Anda di halaman 1dari 11

International Journal of Mechanical Sciences 75 (2013) 123133

Contents lists available at ScienceDirect

International Journal of Mechanical Sciences


journal homepage: www.elsevier.com/locate/ijmecsci

Reliability-based-design-optimization of base isolated buildings


considering stochastic system parameters subjected to
random earthquakes
Sudib Kumar Mishra a,n, Bijan Kumar Roy b, Subrata Chakraborty b
a
Department of Civil Engineering, Indian Institute of Technology, Kanpur, UP 208016, India
b
Department of Civil Engineering, Bengal Engineering and Science University, Shibpur 711103, Howrah, India

art ic l e i nf o a b s t r a c t

Article history: Base isolation (BI) is established as an effective control strategy for improving seismic performance of
Received 15 March 2012 structures. The studies on performance of BI system under stochastic earthquake load are notable.
Received in revised form However, most of the studies consider system parameters as deterministic and the optimal isolator
21 June 2013
characteristics are obtained accordingly. A major limitation of such deterministic approach is that the
Accepted 25 June 2013
uncertainties in the performance related decision variables cannot be included in the parameters
Available online 5 July 2013
optimization process. But, the safety of a BI system can be signicantly affected due to uncertainty in the
Keywords: system parameters. The present study deals with the Reliability-Based-Design-Optimization (RBDO) of BI
Base isolation system to mitigate seismic vibration effects considering system parameter uncertainty. With the aid of
Reliability
matrix perturbation theory and rst order Taylor series expansion, the concept of total probability theory
Stochastic
is used to evaluate the unconditional response of structures under parameter uncertainty. For this, the
Earthquakes
Optimization conditional second order information of responses is obtained in the random vibration framework.
Subsequently, the unconditional failure probability of the primary structure is used as the objective
function in order to obtain the optimum parameters of the isolator. The proposed design is tolerant to the
uncertainty and provides estimate of the enhanced risk unforeseen in the deterministic system.
A multistoried building frame isolated by Lead-Rubber-Bearing (LRB) is taken up for numerical illustration
and to elucidate the effect of parameter uncertainty on the optimum performance of BI system.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction In BI system, the building rests on a system of isolators which


uncouple the building from the horizontal component of the
Passive control is a viable alternative to the traditional design ground motion, thereby, effectively reducing the load transmitted
which relies on the energy dissipation by inelastic deformations of to structure. These devices adopt different materials and design
structural elements. In contrast, the passive vibration control methodologies in order to disconnect the superstructure motion
substantially reduces the structural response to ensure minimal from the ground. A number of BI devices, such as, Rubber Bearings,
damage to the structures. The concept of passive vibration control Lead Rubber Bearings (LRB), High Damping Rubber Bearings,
is widely accepted and is frequently implemented in buildings and Friction Pendulum, and Resilient Friction Bearing are available
bridges over last three decades due to its low maintenance nowadays for seismic protection of buildings and bridges. The
requirements, leading to overall economy. Many such devices, effectiveness of BI systems and their performances have been
such as Tuned Mass Damper, Fluid Viscous Damper, Viscoelastic extensively studied in the past [14]. The studies on stochastic
Damper, Friction Damper, Base Isolation (BI) system, Metallic Yield response of BI system under random earthquakes are notable [5,6].
Damper, Tuned Liquid Column Damper have been implemented to It is well established that the performance of BI system largely
mitigate structural vibration effects. Amongst these, the BI systems depends on the characteristics of the isolator such as the
have been used since late 1970s and globally accepted as an yield strength for LRB, Rubber Bearings and optimal damping
effective technology to reduce the seismic vibration effects on for Resilient Friction Bearing. Attempts are made in order to
strategically important structures as well as in seismic retrot. characterize such optimal parameters [79] to ensure optimal
performance.
The most commonly adopted approach of designing BI system
n
Corresponding author. Tel.: +91 512 259 6731. is to consider the earthquake load as the only source of random-
E-mail address: smishra@iitk.ac.in (S.K. Mishra). ness. The stochastic structural responses are obtained in the

0020-7403/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijmecsci.2013.06.012
124 S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133

