] utilized the simulated annealing approach, to search
for the optimal placement of dampers. Gluck et al. [7] utilized the optimal solution for the
linear quadratic regulator problem to obtain the optimal damping matrix, which was then
used to determine the damper coeMcients in di6erent storeys. Takewaki [8] and Takewaki
et al. [9] have used a gradient-based approach to search for the optimal solution that would
minimize a desired system transfer function. Recently, Singh and Moreschi [10] have also
used a gradient-based approach to obtain the optimal distribution of classical viscous dampers
for a continuous description of the performance function.
Another versatile and Nexible option to solve such an optimal design problem is to use the
genetic algorithms. This approach is especially suitable for problems where the performance
index is not a continuous function of the design variables and=or the variable design space is
discrete. Also, in practical applications, the damper capacity and sizes may be @xed because
of commercial and manufacturing constraints. Thus, the option to chose any capacity damper
may not be available; rather, one must choose from those that are available commercially. The
optimal placement of a given number of such @xed size dampers can be viewed as a combi-
natorial optimization problem since the design space—that is, the space de@ned by variables
of the number of devices and their locations within a structure—is discrete. In principle, the
optimal solution of such a discrete problem can be found by an exhaustive or enumerative
search of every possible combination of dampers locations. However, practical implementation
of this search is impossible due to the high number of feasible design solutions. For example,

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 957

for optimal placement of 50 di.erent damper devices in 10 possible storeys of a building

structure, one will have to examine 1050 possible combinations. For m identical devices to be
placed in n possible storeys, there are (m+n − 1)!=m!(n − 1)! di6erent combinations. For m = 50
and n = 10, this number is 1:25 × 1010 . Searching each possible combination is obviously a
daunting task even for the current computing facilities and, therefore, a more systematic and
an eMcient approach must be used. In this paper, we use the genetic algorithms to obtain
the optimal solution of our problem. To demonstrate the application of this approach and to
illustrate the bene@ts of adopting an optimal design, several numerical examples are presented.

PROBLEM DEFINITION

The equations of motion of a building with added damping devices subjected to base motion
can be written as

nl
S + Cs u̇(t) + K s u(t) + rj nj Pj (t) = −MEf(t)
Mu(t) (1)
j=1

where M; K s and C s represent, respectively, the N × N mass, structural sti6ness and inherent
structural damping matrices, f(t) is an l-dimensional vector representing the seismic excitation,
E is a N × l matrix of ground motion inNuence coeMcients, u(t) is the N -dimensional relative
displacement vector with respect to the base and a dot over a symbol indicates di6erentiation
with respect to time. The contribution of the force caused by a single damper Pj to a degree
of freedom is considered through the inNuence vector rj . The number of identical dampers
installed at the jth location is denoted by nj , and nl is the number of possible locations for
a damper in the structure. Given an energy dissipation device with predetermined mechanical
properties, it is of interest to determine the optimal number and locations of such devices to
achieve a desired reduction in the structural response. Such a question can be answered by
posing this design problem as an optimization problem.
The reduction of a desired response quantity or the performance expected from a structure
could be expressed in terms of the desired value of a performance index or function, f( ). In
terms of a performance index, the optimal design problem at hand can be stated as

minimize f[R(n; t)] (2)

n

nl
subject to nj = n T (3)
j=1

where n is the vector of design variables nj , n T is the total number of dampers to be placed in
a structure and R(n; t) is the structural response vector such as the Noor accelerations, shears,
etc. on which the performance function depends.
The performance index in Equation (2) could be stated in di6erent forms, depending on the
objectives of the design. For example, it could be de@ned in terms of a single response quantity
of interest, such as the acceleration of a Noor or the base shear or the over turning moment, if
the objective is to reduce such a response quantity. It could also be de@ned in terms of several
similar response quantities such as the sums of the squares of the Noor accelerations, or the

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
958 M. P. SINGH AND L. M. MORESCHI

squares of the sums of the inter-storey drifts, where the interest would be to reduce these
quantities for the entire structure. For the performance-based seismic design of structures,
where the structure is expected to perform in a certain desired manner at di6erent levels
of excitation intensities, the performance index could take forms that are more complex. For
example, it could be de@ned in terms of the life cycle cost estimates for the building structure,
considering di6erent levels and types of hazards a building structure can experience. The
objective may be to minimize the life cycle cost by altering the structure design parameters,
including the parameters of the protective devices such as dampers and other control devices.
We do not intend to go into the details of the formation of these performance indices, but
just to indicate that they can take di6erent forms serving di6erent objectives. The forms of
performance indices that are chosen in this study are de@ned later when the numerical results
are presented. The primary objective of this study is to obtain the best combination of nj ,
that is the number of dampers at locations, such that the performance index of Equation (2)
is minimized.

OPTIMAL SOLUTION APPROACH—GENETIC ALGORITHMS

The optimization problem stated above could be solved by using an appropriate gradient-based
approach if the performance function is di6erentiable and its gradients can be conveniently
calculated. Depending upon the initial solution used, however, a gradient approach might
converge to a local minimum solution, if there is more than one such solution. This may
not be a special problem as a local minimum will still be better than a non-optimal solution.
If desired, the local minimum solutions may be improved by starting with several di6erent
initial solutions. Genetic algorithms essentially do this but without computing any gradients.
We propose to use this alternative approach to solve our problem.
In the past few years, genetic algorithms have enjoyed a growing interest in the combi-
natorial optimization community. They have been successfully applied to a wide range of
engineering applications. They o6er robust search and optimization techniques that are based
on the principles of natural biological evolution where the stronger individuals are likely to
be the winners in a competing environment. In their search for the best design solution, they
operate on a population of potential solutions (designs) simulating evolution by means of
random genetic changes that produce successively better approximations ultimately leading to
a @nal design solution. A strong advantage of the genetic algorithm in the context of our
problem is that they can be used with any type of performance functions and with linear
or non-linear systems as long as the system can be analysed and performance function can
be calculated. They are also less likely to converge to a locally optimum solution. They are
especially suitable for the problems with performance functions that are not di6erentiable or
are noisy such that their gradients change drastically over the range of the design variables
and thus they have several local optima. Also as mentioned before, the use of this discrete
optimization approach may become necessary if one is restricted to use only those dampers
that are commercially available. On the other hand, a disadvantage of this approach, espe-
cially for very large problems, is that the computational e6ort could be high compared to the
gradient-based solutions.
The basic principles of genetic algorithms were @rst proposed by Holland [11] in 1975.
Since then, many di6erent applications of genetic algorithms have been explored. Besides the

