Journal of Constructional Steel Research 98 (2014) 100113

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Effect of buckling-restrained brace model parameters on seismic

structural response
Quan Gu a,c, Alessandro Zona b,, Yi Peng a, Andrea Dall'Asta b
a
b
c

School of Architecture and Civil Engineering, Xiamen University, PR China

School of Architecture and Design, University of Camerino, Italy
Key laboratory of concrete and pre-stressed concrete structures Ministry of Education, Southeast University of China, PR China

a r t i c l e

i n f o

Article history:
Received 20 September 2013
Accepted 28 February 2014
Available online 28 March 2014
Keywords:
Braced frames
Buckling-retrained braces
Direct differentiation method
Nonlinear constitutive model
Seismic analysis
Sensitivity analysis

a b s t r a c t
This paper illustrates the derivation of response sensitivities for a hysteretic model specically developed for
buckling-restrained braces (BRBs) in order to provide a tool that can be used to evaluate the effect of BRB constitutive parameters on structural response as well as a tool in gradient-based methods in structural optimization,
structural reliability analysis, and model updating. The adopted BRB model, shown in an earlier study to give accurate predictions of the experimental behaviour of BRBs, is differentiated with respect to its material constitutive parameters using the direct differentiation method (DDM) and the obtained response sensitivities are
validated by comparisons with the nite difference method (FDM). Results for a case study consisting of a
steel frame with BRBs subjected to seismic input are reported to illustrate the inuence on global and local structural response quantities of the BRB constitutive parameters. In addition, the derived response sensitivities are
used in a simulated nite element model updating problem to show the efciency of DDM over FDM. This
work opens the way to many applications and potentialities such as sensitivity analysis of complex BRB design
solutions, performance-based selection of optimal BRB properties, development and use of optimization-based
design procedures.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Buckling-restrained braces (BRBs) are used in seismic areas for both
new construction and rehabilitation of existing structures due to their
effective and stable energy dissipation capacity, as shown in many experimental and numerical studies, e.g., [123]. Although different typologies are available, basically a BRB is made of a ductile core and an
external sleeve that precludes global buckling of the core in compression. An unbonding material and/or a small gap between the steel
core and the external sleeve is provided to avoid the transfer of axial
force between the two components. Because buckling is prevented during the compression phase, the BRB core can yield both in tension and
compression; thus, it dissipates seismic energy through the hysteretic
behaviour of the core material. However, the global behaviour of the
BRB can't be relied upon to replicate the local behaviour of its core material. Tensioncompression asymmetry is observed with force resisted
in compression about 10 to 15% higher than forces resisted in tension
due to friction between core and sleeve in compression fostered by a
limited core buckling made possible by the clearance left between the
Corresponding author at: University of Camerino, School of Architecture and Design,
Viale della Rimembranza, Ascoli Piceno 63100, Italy. Tel.: +39 0737 404280; fax: +39
0737 404272.
E-mail address: alessandro.zona@unicam.it (A. Zona).

http://dx.doi.org/10.1016/j.jcsr.2014.02.009
0143-974X/ 2014 Elsevier Ltd. All rights reserved.

core and the sleeve [1,3,14]. Other aspects requiring attention in

model-based analysis of BRBs are the description of their isotropic hardening that can be signicant due to the capacity of sustaining stable hysteresis loops at high strains [13], and the calculation of the cumulative
plastic deformation needed for verications based on BRB capacity
models [2426]. In order to incorporate within a mathematically simple
and physically consistent formulation the tensioncompression asymmetry, the hardening behaviour as observed in experimental tests,
and the direct evaluation of plastic strain, a constitutive elastoplastic
model specically developed for BRBs was presented in [27]. Such a
constitutive model requires only one internal variable (plastic strain),
has a simple physical interpretation, allows straightforward control of
the dissipative properties of the model and direct computation of the response quantities related to failure and dissipated energy as derived
from the plastic strain. Various experimental test results available in
the literature were compared to the response results obtained using
the proposed model, showing good predictions of the experimental behaviour of different BRBs subjected to symmetric and non-symmetric
cyclic loadings with variable amplitudes [27].
The availability of a proper cyclic model for BRBs in a nite element
(FE) software for nonlinear dynamic analysis is important for an accurate seismic assessment of the structural behaviour of constructions
equipped with BRBs, especially when BRBs are the only lateral resisting
components, as is the case in braced non-moment resisting steel frames.