random vibration framework. Subsequently, a standard nonlinear control is very limited. Moreover, in such studies, the root mean
optimization problem is formulated to minimize the stochastic square (RMS) displacement of the primary structure is taken as the
response [10], referred as Stochastic Structural Optimization (SSO). performance index in the assessment of BI system. However, perfor-
The dynamic load is considered to be the only source of random- mance of a BI system based on the RMS displacement may not ensure
ness and the system parameters are assumed to be deterministic. that the associated risk with regard to the concerned limit state is
A major limitation of such deterministic approach is that the also minimum [16]. Hence, minimization of risk of exceedence of a
uncertainty in the performance related decision variables cannot predened serviceability or strength limit state quantifying the
be included in the stochastic response analysis and the related structural performances is more desirable in many cases.
optimization procedure. It has been demonstrated that the inter- Keeping the above in view, the present study deals with the
play among the parameter uncertainty and loading uncertainty optimum design of BI system considering the effect of uncertainty
can markedly change the response of a system [11] and its of various system parameters in the vibration control of building
safety [12]. The optimal design is also observed to be changed structures subjected to stochastic earthquake load. The probability
signicantly in presence of system uncertainty [13]. In case of a of failure with respect to maximum top oor displacement of the
control system, particularly, the efciency may be drastically building structure is considered as the performance index. With
reduced if the parameters are off tuned to the vibrating mode, it the aid of matrix perturbation theory, using rst order Taylor
is designed to suppress due to unavoidable presence of uncer- series expansion, the total probability concept [12,23] is used to
tainty. Thus, the passive vibration control of structures with BI evaluate the unconditional response of structure considering the
considering uncertain parameters has attracted interests of the effect of system parameter uncertainty. For this, the conditional
structural vibration control community. (conditioned on the fact that system parameters are deterministic)
The developments in the eld of vibration control by various second order information of response quantities are obtained in
passive control devices considering system parameter uncertainty are the random vibration framework using state space formulation.
notable [1423]. However, the same is not the case for the BI system. Subsequently, the unconditional top oor RMS displacement of the
The studies on the performance of BI system in connection with isolated structure is evaluated. These unconditional responses are
passive vibration control strategy are very limited. Juhn and Manolis then used to obtain the unconditional failure probability based on
[24] have indicated the effect of uncertainty on the response of BI rst passage failure criterion, serving as the objective function in
system. Kawano et al. [25] have studied the effects of uncertainty on the BI system optimization. The design optimization is also
the nonlinear dynamic response of BI structure in the framework of performed based on minimization of the unconditional RMS
Monte Carlo Simulation. It is illustrated that the uncertain parameters response of the top oor in order to demarcate the importance
have signicant effects on the maximum response of BI system. of RBDO. A multi-storeyed shear building, isolated by LRB is
Scruggs et al. [26] proposed a probability-based active control synth- considered for numerical study to elucidate the effect of parameter
esis algorithm for seismic isolation of an eight-storey benchmark uncertainty on the optimal performance of BI system.
structure using an array of ideal active control devices working with
the passive isolation bearings, considering uncertain model para-
meters. Zhou et al. [27], Zhou and Wen [28] presented two adaptive 1.1. Stochastic response of base isolated structure
back stepping control algorithms for active seismic protection of
building structures using BI system considering uncertainty in the The two dimensional shear building, isolated with LRB as shown
hysteretic structural parameters. The reliability assessment of BI in Fig. 1a is considered in the present study. Apart from shifting the
system under random earthquake was presented by Chen et al. fundamental natural period of the structure, the LRB also provides
[29,30]; Zhang et al. [31]. However, the system parameters are treated additional hysteretic damping through yielding and viscous damping
deterministically in these works. It appears that even though the study of rubber to dissipate the input seismic energy. Since the BI
on performance of BI system in connection with active vibration substantially reduces the structural response, its behavior can be
control strategy is notable, the effect of uncertain parameters and their reasonably assumed to be linear. However, the behavior of BI system
implications on optimal design of BI system for passive vibration is highly non-linear as the energy is dissipated through large shear

Fig. 1. (a) Idealized model of the base isolated structure, (b) idealization of the LRB and (c) bilinear forcedeformation hysteresis of LRB.
S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133 125

deformation and yielding of the lead core. The idealized mechanical The post-yield stiffness (kb ) of the isolator is chosen to provide
model of the non-linear bearing is depicted in Fig. 1b and associated specic isolation time period T b , given by
forcedeformation behavior is shown in Fig. 1c. The structure is q
assumed to be excited by the horizontal component of earthquake T b 2 M=kb 5
ground motion only.
where M ni 1 mi mb is the total mass, i.e. sum of all the oor
Assuming linear superstructure behavior, the equation of
mass mi and mass of the LRB system mb . The viscous damping
motion can be written as
cb in the isolator can be expressed as
   
M  x C  x_ K fxg M fr gx g x b 1 cb 2b Mb 6

where, M ; K  and C  are the mass, stiffness and damping in which b is the viscous damping ratio and b is the frequency of
matrices of the xed-base structure respectively; fxg is the the bearing, expressed as b 2=T b .
displacement vector containing the lateral displacement of any The yield displacement of the isolator bearing q is obtained
oor relative to the isolator, as shown in Fig. 1a. The vector fr g is from the loaddeformation relation of the isolator shown in Fig. 1c.
the inuence coefcient vector. It is used to represent the inertial It is the displacement corresponding to the yield strength of the
forces distributed amongst all the degrees of freedom (dof) of a isolator F Y . The yield strength is conveniently normalized F 0
structure due to unit acceleration at the base. For the shear with respect to the weight of the structure
 T
building, this is given as fr g 1 1 :: 1 as the forces pro- F 0 F Y =W 7
duced at any dof due to unit acceleration at the base is equal to the
mass associated with that dof [32]. The acceleration of the where W is the total weight of the isolation-superstructure system
isolation bearing with respect to the ground is given by x b and W Mg, g is the gravitational acceleration.
x g is the acceleration of the earthquake ground motion. The The nonlinear forcedeformation characteristic (Eq. 4) of the
governing equation of motion for the isolator (Fig. 1b) can be LRB is too complicated to be readily incorporated in the state-
written as space formulation for evaluating the stochastic response and the
sensitivity statistics. This equation can be linearized through
mb x b cb x_ b F b c1 x_ 1 k1 x1 mb x g 2 equivalent stochastic linearization [37,38]. The equivalent linear
form of the nonlinear Eq. (4) is expressed as
where mb is the mass of the isolator base, F b is the restoring force
developed in the LRB, cb is the viscous damping of rubber and k1 qZ_ C e x_ b K e Z 0 8
and c1 are respectively the stiffness and damping of the rst storey where C e and K e are the stochastically equivalent linear damping
of the superstructure. Further details of modeling and design of and stiffness obtained by minimizing the expectation of the
isolation systems for building structures may be found in the least square error between the linear and nonlinear terms in Eqs.
Building Codes [33,34]. The bilinear forcedeformation behavior of (8) and (4), respectively. The equivalent linearization as proposed
the LRB is adopted herein, where the forcedeformation behavior here incurs some error. However, it has been demonstrated through
is expressed through the differential BoucWen model [35,36]. In simulation studies that the error introduced due to this approxima-
this model, the isolator restoring force is expressed as tion is insignicant [3538]. It may also be noted that though the
linearization is applied for nonlinear isolator behavior, the system
F b xb ; Z kb xb 1F Y Z 3
will still portray nonlinear characteristics due to the fact that the
where cb and kb are the viscous damping and the elastic stiffness equivalent linearized stiffness and damping are still functions of the
of the isolator respectively. The elastic stiffness is the initial response. More details about the linearization procedure and its
stiffness of the LRB and the viscous damping is provided by the applicability may be found elsewhere [38].
rubber and lead core. The symbols xb and x_ b are the relative For the adopted value of 1, close form expression of the
displacement and velocity of the LRB. The parameter represents equivalent linear damping and stiffness of the isolator can be
the ratio of the post to pre yield stiffness of the LRB, referred as obtained as [37,38]
rigidity ratio and F y is the yield strength of the bearing. The 8 9
r>>
< r
 > >
=
variable Z quanties the hysteresis of the LRB and is governed by 2 Ex_ b Z  2
the following differential equation: Ce r
h i E Z  9
>> >
>
: E x_ 2b ;
qZ_ jx_ b jZjZj1 x_ b jZj x_ b 4
8 9
r>
>
< r >
>
h i Ex_ b Z  =
in which q is the yield displacement of the bearing. The para-
2
meters ; ; and appears in Eq. (4) control the nature of Ke E x_ 2b r
h i> 10
> >
the hysteretic cycle of the isolator. The parameters and control : E Z2 > ;
the shape and size of the hysteresis loop whereas the parameter
controls the transition from elastic to plastic phase. With where E  is the expectation operator. In the stochastic lineariza-
increasing value of the elastic to plastic transition becomes tion procedure, the responses xb ; x_ b of the isolator are assumed to
increasingly sharp. However, due to smooth nature of the of the be jointly Gaussian. This assumption might not be correct given
BoucWen model this transition is smooth. The ideal sharp the fact that the system is nonlinear in presence of hysteresis of
bilinear nature can only be attained at -1 (innity). However, the isolator. However, it has been demonstrated [38] that this does
with the presently adopted value of 1, the smooth transition not result in serious error insofar as the stochastic response
can adequately be taken close enough to the ideal bilinear evaluation is concerned.
behavior. The parameter controls the nature of the hysteresis i. The well-known KanaiTajimi stochastic process [39,40] which
e. 4 0 implies hardening and o 0 results in softening. The characterizes the input frequency content for a wide range of
values of other parameters adopted in the present study are practical situations is adopted in the present study to represent
0:05; 0:5 and 1. These are corresponds to the the stochastic earthquake. The stationary model for x g is obtained
smooth bilinear force deformation characteristics of the isolator, by ltering a white noise process acting at the bed rock through a
shown in Fig. 1c [35,36]. linear lter representing the ground. The lter equations can be
126 S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133