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 959

book by Holland, there are now several other books available on this subject; for example,
Goldberg [12], Davis [13], Gen and Cheng [14], and others. Signi@cant applications of this
technique in structural engineering have been made by Chan [15], Cheng and Li [16], Furuya
et al. [17], Kim and Ghaboussi [18], and Singh and Moreschi [19; 20]. In the context of
placement of damping devices in aerospace applications, the papers by Rao and Pan [21],
Onoda and Hanawa [22], and Furuya and Haftka [23] are quite relevant. Hadi and Ar@adi
[24] have also utilized genetic algorithms for optimum design of vibration absorbers. A brief
outline of the approach, as it is applied to the problem of optimal placement of dampers in
a building structure, is given in this section.
To start a genetic algorithm search, an initial population of feasible designs (individuals)
is randomly generated. In the context of placement of damping devices, a feasible design
is a possible distribution of the damping devices at di6erent Noors. In the terminology of
the genetic evolution, this arrangement represents a chromosome, and the placement of a
particular device on a Noor represents a gene. There is no set rule for choosing the size of
the population. However, a smaller size population is likely to take longer (more generations)
to converge than a larger population. A larger population, on the other hand, would mean
more number of performance evaluations in each generation. Each design in the population is
next evaluated by calculating its performance function or index. The design with the highest
performance index is considered the best (@ttest individual).
The next step is to create a new population of next generation by the basic genetic oper-
ators of mating by crossover and mutation of genes of individuals in the current population.
This essentially modi@es the genetic information contained in the current population to create
a new population. The mating is accomplished by pairing of two individuals (two designs).
This pairing is done according to a roulette-wheel selection scheme in which the individuals
with high performance functions are given higher chances for mating with the best. The paired
individuals generate a new population of new individuals (o6spring) through a crossover of
the genes (device locations). The crossover consists of an interchange of the parental genes
according to some simple scheme. Several variations of the crossover schemes are prevalent.
In our study, the simplest single-point crossover scheme was used. Usually, a large fraction
of the population (95–98 per cent) is used for mating and crossover. A small fraction of the
generated population is also mutated to introduce new designs in the population. This alters
the genetic representation of a few selected individuals in which a randomly selected gene (a
damping device location) in the chromosome (an arrangement of devices) is changed to
take on a new value from the admissible set of values. The fraction of the population
to be mutated is generally kept low (about 5 per cent) to avoid too many o6spring los-
ing the resemblance to their parents, and thus losing their ability to learn from their gene
history. Often, the newly generated population is also subjected to some elitist selection
scheme to retain the best individual characteristics of the previous generation. In our nu-
merical study, this elitist scheme consisted of dropping the last ranked individual and replac-
ing it by the best individual from the parental population. The recently formed population
is subject to rank ordering, pairing, and mutation to generate a new population as before.
This process is repeated for a number of cycles (generations) until no further improvement
is observed in the best individual in the subsequent generations. The best designs of the
successive populations gradually converge to the optimal solution. For further discussion of
the operation of the genetic algorithms, the reader is referred to some of the references cited
earlier.

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
960 M. P. SINGH AND L. M. MORESCHI

PERFORMANCE EVALUATION

The majority of e6ort in optimal analysis with genetic algorithms is spent in the calculation of
the performance function. To evaluate a performance function, one must analyse the structure.
The analysis procedure will depend on whether the structure behaves linearly or goes into
the inelastic range. Some damping devices can also introduce non-linearity in the structural
system. Here, however, we will focus on the use of linear viscous and viscoelastic dampers
to control the structural performance. Normally, the structural systems designed according to
the current codes will go in the inelastic range when subjected to the design level ground
motion. However, to avoid permanent damage associated with inelastic design, the added
damping devices may be designed such that the main structural elements remain elastic. In
the following development, it will be assumed that such is, indeed, the case, and we will
take the full advantage of this linearity in selecting a response analysis scheme. To include
the stochastic nature of the seismic input in the analysis, we propose to employ a modal
analysis-based random vibration approach for the seismic input de@ned in a stochastic form,
or a response spectrum approach for input de@ned by design response spectra. These inputs
enable us to consider an ensemble of possible seismic motions collectively in the optimality
analysis. The details of the modal analysis approaches to be used, however, depend upon
the models employed to characterize the damping devices as they change the form of the
equations of motion. In the following, therefore, we describe the force–deformation models
and associated equations of motions for the classical viscous Nuid dampers, viscoelastic solid
dampers, and viscoelastic Nuid dampers that we propose to use in this study.