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

In these situations, the post-elastic stiffness of BRBs is limited and hardening might not be sufcient to avoid soft-storey formations [2830]. In
addition, given that various BRB typologies are available, each with its
cyclic behaviour dened by a specic set of constitutive parameters, it
is important, together with accurate response models, to have qualitative and quantitative information on the inuence of such parameters
on the seismic response of the structure. This information can be
efciently obtained through response sensitivity analysis [31], a very efcient tool for gaining deeper insight into the effect and relative importance of a large number of modelling and design parameters that
otherwise would require extensive parametric analyses. In addition, response sensitivities have useful applications in gradient-based methods
used in structural optimization [32], structural reliability analysis [33],
and model updating [34]. A rst application of response sensitivity analysis for the study of the behaviour of steel frames with BRBs was presented in [35] where response sensitivities were computed with
respect to the brace section area (geometric parameter at the element
level) and used to gain insight into the tendency to soft storey formation, as inuenced by brace over-strength distributions. Conversely, response sensitivities with respect to the BRB constitutive parameters
(material level) remained unexplored despite their important applications. Thus, the objective of this paper is to provide a sensitivity-based
tool for studying the inuence of the BRB constitutive parameters on
the seismic response of structures equipped with such bracing systems.
To this end, the response sensitivities of the BRB cyclic model presented
in [27] are derived with respect to its constitutive parameters (elastic
modulus, initial yield stress, ultimate stresses in tension and compression, hardening moduli in tension and compression, hardening rates in
tension and compression, elastic-to-plastic transition shape parameters
in tension and compression) using the direct differentiation method
(DDM), i.e., response gradients are obtained by analytically differentiating the governing equations, a method proven accurate and efcient especially for nonlinear static and dynamic FE response sensitivity
analysis [31,3638]. Both the BRB model [27] and the response sensitivities illustrated in this paper are implemented into the open source FE
software OpenSees [39] to make them available to the structural engineering community, taking advantage of the fact that OpenSees provides capabilities for DDM-based response sensitivity analysis [40,41].
In this paper the analytical formulation of the elastoplastic constitutive model for BRBs [27] is briey reviewed, its time-explicit and timeimplicit integration algorithms are presented, and the derivation of response sensitivities to the material constitutive parameter is illustrated.
Response sensitivity results obtained using DDM are validated by comparisons with those obtained using the nite difference method (FDM)
[31]. Results for a braced non-moment resisting steel frame under seismic excitations are reported to illustrate the use of the presented approach to evaluate the inuence of the constitutive parameters
considered independent for each BRB on the predicted global and
local structural response quantities. In addition, response sensitivity results are used in the considered case study in a simulated FE model
updating problem to show the efciency of DDM over FDM.

2. BRB model for response analysis

2.1. Time-continuous formulation
The BRB constitutive behaviour is modelled with a rheological scheme
(Fig. 1) consisting of a spring 0 (with stiffness E0) in series with a friction
slider (with one internal variable, i.e., its plastic deformation pl) in parallel with a spring 1 (with stiffness tE0 in tension and cE0 in compression) from which the following evolution laws are derived [27]:



t el t E0 el t E0 t pl t

101

Elastic spring 1

Elastic spring 0

Friction slider

Elastic component

Plastic component

Fig. 1. Elastoplastic rheological model.

8
< t E0 pl t if t N0 and t N0
1 t c E0 pl t if t b0 and t b0
:
0
otherwise




t  pl t 

where is the stress resisted by the model, el is the stress and el the
strain in spring 0, 1 is the stress in the spring 1, the cumulative
plastic deformation, and a superimposed dot represents the derivative
with respect to time. The plastic ow rules furnishing the time evolution
of the internal variable pl are [27]:
8

 t t  t
>
>


1
>
>
 t
>
>


>
<  y;t t 

pl t  t 1 t  c
>
t

>
>
>
>  y;c t 
>
>
:
0

if t N0 and t N0
if t b0 and t b0

otherwise

where y,t and y,c are the yield stresses in tension and compression respectively, is a positive nondimensional constant that controls the
trend of the transition from the elastic to the plastic range, i.e., a higher
value of fosters the tendency to a sharper transition from the elastic
to the plastic range, and a lower value of gives a smoother and more
progressive transition from the elastic to the plastic range (subscript t
refers to tension and c to compression) [27]. The hardening rules that
give the increments of the yield stresses in tension and in compression
are nonlinear functions of the cumulative plastic deformation and are
given by [27]:
!



t t
y;t t y max;t y0 exp
r;t r;t

!



t t
y;c t y max;c y0 exp
r;c r;c

where y0 is the initial yield force, ymax the maximum yield force for the
fully saturated hardening condition, r is a positive non dimensional constant that inuences the rate of hardening, i.e., a higher value of r results
in a slower hardening [27]. Damage models, e.g. softening after necking,
plastic fatigue and fractures, are not included in order to keep the analytical formulation as simple as possible for the sake of numerical efciency
and to allow its subsequent analytical differentiation with respect to the
sensitivity parameters.
The above time-continuous analytical formulation must be
discretized when used within time-discrete solution frameworks, as is
the case in nonlinear dynamic nite element analysis of structural