expressed as Also, the lter equations from (11b) can be rewritten as


x g x f w
11a x f 2g g x_ f 2g xf w
18

x f 2g g x_ f 2g xf w
11b Now, the state variables are augmented in a state space vector
as
Substituting Eq. (11b) in Eq. (11a) yields h   iT
fY g fxg xb Z xf x_ x_ b x_ f 19
x g 2g g x_ f 2g xf 12

where w is the white noise intensity at the rock bed with PSD S0 ,
g and g are the frequency and damping parameters representing The Eq. (15) to (18) can be written in the state space form as
the ground soil strata over the rock bed and underlying the d
fY g AfY g fwg 20
building. The parameters x f , x_ f and xf are the acceleration, velocity dt
and displacement of the KanaiTajimi lter, respectively.
where A is the augmented system matrix and
The KanaiTajimi model for the earthquake ground motion
adopted herein assumes the earthquake as a stationary stochastic h iT
fwg f0g 0 0 0 f0 g 0
w 21
process. However, ground motions are nonstationary in nature in
reality and this nonstationarity could be either in amplitude or in
frequencies. The amplitude nonstationarity in the proposed for- In stochastic analysis, rather than the response, the statistics of
mulation can be easily incorporated by introducing a time- responses, such as covariance are evaluated. It can be shown that
dependent modulating function. However it is seen that the the covariance matrix C YY  of the state vector fY g (assumed as
amplitude nonstationarity does not have much effect and a Markovian) satises an equation of the form [44]
stationary model is good enough to provide the representative
statistics quite well [7,38]. d
C YY  AC YY T C YY AT Sww  22
The frequency nonstationarity might have signicant inuence dt
on the response behavior. The frequency nonstationarity can be in which the elements of C YY  is given by
incorporated by time dependent frequency content and damping  
CYiYj E Y iY j 23
in the model Eqs. ((11a) and (11b)). Among several alternatives,
the FanAhmadi model [41,42] includes such aspect, which can be where E  is the expectation operator. The matrix Sww  is the
directly incorporated in the present formulation. However, it covariance matrix of the rock bed white noise intensity. Following
needs evaluation of time dependent response statistics and the structure of the vector fwg, the matrix Sww  has all terms zero
performance function in the optimization procedure which needs except the last diagonal, given as 2S0 .
further study. The matrix equation in (22) is solved by fourth order Runge
The seismic motion is introduced in the formulation by Kutta technique to obtain the covariance of responses. It may be
incorporating the Eq. (11b) and substituting expression of x g from noted that the stochastically linearized stiffness and damping
Eq. (12) in the Eqs. (1) and (2). The white noise intensity S0 of the properties of the nonlinear LRB are functions of the response
rock bed excitation is related to the RMS value of the ground itself. Thus the solution of Eq. (22) needs an iterative procedure.
acceleration time history as [43] This includes updating the equivalent linear properties with
calculated response statistics obtained from the previous step.
S0 2g s2u g =1 42g g 13
The iteration stops as the estimate of the response covariance from
where s2u g is the RMS of the ground acceleration time history of an the two successive steps converge, which is based on the L2 norm
earthquake. of the error among the response covariance matrices obtained
The equations derived above are now rearranged to express in from two successive iterations. The RMS response are obtained
the state space form suitable for evaluating stochastic responses. from the covariance as
Substituting Eq. (3) in (2) and normalizing with respect to mb , the q
equation of isolator reduces to sY i C Y i Y i 24

cb k 1F y c1 k1 The peak value of any response quantity, such as the isolator
x b x_ b xb Z x_ 1  x1 x g 14
mb b mb mb mb mb displacement xb (denoted as uxb ) is obtained by
Substituting the expression of x g x b from the above in uxb ksxb 25
Eq. (1) and multiplying both sides of Eq. (1) with M 1 , it can be
rewritten as in which k is the peak factor, obtained as [43]
p p
    k 2lnT d 0:577= 2lnT d 26
x M 1 C  x_ M 1 K fxg