Classical viscous /uid dampers

For such devices, the damper force can be simply de@ned in terms of the deformation velocity
of the damper as follows:
Pj (t) = cU̇j (t) = crjT u̇j (t) (4)
where c represents the damping coeMcient value of the selected device. In the last part of this
equation, the damper velocity is expressed in terms of the inNuence vector rjT and the relative
velocity vector u̇j of the structure. Substitution of Equation (4) into Equation (1) gives the
following equation of motion:
 

nl
T
MuS (t) + Cs + rj nj crj u̇(t) + K s u(t) = −MEf(t) (5)
j=1

The installation of these dampers thus only modi@es the damping matrix of the structural sys-
tems. Because of the arbitrary addition of the dampers, the system will become non-classically
damped even if it was classically damped initially. Such a system can be conveniently anal-
ysed by the modal analysis approach for inputs de@ned by time histories, random descriptions
or by response spectra. The state vector approach developed by Singh [25] and Maldonado
and Singh [26] can be used for such an analysis. Since the system matrices are symmetric,
the following symmetrical state vector formulation can be used:
 
0
A s ż(t) + B s z(t) = −Ds f(t) (6)
E

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 961

where z(t) is the 2N -state vector consisting of the relative velocity vector u̇(t) in its @rst N
elements and the relative displacement vector u(t) in the remaining N elements. The symmetric
system matrices A s ; B s and Ds of dimension 2N × 2N are de@ned as
     
0 M −M 0 0 0
As = ; Bs = ; Ds = (7)
M C 0 Ks 0 M

Solid viscoelastic dampers

These dampers dissipate energy through shear deformation of a polymeric material. Besides
damping, these devices also add sti6ness to the structure. As suggested by FEMA-273 [27],
the force–deformation relationship of solid viscoelastic devices may be modelled by a classical
Kelvin model as follows:
Pj (t) = kUj (t) + cU̇j (t) (8)
The coeMcients in Equation (8) can be de@ned in terms of the storage and shear modulii of
the viscoelastic material, and the physical dimensions, such as the shear area and thickness, of
the material used in the damper. Although these coeMcients are usually frequency dependent,
it has been shown [28] to be acceptable to use frequency-independent properties for practical
applications. Substitution of Uj (t) = rjT u(t) in terms of the structural displacement vector in
Equation (8) and then substituting into Equation (1), we obtain
   

nl
T
nl
T
S + Cs + rj nj crj u̇(t) + K s + rj nj krj u(t) = −MEf(t)
Mu(t) (9)
j=1 j=1

Thus, the viscoelastic devices modify both the sti6ness and damping matrices of the system.
However, since the system matrices are still symmetric, a self-adjoint state vector formulation
and solution approach as mentioned earlier can still be used to analyse this non-classically
damped system.

Viscoelastic /uid dampers

The cyclic response of Nuid viscoelastic devices is shown to depend on the frequency of
deformation. This characteristic can be adequately captured by the classical Maxwell model
in which a dashpot and a spring are joined in series. In this model, the force–deformation
relationship can be de@ned by a @rst-order di6erential equation. For the jth damping element,
it can be written as follows:
c
Pj (t) + Ṗj (t) = cU̇j (t) (10)
k
where k is the sti6ness of the selected device at in@nitely large frequency and c is the damping
coeMcients at zero frequency. The ratio  = c=k is referred in the literature as the relaxation
time constant. For a low frequency of deformation, the device exhibits a purely viscous
behaviour, but as the frequency increases, the damper also develops signi@cant sti6ness. The
classical viscous damper is a special case of this model with k approaching in@nity. Other
combinations of linear springs and dashpots have also been proposed to model linear damping
devices that exhibit sti6ening at very low frequencies such as bituminous Nuid dampers.
Although these other mechanical models have not been used in this study, they are mentioned

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
962 M. P. SINGH AND L. M. MORESCHI

here to indicate the di6erent levels of re@nement one can incorporate for modelling these
devices.
It is mentioned that the force–deformation relationship given by Equation (10) is also
encountered when a classical viscous damper is used within a deformable bracing, as bracing
sti6ness and dashpot elements are in series. If the bracing is relatively very rigid, its e6ect
may be ignored and Equation (4) can be used. In a diagonal bracing, the e6ect of the brace
angle with the horizontal must also be considered to de@ne the force applied by the damper
to the structure.
Substituting for the deformation of the damper element Uj (t) in terms of the structural
displacement vector as Uj (t) = rjT u(t), the Equation (10) can be written as

Ṗj (t) + Pj (t) − cj rjT u̇(t) = 0 (11)

Each damping device will be governed by a similar equation. These device equations can
be combined with the equation of motion (1) in the state-space format. However, unlike
the state-space equations for the two previous cases, these equations will not be symmetric.
Herein, these are expressed in the standard state-space format as follows:
ż(t) = Az(t) + Bf(t) (12)
where the description of the dynamic behaviour of the overall structural system is done in
terms of (2N + nl ) × (2N + nl ) non-symmetric system matrix A de@ned as
 
−M−1Cs −M−1 K s LN ×nl

A =  IN ×N 0N ×N 0N ×nl   (13)
u̇(nl ×N ) u(nl ×N ) P(nl ×nl )
The state vector z(t) now includes the device force vector P(t) and is of dimension (2N +nl ).
It is de@ned as
 
 u̇(t) 
 
z(t) = u(t) (14)

 

P(t)
The inNuence matrices L and B specifying, respectively, the locations and number of linear
passive devices and l-components of the seismic excitation in the state-space are given by
 

−EN ×l 
 
L = −n1 M−1 r1 · · · − nnl M−1 rnl N ×nl ; B = 0N ×l (15)

 

0nl ×l
The matrices M; C; K; E and vectors u(t); and f(t) are the same as in Equation (1). The
matrices u̇ , u and P characterize the dynamic properties of the damping devices. For Nuid
viscoelastic devices modelled by Maxwell dampers, the matrices u̇ , u and P reduce to
 T  
r1 1 ··· 0
c .   
 ; u = 0; P = − 1  .. . . . .. 
u̇ =  .
. . . (16)
   
T
rnl 0 ··· 1

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 963

For dampers in diagonal bracings, the coeMcient c must be multiplied by the square of the
cosine of the angle of the brace with the horizontal.