c k 1F y c1 k1 where T d is the duration of the ground motion and is a factor


fr g b x_ b b xb Z x_ 1  x1 15
mb mb mb mb mb dened as
Substituting the expression of x g from lter Eq. (12), the sx_ b =sxb 27
equation of base mass/isolator given by Eq. (2) can be rewritten as
where sx_ b is the RMS velocity of isolator.
c k 1F y c1 k1
x b  b x_ b  b xb  Z x_ 1 x1
mb mb mb mb mb 1.2. Reliability of BI system
2f f x_ f 2f xf 16
In seismic reliability analysis, the BI system is assumed to fail if
The stochastically linearized Eq. (8) for the hysteretic degree of
the response crosses some threshold barrier. The reliability for the
freedom can be expressed as
i-th limit state is thus determined by the rst crossing of any
Ce Ke structural response xi to certain barrier level i . For the BI system
Z_  x_ b  Z 17
q q considered herein, the rst time bilateral crossing of any response
S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133 127

xi to the barrier level i within a time interval T d is expressed as where


  T
F i xi =i ; t0; T d  28 A A Sww
B C YY T C YY  34b
i i i
The conditional failure probability PF i = with regard to  
response xi , based on the structural and excitation model para- In the above, C YY =i is the sensitivity of the response
meters specied by fg (detailed in Eq. 32) can be estimated by the covariance C YY with respect to the parameter i . It may be noted
following classical Rice's formulation [45,30]: here that Eq. (34a) has the same form as that of Eq. (22) and can be
  solved similarly.
PF i 1exp i fgT d 29 The second order sensitivity is obtained by differentiating Eq.
where i fg is the conditional threshold-crossing rate for i-th (34a) with respect to parameter j and with rearrangement of the
failure mode and T d is the duration of the earthquake motion. terms it can be written as
With further assumption that the response is a stationary stochas- 2 T 2
d 2 C YY C YY C YY
tic Gaussian process with zero mean, the conditional crossing rate A AT D 35
dt i j i j i j
can be written as [43]
! in which D is given by
sx_ i 2 h ih iT h ih iT h 2 i h ih iT
i f g exp  i 2 30 A C YY C YY A A T A C YY
sxi 2sxi D j i
i j
i j
C YY  i j
h 2 iT h ih iT h 2 i 36
where sxi and sx_ i are the RMS of xi and x_ i . In case of multiple C YY A
C YY  i j j i i j
A Sww

failure events (e.g. crossing of the top oor displacement as well as


nq
inter storey drift) with nq number of limit states, F i 1 F i . This It is noted that some of the matrices involved in Eq. (36) for
implies that the system fails if any one of the response xi exceeds evaluating D are null because A and Sww  contain rst order
its threshold i . The mean up-crossing rate of the system is then terms of random variables i which vanishes after rst/second
approximated by order differentiation.
nq The sensitivity of RMS response of a quantity Y m (m-th term of
i 31 the state vector fY g) with respect to the i and j-th random variables
i1 are obtained by differentiating expression (24) as
and the probability of failure P F can be approximated as sY m 1 1 C Y m Y m
" # p 37a
nq i 2 C Y m Y m i
PFjfg1exp  i fgT d 32 "
i1
#
2 sY m 1 1 2 C Y m Y m 1 1 C Y m Y m 2
p  37b
i j 2 C Y m Y m i j 2 CYmYm i

in which sY m is the RMS response of the quantity Y m , sY m =i and


2. Response evaluation considering system parameters 2 sY m =i j are the rst and second order sensitivity with respect
uncertainty to i and j respectively.

The stochastic responses of structures under random earth- 2.2. Unconditional response considering uncertain system
quakes are obtained by solving Eq. (22). Such responses are parameters
conditional because the response evaluation is conditioned on
the fact that the system parameters are deterministic. In engineer- The total probability theorem is employed in evaluating the
ing systems, the parametric variations are bound to occur. The unconditional failure probability of the structure under parametric
uncertainty in these parameters may lead to large and unexpected uncertainty. If the information on the conditional failure prob-
excursion of responses that may lead to drastic reduction in ability P c is known along with the joint probability density of the
accuracy and precision of the safety evaluation [11,12]. In design random system parameters f fg, the unconditional failure
of such system, apart from the stochastic nature of the earthquake probability P un is obtained as
loading, the uncertainty with regard to these parameters are Z Z

expected to have inuences. In order to include the effect of P un P c F=fg f fgd1 :::::dnv 38
parametric uncertainty on the safety evaluation of system, the 1;::nv

sensitivities of the stochastic responses with respect to the where nv is the number of random parameters. It is observed that
random parameters are essential [18,20,21]. The relevant formula- the unconditional probability itself is random with associated
tions are presented subsequently. variance. However, in RBDO the mean of the unconditional failure
probability is only minimized. The associated variance can only be
2.1. Sensitivity of stochastic structural responses included in the Robust Design Optimization (RDO) framework
[18,20,23], not in the scope of the present study.
The uncertain system parameters considered in the present Evaluation of the integral in Eq. (38) is prohibitively exhaustive
study are denoted by the vector of random design parameters for practical engineering system. A number of simplications are
fg as applied in order to approximate this integral. Presently, this is
attained by evaluating the unconditional responses following the
h iT
fg k c kb cb F Y g g s0 33 preceding approach. The unconditional responses, so obtained, are
then employed in Eq. (32) to estimate mean of the unconditional
First order sensitivity of the basic state space Eq. (22) with failure probability, as employed in RBDO.
respect to i-th parameter i can be readily obtained as The stochastic response of the structure under random earth-
quake incorporating the effects of the uncertain system para-
d C YY C YY T C YY meters can be expanded in Taylor series about the mean value of
A AT B 34a
dt i i i the uncertain parameters (with the assumption that the random
128 S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133

variability is small). The random design parameter i can be system parameters characterizing the structure, isolator and
viewed
 as the superposition of the deterministic mean component stochastic earthquake load model as deterministic. Such optimiza-
i with a zero mean deviatoric component i . The Taylor tion approach is termed as conditional where the system para-
series expansion of the RMS response of the mth term in the meters are taken as deterministic and the objective function is
state vector fY g can be expanded by Taylor series about the mean sX n based on the conditional response. However, in the design of
value of the random variables as optimum BI system, apart from the stochastic nature of the
 nv s earthquake load, the uncertainty with regard to these parameters
Ym 1 nv nv 2 sY m
sY m sY m i i i j 39 are expected to have inuences. Thus, one should use the
i1 i 2 i 1 j 1 i j
unconditional response, sX n (dened by Eq. (39)) to redene
where nv is the total number of random variables involved in the optimization problem as
the problem and the derivatives of the respective response 1 nv 2 sX n 2
quantities are the sensitivities of RMS responses addressed in Find F 0 to min: : f sxn sxn i s i 41
2 i 1 2i
the previous Section 3.1. As i is a zero mean random variables,
its expectation vanishes, i.e. Ei 0. It is also assumed that the If the optimization problem is formulated to minimize the
uncertain random variables are uncorrelated and therefore the failure probability instead, then the optimization is based on
expectation of the cross terms in Eq. (38a), i.e. Ei j 0 if ij. minimizing the failure probability PFj. It can be noted that
Therefore, the quadratic approximation provides the expected the failure probability obtained by Eq. (32) is conditional also
value of the unconditional RMS response as because the structural and excitation model specied by fg are
assumed to be known a priori. Based on this assumption, the
1 nv 2 sY m 2
sY m sY m i s i 40 threshold crossing rate was computed using Eq. (29) and subse-
2 i 1 2i
quently the probability of failure was evaluated from Eq. (28).
where si is the standard deviation of the i-th random parameter. However, knowing the conditional second order information of
Using the formulation described above, the unconditional top oor response quantities, the unconditional response can be obtained
RMS displacement sxn can be obtained. The other important by Taylor series expansion, described in equation (Eq. (39)). This
quantity of interest is the unconditional RMS displacement of the unconditional response can now be used in the evaluation of
isolator sxb , which can also be obtained similarly. unconditional crossing rate using Eq. (30) and subsequently the
unconditional probability of failure using Eq. (29). Finally, the BI
system optimization as dened by Eq. (41) is redened as follows:
3. Optimization of BI system " #
nq
Find F 0 to min: : f PF 1exp  i T d
Presently, the response quantities of interest are the top oor i1
" (
!) #
RMS displacement sxn . From the description of the bilinear 1 2i
isolator model it is realized that the characteristic parameters of 1exp  s _ exp  2 Td
sX n X n 2sX_
a BI system are the isolation time period T b , coefcient of viscous n

damping b and the normalized yield strength F 0 of the isolator. 42


It has generally been observed that with increasing value of the The optimization problem as presented herein can be solved
two parameters T b ; b , the responses of isolated structure using the available nonlinear optimization algorithm. In the present
decreases monotonically. However, there exists optimal value of study, the pattern search technique [46] has been adopted. This
F 0 for which the response attains minimum. Thus the normalized particular technique is a global optimization technique and does not
yield strength F 0 is taken as the design variable in the rely on the gradient information of the objective function.
optimization study. It may be pointed out here that the absolute accelerations govern
the design if the safeties of non-structural components are of major
3.1. Conventional stochastic optimization of BI system concern, whereas when the overall design of the structural system is
of concern, the displacement values at various oor levels will be
The conventional optimization problem, so dened, is usually more important. In the present study we have concentrated on this
transformed into a standard nonlinear programming problem. One of second aspect. The limit states with respect to inter-story drifts will
the much used approaches is to minimize the RMS response of the be more important in this regard. For simplicity, we have studied the
isolated structure. The objective function is normally taken as the probability of failure for maximum roof level displacement para-
RMS displacement at the top oor (Eq. 25) or exceedence of some meter. However, the formulation presented here is generic in nature
predened serviceability or strength limit state by the structural and can be easily applied for multiple failure events to consider the
performance variables (Eq. (26)). The response or reliability being inter-story drift as the criteria and also the absolute acceleration as
nonlinear functions of the design variables requires the solution of a the criteria in the rst passage failure probability evaluation.
nonlinear optimization problem, termed as Stochastic Structural
Optimization (SSO). More details may be found elsewhere
[10,43,46]. The response based optimization is stated as 4. Numerical illustrations
Find F 0 to minimize sxn i
If the optimization problem is formulated to minimize the A ve storied shear building is taken up to illustrate the
failure probability instead, then the optimization is based on proposed RBDO of BI system for mitigating seismic vibration effect
minimizing the failure probability PFj. These optimization under system parameter uncertainty. The mass, stiffness and
problems are nonlinear unconstrained optimization and can be damping ratio mi ; ki ; i of each oor of the building are assumed
solved by available optimization algorithms. to be identical for simplicity. However, applicability of the pro-
posed approach is not restricted to such simplication and can
3.2. Optimization of BI system considering parameter uncertainty tackle different combination of these parameters. The value of the
damping ratio of the primary structure is assumed to be 2%,
It can be noted that the optimization problems described so far unless specically mentioned.
The base isolator is characterized by
are based on the stochastic response statistics assumes that the using the mass ratio mb =mi , taken as 1. The viscous damping
S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133 129