Response analysis
Since the system matrix A in the previous case is not symmetric, a generalized modal anal-
ysis approach will have to be used [29; 30]. In this approach, the eigenproperties of the
following adjoint eigenvalue problems are used to uncouple Equation (12) using a similarity
transformation:
A j = j j ; AT j = j j ; j = 1; : : : ; 2N + nl (17)
where j is the jth eigenvalue, and j and j are the corresponding right and left eigenvectors.
Usually, the eigenvalues, and the corresponding eigenvectors, will occur in complex conjugate
pairs. However, some of them could also be real if the corresponding modes are critically
or over-damped. For a stable structural system, the eigenvalues must also have negative real
parts.
By using the following standard transformation of co-ordinates in Equation (12):
z(t) = ^(t) (18)
in which ^(t) is the vector of modal co-ordinates and  is the modal matrix containing the
right eigenvectors j , and pre-multiplication by the modal matrix of the left eigenvectors T ,
one obtains 2N + nl uncoupled equations for the principal co-ordinates  j (t) as follows:

˙j (t) − j j (t) = −jT Bf(t); j = 1; : : : ; 2N + nl (19)

The solution of Equation (19) for a given ground motion and zero initial conditions can be
obtained for j (t) directly in the form
 t
j (t) = e j (t−) jT Bf() d (20)
0

For a given installation of devices, any response quantity can be obtained as a linear combi-
nation of the states of the system as
R(n; t) = Tz(t) (21)
where T is a transformation matrix of appropriate dimensions. For a force-related response
quantity, the elements of the transformation matrix T will consist of sti6ness-related quantities.
For calculating the absolute acceleration vector of the structure, one can de@ne such a response
transformation matrix in the following form:
 
T = −M−1C − M−1 K L (22)
Equations (18) and (21) relate the solution in the principal co-ordinates j (t) to any desired
response quantity as follows:
R(n; t) = T^(t) (23)
Equation (23) can be used to de@ne the mean square response for a stochastic description of
the input, and to obtain the design response for a response spectrum description of the input.

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
964 M. P. SINGH AND L. M. MORESCHI

In this study, the maximum values of the structural response vector R(n; t); denoted as R(n);
are utilized in the calculation of the performance indices described later. For a stochastic
input, these quantities can be expressed as an ampli@ed value of the root-mean square (RMS)
response. That is,

∼ F E[R2 (n; t)]
R(n) = max |R(n; t)| = (24)
t

where F is a peak factor. It is straightforward to show that for a stationary stochastic input
de@ned by l-uncorrelated stationary random processes with spectral density functions Vl (!);
the stationary mean square value of the response vector in Equation (24) can be expressed
as follows:
 ∞    
 2 
2N
+nl 2N

+nl {q l } j {ql }j
E R (n; t) = Vl (!) d! (25)
−∞ l j=1 k=1 j − i! k − i!

where the vector {ql }j is the jth column of the complex matrix Ql ; de@ned as

Ql = TT bl (26)

and vector bl is the lth column of the B matrix.

This formulation can also be specialized for the self-adjoint system of Equation (7) with
the pair of symmetric matrices A s and B s . The only modi@cation is that the eigenproperties
required for the evaluation of modal response quantities are now obtained by solving the
reduced 2N × 2N -symmetrical eigenvalue problem,

−j A s j = B s j ; j = 1; : : : ; 2N (27)

In this case, the left eigenvectors coincide with the right eigenvectors, and the complex matrix
Ql is now de@ned slightly di6erently as

Ql = TT MEl (28)

For a given spectral density function, one can evaluate the integral in Equation (25) by
the residue analysis. However, if one wants to express the integral in terms of the ground
response spectra of the input motion, then Equation (25) must be explicitly expressed in terms
of the conventional frequencies !j and damping ratios j of the system modes.
Let the number of real eigenvalues be n r and the number of complex conjugate eigenvalues
be 2nc . Also, let the real and complex eigenvalues and their corresponding modal coeMcients
{ql }j and {gl }j , be de@ned as

j = −j ; {el }j = {ql }j ; j = 1; : : : ; n r (29)


j = −j !j + i!j 1 − j ; j = 1; : : : ; n c (30)

Real(j )
!j = |j |; j = − ; j = 1; : : : ; n c (31)
!j

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 965

{ql }j = {al }j + i{bl }j ; j = 1; : : : ; n c (32)

  
{gl }j = 2!j {bl }j 1 − j2 − {al }j j ; j = 1; : : : ; n c (33)

Realizing that the summation terms in Equation (25) is a function of the real and complex
quantities de@ned in Equations (29)–(33), one can simplify the expression for the mean square
value as follows:
E[R2 (n; t)] = S1 + S 2 + S 3 (34)
where the components of the quantities S1 ; S 2 and S 3 are de@ned as follows:

nr
n

r −1

nr {eli }j {eli }k
S1i = {eli }2j Jlj + 2 (j Jlj + k Jlk ) (35)
l j=1 l j=1 k=j+1 (j + k )

nr

nc
S2i = 2 {eli }j (Alijk Jlj + Blijk I1lk + Clijk I2lk ) (36)
l j=1 k=1

  

n

c −1

nc I1lk
S3i = 2 Wlijk I1lj − 4 + Qlijk (I2lj − I2lk )
l j=1 k=j+1 W

{gli }j {gli }k
+ I1lk + 4{ali }j {ali }k I2lk (37)
W2

The explicit expressions for Ajk ; Bjk ; Cjk ; Wjk ; and Qjk required in Equations (35)–(37) are
given in Appendix A. Equations (35)–(37) require the calculation of following frequency
integrals:
 ∞
Vl (!)
Jlj = d! (38)
−∞ (j2 + ! 2 )
 ∞  ∞
Vl (!) Vl (!)! 2
I1lj = d!; I2lj = d! (39)
−∞ (!j2 − ! 2 )2 + 4!j2 j2 ! 2 2
−∞ (!j − ! 2 )2 + 4!j2 j2 ! 2