ratio for the LRB is considered as b 5%. The values of parameters To further emphasize the importance of RBDO approach, the
characterizing the stochastic earthquake load are as: g 5 rad/s, optimal yield strengths are obtained based on the criteria of
g 0:6 and S0 0.05 m2/s3. The duration of the seismic ground response and failure probability minimization. Then, adopting
motion T d is taken as 30 s for evaluation of peak factor. The these two different optimum yield strengths, the probabilities of
barrier level for the top oor displacement is assumed to be failure of the isolated structure with top oor displacement limit
0.045 m. The random parameters considered in the study are state are obtained for several scenario of parameter variations in
mentioned in the random design vector fg and are assumed to structural parameters and seismic loading. The results are shown
be statistically independent and follow normal distribution. The in Fig. 3ac. It can be observed from these plots that the failure
coefcient of variation (COV) of these random variables is assigned probabilities associated with the response based optimal design
as 1015% for parametric study. are underestimated compared to the estimate from the risk based
The variation of the top oor RMS displacement is investigated optimal design for all ranges of system parameters. Thereby,
with respect to different period of isolation and viscous damping response based criteria might lead to an unsafe design.
of the isolator and it is observed (though not shown herein) that It has been revealed through numerical simulation that the
the variations are monotonic. With increasing time period of the optimal yield strength values differ depending on whether the
isolator and damping, the oor displacement decreases (isolation objective is response minimization or minimizing the rst passage
efciency increases). Thus, like yield strength, there is no such failure probability. Further, it was also observed that the optimal
optimal value for the isolation time period and viscous damping of values are relatively insensitive to the parameter uncertainty and
the LRB. Hence, the optimal design parameter studied is the yield not much disparity among the optimal yield strength values are
strength of the LRB only. observed among the conditional and unconditional RBDO. None-
The top oor RMS displacement and the rst passage failure theless, signicant disparities among the failure probabilities are
probability of the isolated structure are shown with respect to observed corresponding to different level of parameter uncer-
the normalized isolator yield strength in Fig. 2a and b. The tainty. Keeping this in view, in the next, the optimal parameters
conditional and unconditional RMS displacements, along with are shown for response based optimization and reliability based
the conditional and unconditional failure probabilities corre- optimization; whereas, the failure probabilities from the respec-
sponding to the top oor displacement limit state are obtained. tive RBDO-optimal system are presented for different level of
The results indicate that the yield strength of the isolator possess parameter uncertainty.
optimum value for ensuring minimum oor displacement or The optimum values of isolator yield strength, obtained by
probability of failure. It is noteworthy that the optimal F 0 minimizing the unconditional failure probability are shown in
(Fig. 2a) obtained by minimizing the RMS displacement is not Fig. 4a with respect to varying superstructure exibilities. The
necessarily identical to the optimal F 0 obtained by minimizing the optimum yield strengths obtained by minimizing the uncondi-
failure probability (Fig. 2b). This implies that minimizing the tional top oor displacement are also shown in the same gure for
response does not necessarily comply to risk minimization. comparison. The differences between the optimal parameters
Fig. 2b further indicates the failure probabilities are more sensitive estimated based on the two objective functions (response and
to the parameter uncertainty comparing to the responses itself. reliability/risk based criteria) are denite for all time period
This is obvious as the rst passage failure involves uncertainty in considered in the present study. Fig. 4b shows the variation of
the two response quantities, i.e. in the RMS top oor displacement failure probability of optimal LRB considering the unconditional
and velocity as well. failure probability. It is observed that the optimal yield strength is

Fig. 2. (a) RMS displacement of the top oor and (b) failure probability for the top oor displacement limit states with the normalized yield strength of the isolator.

Fig. 3. Probability of failure with respect to top oor displacement for (a) different exibilities of the superstructure, (b) different damping of the superstructure and
(c) earthquake intensity.
130 S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133

Fig. 4. (a) Comparison of response and reliability based optimal isolator yield strength for varied superstructure exibility and (b) probability of failure for different COV
of the random variables.

Fig. 5. (a) Comparison of response and reliability based optimal isolator yield strength for varied superstructure damping and (b) probability of failure for different COV
of the random variables.

Fig. 6. (a) Comparison of response and reliability based optimal isolator yield strength for varied isolation time period and (b) probability of failure for different COV of the
random variables.

relatively insensitive to the superstructure exibility, but the are notable deviations in risk based optimization compared to the
failure probability varies considerably and increases for increas- conventional response based optimization. The failure probabil-
ingly exible structures. The optimum yield strength obtained by ities correspond to the reliability based optimal design is shown in
minimizing the conditional probability of failure assuming deter- Figs. 6b and 7b. The enhanced risk associated with the system
ministic system parameters indicates the possibility of obtaining a considering uncertain parameters is obvious. Expectedly, the risk
less probability of failure, which in reality will produce larger increases with increasing level of system uncertainty.
values. This is a potential problem of deterministic analysis and Fig. 8a shows the optimal design parameters for varying
warns its application to the BI system without proper considera- intensity of earthquake, characterized by the white noise inten-
tion of uncertainty. sity S0 . The optimal yield strength is higher for mitigating
The optimal yield strength based on reliability and response are earthquake of higher intensity. This trend is a direct consequence
now studied in Fig. 5a and b with respect to varied superstructure of nonlinear forcedeformation characteristic of the isolator
damping. Following the previous trend, the optimal yield strength bearing. The difference in the reliability and response based
does not change signicantly with the superstructure damping, optimal design is noteworthy. The discrepancy increases with
yet the enhanced risk associated with uncertain system parameter increasing level of seismic intensity. The respective risks asso-
is notable. As expected, with increasing damping the failure ciated with optimal parameters are shown in Fig. 8b. It is observed
probability decreases. that enhanced risks are associated with higher level of uncertainty
The optimization results are further shown for different isola- and the failure probability corresponding to optimal design vary
tion time period and damping in Figs. 6 and 7, respectively. There signicantly with increasing intensity of earthquakes.
S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133 131

Fig. 7. (a) Comparison of response and reliability based optimal isolator yield strength for varied isolation damping and (b) probability of failure for different COV of the
random variables.

Fig. 8. (a) Comparison of response and reliability based optimal isolator yield strength for varied intensity of earthquake and (b) probability of failure for different COV of the
random variables.

Fig. 9. Effect of bearing displacement on (a) optimal yield strength and respective (b) probability of failure for different values of the isolator displacement constraint.

The results presented so far focuses on the unconstrained case displacement constraint in Fig. 9a and b, respectively. It can be
where no consideration has been imposed on the maximum observed from Fig. 9a and b that the parameter uncertainty does
possible displacement of the isolator. The effect of the isolator not have signicant inuence on the optimal yield strength values
displacement constraint on the RBDO is now demonstrated. in the unconstrained regime. However, the disparities among the
Keeping in view that the displacement of isolator is indeed an optimal strength values in the constrained regime could be
important design aspect in an isolation system; the problem can signicant, particular, jumps among the values are noticed around
also be reformulated as a constrained nonlinear optimization 0.3 m. This is because the actual bearing displacement lies around
problem. For this, the respective unconstrained optimization this values and the activation or deactivation of constraint can be
problem (as stated in Eq. 42) should also incorporate the allowable switched by the variations of bearing displacement under para-
isolator displacement, restated as metric uncertainties. The effect on the respective failure probabilities
are shown in Fig. 9b and follow the general trend of enhanced failure
Find F 0 to minimize : f PF
probability with increasing level of uncertainty. Additionally, the
so that uxb d disparities among the conditional and unconditional failure prob-
where uxb is the unconditional peak displacement of the isolator, abilities steeply increase with increasingly lower value of allowable
obtained by incorporating the aspect of system parameter uncer- isolator displacement, i.e. for increasingly stringent constraint.
tainty and d is the maximum allowable isolator displacement The prior observations clearly indicate the difference among
governed by practical design considerations. the response based and the reliability based optimization and the
The optimal yield strength and respective probability of failures implications of parameter uncertainty. To exhibit precise estimate
are shown for different values of the allowable isolator of the discrepancies, the results are further provided in Table 1a.
132 S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133

Table 1a
Optimal yield strength for different values of isolator displacement constraint and level of parameter uncertainty.

Allowable bearing displacement Optimal yield strength Maximum difference (%)

Conditional Unconditional

Response based Reliability based COV 5% COV 10% COV 15%

0.50 0.0958 0.0934 0.0934 0.0934 0.0934 2.5


0.40 0.0958 0.0934 0.0934 0.0934 0.0941 2.5
0.35 0.1006 0.0967 0.0949 0.0954 0.0961 5.7
0.30 0.1018 0.1063 0.1065 0.0934 0.0935 8.25
0.25 0.1048 0.1049 0.1051 0.1057 0.1068 2.0
0.20 0.1409 0.1411 0.1414 0.1421 0.1437 2.0

Table 1b
Probability of failure for different values of constraint under varying level of uncertainty.

Allowable bearing displacement Failure probability Maximum difference (%)

Conditional Unconditional

Response based Reliability based COV 5% COV 10% COV 15%

0.50 0.00083 0.00082 0.00088 0.00110 0.00150 81.05


0.40 0.00083 0.00082 0.00088 0.00110 0.00150 80.94
0.35 0.00088 0.00083 0.00088 0.00110 0.00150 69.97
0.30 0.00090 0.00100 0.00110 0.00110 0.00150 65.79
0.25 0.00098 0.00098 0.00110 0.00130 0.00190 94.45
0.20 0.00670 0.00680 0.00730 0.00900 0.01270 89.55

The estimate of the conditional and unconditional failure prob- design of BI system without proper considerations of such para-
abilities pertaining to the optimal systems are shown in Table 1b. meter uncertainty. The formulation presented here involves linear
It may be noted from Table 1a that the optimal yield strengths perturbation based approximation of the responses around the
obtained by the reliability based optimization vary notably from mean values of the uncertain parameters. For larger level of
that of obtained by the response based optimization. The optimal uncertainty, alternative approach to linear perturbation analysis
strengths by the unconditional RBDO do not show signicant should be applied, which requires further study.
deviations from the conditional RBDO, except under the constraint
imposed on the bearing displacement. However, the parameter
uncertainty shows signicant inuence on the probability of Acknowledgments
failure corresponds to the optimal design as indicated in Table 1b.
We would like to sincerely acknowledge to the anonymous
Reviewer for his meticulous reading, constructive comments and
5. Conclusions valuable suggestion which has substantially helped in improving
the quality of the manuscript.
The reliability based optimization of BI system of protection is
studied for mitigation of vibration effect on structures subjected to
stochastic earthquake loading. The parameters uncertainty char- References
acterizing the mechanical model of the system and stochastic
earthquake load is incorporated in the response and reliability [1] Kelly JM. Aseismic base isolation: review and bibliography. Soil Dyn Earth-
quake Eng 1986;5:20216.
based optimization in the framework of total probability theorem. [2] Buckle I, Mayes R. Seismic isolation: history, application and performance-a
The associated failure probabilities for RBDO-optimal system are world view. Earthquake Spectra 1990;6(2):161201.
obtained based on the rst passage failure criteria. The results of [3] Symans MD, Constantinou MC. Semi-active control systems for seismic
protection of structures: a state-of-the-art review. Eng Struct 1999;21:46987.
RBDO are compared to the response based design to study the [4] Jangid RS, Datta TK. Seismic behavior of base-isolated buildings: a state-of-
effectiveness of the former one. Even though, the trend followed the-art review. Proc Inst Civil Eng, Struct Build 1995;110:186202.
by the optimization results obtained from both the approaches are [5] Constantinou MC, Tadjbakhsh IG. Hysteretic dampers in base isolation:
random approach. J Struct Eng (ASCE) 1985;111:70521.
mostly similar; there is a denite change in the optimum BI [6] Lin BC, Tadjbakhsh IG, Papageorgiou AS, Ahmadi G. Response of base isolated
conguration obtained by RBDO approach compared to that buildings to random excitation described by the C-P spectral model. Earth-
obtained by the response based optimization approach. It is seen quake Eng Struct Dyn 1989;18:4962.
[7] Jangid RS. Stochastic response of building frames isolated by lead-rubber
that the failure probability associated to the response based bearings. Struct Control Health Monit 2008;17:122.
optimization could be lower than the risk involved in the relia- [8] Jangid RS. Optimum damping in a non-linear base isolation system. J Sound
bility based optimization. This justies the need of RBDO approach Vib 1996;189:47787.
[9] Baratta A, Corbi L. Optimal design of base-isolators in multi-storey buildings.
over the response based optimization for design of BI system. It is
Comput Struct 2004;82(2326):2199209.
generally observed that the risk in case of RBDO approach [10] Nigam NC. Structural optimization in random vibration environment. AIAA J
signicantly increases due to system parameter uncertainty. 1972;10(4):5513.
Deterministic system provides considerably lower estimate of risk [11] Jensen HA. Design and sensitivity analysis of dynamical systems subjected to
stochastic loading. Comput Struct 2005;83:106275.
than what is actually involved. This might lead to a non- [12] Chaudhuri A, Chakraborty S. Sensitivity evaluation in seismic reliability
conservative design and could be a potential problem in optimum analysis of structures. Comput Methods Appl Mech Eng 2004;93:5968.
S.K. Mishra et al. / International Journal of Mechanical Sciences 75 (2013) 123133 133