These integrals can be evaluated for a given spectral density function by any suitable method.
It is noted that I1lj and I2lj are the mean square values of the relative displacement and relative
velocity responses, respectively, of a single-degree-of-freedom oscillator of parameters !j ; j
excited by ground motion component fl (t). Jlj represents the mean square response E[ 2 (t)]
associated with the following @rst-order equation:
˙(t) + j (t) = fl (t) (40)
The mean square values de@ned by I1lj and I2lj can be expressed in terms of the usual
response spectra used in seismic design practice. Similarly, Jlj ; can also be expressed in terms

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
966 M. P. SINGH AND L. M. MORESCHI

Axis of
Axis of mass centers
resistance centers y

ex

ey x

m6

3
4 2
1
m5

3
4 2
1
m4

3
4 2
1
m3

3
4 2
1
m2

3
4 2
1
m1

3
..
X g (t )
4 2
1
..
Yg (t )

Figure 1. Schematic representation of six-storey torsional building.

Table I. Mechanical properties of six-storeys torsional building.∗

Storey Mass Mass moment Storey sti6ness
(kg × 104 ) of inertia
(kg m2 × 105 ) Kx Ky
(N=m × 107 ) (N=m × 107 )
1 4.80 8.00 8.34 8.34
2 4.80 8.00 8.34 8.34
3 4.32 7.20 5.34 5.34
4 4.32 7.20 5.34 5.34
5 3.84 6.40 3.24 3.24
6 3.84 6.40 3.24 3.24
∗Note: Radius of gyration r = 10 m; 3% modal damping ratio for all modes.

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 967

0.70
worst design
0.60

average design
0.50
Performance Index
0.40

0.30 best design

0.20

0.10

0.00
0 50 100 150 200 250 300 350 400 450 500
Generation

Figure 2. Optimization history for base shear response reduction.

of a response spectrum input associated with Equation (40). Such a spectrum can be easily
developed for design purposes using an ensemble of time histories. This way, the response
of the systems with augmented damping can be expressed in terms of the seismic input as
de@ned by a response spectrum.

NUMERICAL RESULTS

To illustrate the application of the approach described above, the optimal designs have been
obtained for two building structures installed with three types of dampers described earlier.
The @rst building structure is a shear-beam model of a 24-storey building. The building has
non-uniform structural properties, and they are the same as given in Reference [10]. The @rst
six natural frequencies of this building are: 3:43; 8:30; 13:40; 18:26; 23:14, and 28:27 rad=s.
The second building is a 6-storey torsional system. The sti6ness and mass centre of the
building do not coincide; thus, its lateral and torsional responses are coupled. The schematic
of this building is shown in Figure 1, with the mass and sti6ness properties given in Table I.
The @rst six frequencies of this building are: 9:36; 9:47; 14:31; 23:41; 23:70; 35:80 rad=s. In
each of these buildings, 24 di6erent damper locations have been considered.
For numerical studies, the design earthquake ground motion is de@ned by a spectral density
function of the Kanai–Tajimi form [31]:
!g4 + 4g2 !g2 ! 2
Vg (!) = S 2 (41)
(!g2 − ! 2 ) 2 + 4g2 !g2 ! 2

where the parameters !g , g and S 2 are taken as 18:85 rad=s, 0.65 and 38:3 × 10−4 m2 =s3 =rad,
respectively. For this form of the input, the performance functions used in this study could
be calculated in the closed form. One could also use the seismic input de@ned by design

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
968 M. P. SINGH AND L. M. MORESCHI

Table II. Optimal distribution of viscous devices.

Storey Base shear Maximum Noor
acceleration
1 0 0
2 0 0
3 1 0
4 1 0
5 1 0
6 2 0
7 3 0
8 3 2
9 4 5
10 5 1
11 4 0
12 4 0
13 5 3
14 4 6
15 4 5
16 4 4
17 4 6
18 3 9
19 4 8
20 4 7
21 4 6
22 4 5
23 3 3
24 1 2

f(n∗ ) 0.4008 0.3440

Reduction (%) 59.92 65.60

response spectra in this formulation as easily. The input to the torsional system was de@ned
by two uncorrelated orthogonal components, each represented by the same spectral density
function.
First, we show the numerical results for the 24-storey shear building using classical viscous
dampers. The objective is to obtain a 60 per cent reduction in the RMS value of the base shear
of the building. The unit damper device is selected to be the one with the damping coeMcient
value of 2:55 × 107 N s=m. The force in the devices is described by Equation (4). It is desired to
calculate the total number of such unit devices and their placement that will achieve the desired
reduction optimally. The genetic algorithm was used to search for the optimum distribution
with increasing number of devices. The genetic algorithm was started with a population size
of 30, with the probabilities of cross-over and mutation of 0.95 and 0.05, respectively. The
computer code by McMahon et al. [32] was modi@ed for genetic search for our example
problems. The convergence characteristic of the optimization algorithm is depicted in Figure 2.
The improvement in the performance with generation evolution is clearly seen. The solution
seems to converge to practically the best solution in about 150 generations, except with
some slight improvement near about the 400th trial. In the following, the results obtained

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 969

24
23
22
21
20
19
18
17
16
15

Story No.
14 Uncontrolled
13
12
11
10
9
8
7
6
5
4 Controlled
3
2
1
0
0 20 40 60
3
(a) Shear [KN x 10 ]

24
23
22 Controlled
21
20
19
18
17
16
Story No.

15
14
13
12
11
10
9
8 Uncontrolled
7
6
5
4
3
2
1
0.0 0.5 1.0 1.5 2.0 2.5

(b) Acceleration [m/s2]

Figure 3. Comparison of controlled and uncontrolled response quantities: (a) shear forces
corresponding to the design of Table II, Column 2 and (b) Noor accelerations corresponding
to the design of Table II, Column 3.

in the 450th generation are shown as the converged results. This procedure determined that
72 devices are required to achieve the desired objectives. The optimal distribution of these
devices at di6erent levels is shown in Column (2) of Table II. The per cent reduction in the
performance function is shown in the last row of this table.
The distribution of the devices will, of course, be di6erent if a di6erent objective function is
used. Column (3) of Table II now shows the optimal distribution of these 72 devices to achieve
the best reduction in the maximum value of the RMS Noor acceleration (at any Noor) of the
building. It is noted that this objective function is not continuous as the maximum value could
occur at di6erent Noors in di6erent trials. Such a function would not be amenable to treatment
by a gradient-based optimization search. The genetic algorithm, however, can be used with
any performance function as long as it can be calculated. The last row shows that a reduction
of about 66 per cent can be achieved in the maximum RMS value of Noor acceleration. The
distribution of the dampers for the two cases is noted to be di6erent. Figures 3(a) and 3(b)

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
970 M. P. SINGH AND L. M. MORESCHI

Table III. Optimal distribution of viscoelastic devices.

Storey Normed Normed drifts
accelerations
1 0 0
2 0 0
3 0 0
4 0 0
5 0 0
6 1 0
7 0 0
8 1 2
9 3 5
10 3 4
11 3 0
12 3 3
13 2 1
14 4 5
15 4 7
16 6 5
17 4 4
18 5 8
19 4 6
20 6 5
21 6 5
22 5 3
23 3 2
24 2 0

f(n∗ ) 0.4997 0.3548

Reduction (%) 50.02 64.52

compare the reduced storey shears and Noor acceleration with those of the original building
without any dampers.
The next set of results is for an optimal design with viscoelastic dampers, again for the
24-storey shear building. For illustration purposes, the shear modulus G  and storage modulus
G  of the viscoelastic material are taken from Zhang and Soong [2] as: G  = 286:6 × 104 N=m2 ,
G  = 430:3 × 104 N=m2 . For a unit device, the material dimensions are chosen such as to
provide the equivalent sti6ness and damping coeMcient values of k = 3:45 × 107 N=m and
c = 1:5 × 107 N s=m, respectively.
The objective is to reduce the second norm (square root of the sum of the squares) of
the Noor accelerations by 50 per cent by installation of the viscoelastic devices. Column (2)
of Table III shows the distribution of dampers in di6erent storeys. The last row also shows
the per cent reduction in the corresponding norm. It is seen that we need a total of 65
dampers. The total number of dampers shown in Column (2) is now redistributed to achieve
the best reduction in the normed inter-storey drift. This optimal distribution is shown in
Column (3) of Table III. The damper distribution for the two designs is somewhat di6erent.
In Figures 4(a) and 4(b) we cross-compare the per cent reductions in the Noor accelerations

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 971

24
23
22
21
20
19
18
17
16
15

Story No.
14
13
12
11
10 Acceleration-Based
9 Design
8
7 Drift-Based
6 Design
5
4
3
2
1
0 10 20 30 40 50 60

(a) Shear Reduction [%]

24
23
22
21
20
19
18
Acceleration-Based
17 Design
16
15
Story No.

14
13
12
11
10
9
8
7
6 Drift-Based
5 Design
4
3
2
1
0 10 20 30 40 50 60 70

(b) Acceleration Reduction [%]

Figure 4. Cross-comparison of the e6ectiveness of the acceleration-based and drift-based designs.

and storey drifts for the two designs. In Figure 4(a), we compare the reduction in the storey
shears for the drift-based and acceleration-based designs. In this case, the drift-based design
provides a slightly better performance, especially in the lower storeys. It is, perhaps, due to
the reason that storey drift-based performance function is more inNuenced by the drifts in the
lower storeys. In Figure 4(b) are compared the reductions in Noor accelerations for the two
designs, and here the acceleration-based design provides a better reduction. Thus, from these
comparisons we observe that a best reduction in the response will be obtained by the design
made for that norm. That is, the accelerations are likely to be best reduced by the damper
distributions obtained for the acceleration-based norm. It is also noted that the accelerations
and drifts response reductions are about the same, indicating that the two response quantities
have a strong correlation. That is, a design made to reduce the acceleration response will also
reduce the storey shear response and vice versa. The inter-storey drifts are often related to
the expected non-structural damage in earthquakes, and so are the Noor acceleration levels.