[13] Schuller GI, Jensen HA. Computational methods in optimization considering [29] Chen SH, Song M, Chen YD. Robustness analysis of responses of vibration
uncertainties- an overview. Comput Methods Appl Mech Eng 2008;198(1):213. control structures with uncertain parameters using interval algorithm. Struct
[14] Yen YJ, Huang PC, Wan S. Modications on base isolation design ranges Saf 2007;29(2):94111.
through entropy-based classication. Expert Syst Appl 2009;36(3):491522. [30] Chen J, Weiqing L, Peng Y, Li J. Stochastic seismic response and reliability
[15] Jensen H, Setareh M, Peek R. TMDs for vibration control of system with analysis of base-isolated structures. J Earthquake Eng 2007;11(6):90324.
uncertain properties. J Struct Eng, ASCE 1992;118(2):328596. [31] Zhang Y, Web B, Liu Q. First passage of uncertain single degree-of-freedom
[16] Papadimitriou C, Katafygiotis LS, Au SK. Effects of structural uncertainties on nonlinear oscillations. Comput Methods Appl Mech Eng 1998;165:22331.
TMD design: A reliability based approach. J Struct Control 1997;1(4):6588. [32] Clough RW, Penzien J. Dynamics of structures. Singapore: McGraw Hill Book
[17] Taanidis AA, Beck JL, Angelides DC. Robust reliability-based design of liquid Company; 0-07-085098-4.
column mass dampers under earthquake excitation using an analytical [33] International Code Council. International Building Code: 2000.
reliability approximation. Eng Struct 2007;29:352537. [34] International Conference of Building Ofcials, Uniform Building Code, Califor-
[18] Roy BK, Chakraborty S, Mishra SK. Robustness optimum design of base nia; Whittier: 1997.
isolation system in seismic vibration control of structures under uncertain [35] Bouc R., Forced vibration of mechanical systems with hysteresis, In: Proceed-
bounded system parameters. J Vib Control 2012;20:4. ings of the 4-th conference on nonlinear oscillation, Prague, Czechoslovakia,
[19] Debbarma R, Chakraborty S, Ghosh S. Optimum design of tuned liquid column 1967; p. 315.
dampers under stochastic earthquake load considering uncertain bounded [36] Wen YK. Method of random vibration of hysteretic systems. J Eng Mech
system parameters. Int J Mech Sci 2010;52:138593.
(ASCE), 102; 24963.
[20] Debbarma R, Chakraborty S, Ghosh S. Unconditional reliability based design of
[37] Atalik TS, Utku S. Stochastic linearization of multidegree of freedom nonlinear
tuned liquid column dampers under stochastic earthquake load considering
systems. Earthquake Eng Struct Dyn 1976;4:41120.
system parameter uncertainties. J Earthquake Eng 2010;14(7):97088.
[38] Roberts JB, Spanos PD. Random vibrations and statistical linearization. New
[21] Marano GC, Sgobba S, Greco R, Mezzina M. Robust optimum design of tuned mass
York: Wiley; 1990.
dampers devices in random vibrations mitigation. J Sound Vib 2008;313:47292.
[39] Kanai K. Semi-empirical formula for the seismic characteristics of the ground,
[22] Marano GC, Greco R, Sgobba SA. Comparison between different robust
Bulletin of Earthquake Research Institute. University of Tokyo
optimum design approaches: application to tuned mass dampers. Probabilistic
1957;35:30925.
Eng Mech 2010;25:10818.
[40] Tajimi HA. Statistical method of determining the maximum response of a
[23] Mishra SK, Chakraborty S. Performance of a base-isolated building with
system parameter uncertainty subjected to a stochastic earthquake. Int J building structure during an earthquake, In: Proceedings of the 2nd World
Acoust Vib 2013;18(1):719. conference on earthquake engineering. 1960; 11: p. 781798.
[24] Juhn G, Manolis GD. Stochastic sensitivity and uncertainty of secondary [41] Fan FG, Ahmadi G. Nonstationary KanaiTajimi models for El-Centro 1940 and
systems in base-isolated structures. J Sound Vib 1992;159(2):20722. Mexico city 1985 earthquakes. Probabilistic Eng Mech 1990;5(4):17181.
[25] Kawano K, Arakawa K, Thwe M, Venkastaramana K. Seismic response evalua- [42] Hurtado JE, Barbat AH. Equivalent linearization of the BoucWen hysteretic
tions of base-isolated structures with uncertainties, structural stability and model. Eng Struct 2000;22:112132.
dynamics, In: Proceedings of the Second International Conference, Singapore; [43] Crandall S, Mark W. Random vibration in mechanical system. New York:
December 1618, 2002. p. 889894. Academic; 1963.
[26] Scruggs JT, Taanidis AA, Beck JL. Reliability-based control optimization for [44] Lutes DL, Sarkani S. Random vibrations, analysis of structural and mechanical
active base isolation systems. Struct Control Health Monit 2006;13:70523. systems. Burlington, USA: Elsevier; 2004.
[27] Zhou J, Wen C, Cai W. Adaptive Control of a base isolated system for protection [45] Zhao YG, Ono T, Idota H. Response uncertainty and time-variant reliability
of building structures. J Vib Acoust 2006;128(2):261. analysis for hysteretic MDOF structures. Earthquake Eng Struct Dyn
[28] Zhou J, Wen C. Control of a hysteretic structural system in base isolation 1999;28:1187213.
scheme, adaptive backstepping control of uncertain systems. Lect Notes [46] Davidon WC. Variable metric method for minimization. SIAM J Optim 1991;1
Control Inf Sci 2008;372:199213. (1):117.

Anda mungkin juga menyukai