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
972 M. P. SINGH AND L. M. MORESCHI

Table IV. Optimal distribution of Nuid viscoelastic devices for torsional building.
Storey Number of devices at the jth location
no.
Column deformation-based Maximum Noor acceleration-based
performance function performance function
(a) (b)

1 2 3 4 1 2 3 4
1 0 1 0 1 0 0 0 0
2 0 0 0 0 0 0 0 0
3 8 5 4 2 1 0 5 5
4 2 1 1 1 1 0 1 0
5 4 3 2 1 1 0 4 4
6 0 0 0 0 1 0 1 1

No. of 36 25
devices

% 49.81 49.80
Reduction

The next set of results is for the torsional building system, installed with dampers rep-
resented by the frequency-dependent Maxwell model of Equation (10). Figure 1 identi@es
for each storey the possible locations at which a device can be placed (1–4). Therefore,
there are a total of nl = 24 possible places to locate a damper in this particular building. A
viscous Nuid damper is selected from available commercial products (Taylor devices) with
mechanical properties taken from an experimental study [33] as follows: c = 2:02 × 105 N s=m,
and  = 0:014 s. The excitation along the x and y directions are now de@ned by the same
Kanai–Tajimi spectral density functions.
Optimal designs are obtained for two performance functions. The @rst performance function
is related to the deformations of the building columns, and is de@ned in the normalized
form as
!
ncl  2 
! j=1 Rjx (n) + Rjy 2
(n)
f [R(n)] = "
ncl 2 2
(42)
j=1 (Rjox + Rjoy )

where Rjx (n) and R jy (n) are the maximum relative displacements experienced by the jth
column along the x- and y-axis, respectively; R jox and R joy are the respective quantity of
the original (uncontrolled) building, and n cl is the total number of columns in the building.
It is desired to reduce this performance function by an arbitrarily selected value of 50 per
cent. The part (a) of Table IV shows the distribution of devices. The results are for the
eccentricities ratios of ' x = −5 per cent and 'y = 5 per cent. The last but one row of the table
gives the total number of devices required, and the last row gives the actual level of reduction
achieved.
The part (b) of the Table IV similarly shows the distribution of devices required to re-
duce the maximum Noor acceleration of the building by 50 per cent, for eccentricities ratios

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 973

y direction

Story No. x direction

x direction
(modified model)
1
0 10 20 30 40 50 60 70
Response Reduction [%]
(a) Drifts-Based Design

x direction
(modified model)
5
y direction

4
Story No.

x direction

1
0 10 20 30 40 50 60 70
Response Reduction [%]
(b) Acceleration-Based Design

Figure 5. Response reduction in storey drifts and Noor accelerations.

of ' x = 10 per cent and 'y = −5 per cent. For this case, the performance index takes the
form
 x 
R i (n) Ryi (n)
f [R(n)] = max ; (43)
i R0 R0
where Rxi (n) and Ryi (n) represents the maximum Noor acceleration at the ith storey along the
x and y directions, respectively, and R0 is the maximum acceleration along both directions
of the original building. This is again a discontinuous performance function. The per cent

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
974 M. P. SINGH AND L. M. MORESCHI

reductions in the storey drifts and Noor acceleration in the x and y directions are shown in
Figures 5(a) and 5(b), respectively. In Figure 5, we also compare the response reductions that
would be obtained if the frequency dependence of the model were ignored. The results of the
two models are slightly di6erent, indicating the e6ect of using a simpli@ed device model. In
this particular case, this di6erence is not very large.

CONCLUDING REMARKS

The paper demonstrates the use of a genetic algorithm approach for optimal design of passive
dampers for linearly behaving building structures. The study employs the classical viscous
dampers, solid viscoelastic dampers and Nuid viscoelastic dampers for the dissipation of en-
ergy. The genetic approach is used to calculate the required number of a given capacity
dampers and their optimal placement locations in a building to achieve a desired reduction in
the response. For a given number of dampers, the approach can also @nd their most suitable
locations to achieve the maximum reduction in a desired response. The response reduction
performance could be expressed in terms of a reduction in a chosen response quantity such
as base shear, over turning moment or Noor acceleration. It could also be de@ned in terms of
a performance function, depending upon several response quantities. The approach is Nexible
inasmuch as it can work with any performance function as long as it can be numerically
calculated. Numerical results are reported for a shear building model and torsional building
model installed with three types of damping devices and several di6erent forms of performance
measures.

ACKNOWLEDGEMENTS

This study was @nancially supported by the National Science Foundation through grant numbers CMS-
9626850 and CMS-9987469. This support is gratefully acknowledged. The authors are also grateful to
Prof. Zafer GSurdal of the Department of Engineering Science and Mechanics at Virginia Tech for his
help.

APPENDIX A

The partial fraction coeMcients required in Equations (35)–(37) are de@ned as

Wlijk = −{(W2 − W−2 ) lijk − 2[1 − W2 + 2(j2 W2 − k2 )])lijk }=*jk (A1)

Qlijk = {(W2 − W−2 ))lijk − 2W−2 [1 − W2 + 2(k2 W2 − j2 )lijk ]}=(!j2 *jk ) (A2)
Alijk = j (−{gli }k + 2{ali }k j )=+jk (A3)
Blijk = −!k2 [2{ali }k !k2 + {gli }k (j + 2!k k )]=+jk (A4)
Clijk = [{gli }k j + 2{ali }k !k (!k + 2j k )]=+jk (A5)

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
 OPTIMAL PLACEMENT OF DAMPERS 975

with

W = !j =!k (A6)
)lijk = {gli }j {gli }k (1 − 4j2 + 4j k W − W2 ) + 4!j2 {ali }j {ali }k (1 − W2 )
+ 4!j ({ali }j {gli }k − {ali }k {gli }j )(k W − j ) (A7)
lijk = {gli }j {gli }k (W2 − 1) + 4!j2 {ali }j {ali }k [(1 − 4j2 )W2 + 4j k W − 1]
+ 4!j W({ali }j {gli }k − {ali }k {gli }j )(j W − k ) (A8)
*jk = 16(j2 + k2 − j4 − k4 ) + 4(W2 + W−2 )

[1 − 2(j2 + k2 − 2j2 k2 )] − W−4 − W4 − 6 (A9)

+jk = !k2 + j2 + 2j !k k (A10)

REFERENCES

1. Ashour SA, Hanson RD. Elastic seismic response of buildings with supplemental damping. Report No. UMCE
87-01, Department of Civil Engineering, University of Michigan, Ann Arbor, MI, 1987.
2. Zhang RH, Soong TT. Seismic design of viscoelastic dampers for structural applications. Journal of Structural
Engineering 1992; 118(5):1375 –1392.
3. Shukla AK, Datta TK. Optimal use of viscoelastic dampers in building frames for seismic response. Journal of
Structural Engineering 1999; 125(4):401– 409.
4. Wu B, Ou JP, Soong TT. Optimal placement of energy dissipation devices for three-dimensional structures.
Engineering Structures 1997; 19:113–125.
5. Natke HG, Soong TT. Topological structural optimization under dynamic loads. In Optimization of
Structural Systems and Applications, Hernandez S, Brebbia CA (eds). Computational Mechanics Publications:
Southampton, 1993.
6. Milman MH, Chu CC. Optimization methods for passive damper placement and tuning. Journal of Guidance,
Control and Dynamics 1994; 17(4):848 – 856.
7. Gluck N, Reinhorn AM, Gluck J, Levy R. Design of supplemental dampers for control of structures. Journal
of Structural Engineering 1996; 122(12):1394 –1399.
8. Takewaki I. Optimal damper placement for minimum transfer functions. Earthquake Engineering and Structural
Dynamics 1997; 26:1113–1124.
9. Takewaki I, Yoshitomi S, Uetani K, Tsuji M. Non-monotonic optimal damper placement via steepest direction
search. Earthquake Engineering and Structural Dynamics 1999; 28:655 – 670.
10. Singh MP, Moreschi LM. Optimal seismic response control with dampers. Earthquake Engineering and
Structural Dynamics 2001; 30:553–572.
11. Holland JH. Adaptation in Natural and Arti7cial Systems. University of Michigan Press: Ann Arbor, MI, 1975.
12. Goldberg DE. Genetic Algorithms in Search, Optimization and Machine Learning. Addison-Wesley: Reading,
MA, 1989.
13. Davis L. Handbook of Genetic Algorithms. Van Nostrand Reinhold: New York, 1991.
14. Gen M, Cheng R. Genetic Algorithms and Engineering Design. Wiley: New York, 1997.
15. Chan E. Optimal design of buildings structures using genetic algorithms. Ph.D. Dissertation, California Institute
of Technology, Pasadena, CA, 1997.
16. Cheng FY, Li D. Multiobjective optimization design with pareto genetic algorithm. Journal of Structural
Engineering 1997; 123(9):1252–1261.
17. Furuya O, Hamazaki H, Fujita S. Proper placement of energy absorbing devices for reduction of wind-induced
vibration caused in high-rise buildings. Journal of Wind Engineering and Industrial Aerodynamics 1998;
74 –76:931– 942.
18. Kim Y, Ghaboussi J. A new method of reduced order feedback control using genetic algorithms. Earthquake
Engineering and Structural Dynamics 1999; 28(3):193–212.

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976
976 M. P. SINGH AND L. M. MORESCHI

19. Singh MP, Moreschi LM. Genetic algorithm-based control of structures for dynamic loads. Proceedings of the
IA-99, Osaka University, Japan, November 15 –16, 1999.
20. Singh MP, Moreschi LM. Optimal seismic design of building structures with friction dampers. Proceedings of
the U.S.–China Millenium Symposium on Earthquake Engineering, Beijing, China, November 8 –11, 2000.
21. Rao SS, Pan T-S. Optimal placement of actuators in actively controlled structures using genetic algorithms.
AIAA Journal 1991; 29(6):942– 943.
22. Onoda J, Hanawa Y. Actuator placement optimization by genetic and improved simulated annealing algorithms.
AIAA Journal 1993; 31(6):1167–1169.
23. Furuya H, Haftka RT. Placing actuators on space structures by genetic algorithms and e6ectiveness indices.
Structural Optimization 1995; 9:69 –75.
24. Hadi MNS, Ar@adi Y. Optimum design of absorber for MDOF structures. Journal of Structural Engineering
1998; 124(11):1272–1280.
25. Singh MP. Seismic response by SRSS for nonproportional damping. Journal of Engineering Mechanics ASCE
1980; 106(6):1405 –1419.
26. Maldonado GO, Singh MP. An improved response spectrum method for calculating seismic design response.
Part II: non-classically damped structures. Earthquake Engineering and Structural Dynamics 1991; 20(7):
637– 649.
27. Federal Emergency Management Agency—FEMA. NEHRP guidelines and commentary for the seismic
rehabilitation of buildings. Reports No. 273 and 274, Building Seismic Council, Washington, DC, 1997.
28. Soong TT, Dargush G. Passive Energy Dissipation Systems in Structural Engineering. Wiley: New York,
1997.
29. Singh MP, Chang TS, Suarez LE. A response spectrum method for seismic design evaluation of rotating
machines. Journal of Vibration and Acoustics, ASME, 1992; 114(4): 454 – 460.
30. Moreschi LM. Seismic design of energy dissipation systems for optimal structural performance. Ph.D.
Dissertation, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, July 2000.
31. Tajimi H. A statistical method of determining the maximum response of a building structure during an
earthquake. Proceedings of 2nd World Conference in Earthquake Engineering, Tokyo, Japan, 1960.
32. McMahon MT, Watson LT, Soremekun GA, GSurdal Z, Haftka RT. A Fortran 90 genetic algorithm module for
composite laminate structure design. Engineering with Computers 1998; 14: 260 –273.
33. Reinhorn AM, Li C, Constantinou MC. Experimental and analytical investigation of seismic retro@t of structures
with supplemental damping: Part 1—Nuid viscous damping devices. Report No. NCEER 95-0001, National
Center for Earthquake Engineering Research, University of New York at Bu6alo, Bu6alo, NY, 1995.

Copyright ? 2002 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2002; 31:955–976