102

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

systems. Various integration algorithms are available for elastoplastic

constitutive laws, e.g., [42]. Hereafter, two complementary approaches
are briey described to provide the constitutive equations that are afterwards derived within the time-discrete scheme for a consistent nite
element response sensitivity analysis.







c


 t
 n1 E0 t n pl t n c E0 pl t n1 pl t n 
  t


pl t n 



n

t
j t j


y0 y max;c y0 1 exp n1 pl n


r;c
if t n b0 and t n b0

14b

2.2. Explicit time-discrete formulation

In the explicit time-discrete formulation, the response in the BRB
constitutive model at time step n is computed from the material state
of the previous time step n 1 and the assigned strain at time step n.
The algorithm is based on the following sequence. The discrete strain
increment is:
t n t n t n1

afterward the plastic strain increment is obtained:

8

 t
>
t E0 pl t n1  t
>
>  n1
>
  t n if t n1 N0 and t n1 N0
>
>


y;t t n1
>
<

pl t n  t n1 c E0 pl t n1  c
>

  t n if t n1 b0 and t n1 b0
>
>

y;c t n1
>
>
>
>
:
0
otherwise

8
then the updated plastic strain and the current stress is computed:
pl t n pl t n1 pl t n



t n t n1 E0 t n pl t n

10

nally the cumulative plastic strain and the current yielding stress in
tension and compression are updated:




t n t n1 pl t n 
11
"
!#



t n
y;t t n y0 y max;t y0 1 exp
r;t

12

"
!#



t n
y;c t n y0 y max;c y0 1 exp
:
r;c

13

The explicit time-discrete formulation requires few calculations at

each time step, but this advantage is usually limited by the fact that
very small time steps are needed to achieve adequate accuracy and
avoid numerical instability problems. This limitation can be overcome
by enhancing the explicit approach; for example, by using automatic
sub-stepping and error control [43], or using the implicit approach described hereafter as an alternative.
2.3. Implicit time-discrete formulation
In the implicit time-discrete formulation, the response in the BRB
constitutive model at time step n is computed from the material state
and assigned strain at the same time step n. The rst step of the implicit
algorithm requires the computation of the plastic strain increment at
time step n:






t


 t
 n1 E0 t n pl t n t E0 pl t n1 pl t n 
  t


pl t n 



n

t
j t j


y0 y max;t y0 1 exp n1 pl n


r;t
if t n N0 and t n N0

14a

pl t n 0 otherwise:

14c

Since the above equations are nonlinear function of the unknown

pl(tn ), an iterative procedure is required. In this study, the
safeguarded Newton method combined with a bi-section method is
used. Once that pl(tn) is obtained, the updated plastic strain and
the current stress are computed as in the explicit formulation, i.e.,
using Eqs. (9) to (13). Such a implicit time-discrete formulation, implemented in OpenSees and made available to its users, is adopted in
this work and response sensitivities derived as described in the
following section.
3. BRB model differentiation for response sensitivity analysis
3.1. Direct differentiation framework for response sensitivity analysis
The sensitivity of a response quantity r (e.g., displacement, strain,
stress) with respect to a sensitivity parameter (e.g., material constitutive parameter), is dened as the rst derivative dr/d. The DDM-based
FE response sensitivities are computed after convergence of the iterative
procedure adopted at each time step of the nonlinear FE response analysis. This requires consistent differentiation of the FE algorithm for the
response-only computation, with respect to each sensitivity parameter
. Consequently, the response sensitivity computation algorithm involves the differentiation of responses at various hierarchical levels of
FE response analysis: the structure level, the element/section level,
and the material level. Details about the DDM-based sensitivity formulations in displacement-based, force-based and mixed FE methods can
be found in the literature, e.g., [37,38,44,45]. For displacement-based
FE response sensitivity analysis using DDM, after spatial discretization
using the FE method and time discretization using the Newmark timestepping method [46], the following sensitivity equation must be solved
in the unknown dun/d:



1

dun
C

K
M

n
t
d
t 2



d eFn R n 
1 dM
dC

un


2
d
un1
t d t d

15

where un is the vector of nodal displacements at the current time

step n, M is the mass matrix, C is the damping matrix, K denotes
the consistent tangent (static) stiffness matrix of the structure, and
R is the history-dependent internal resisting force vector, e
Fn is dened as




1
1
1 ::
eF F M
un1
u

u n1 1
n
n
2 n1
t
2
t 




::
u
u
u
1
t 1
:
16
C
t n1
n1
2 n1
::
Fn is the applied dynamic load vector, u n1 and un1 denote, respectively, the rst and second derivatives of un 1 with respect to
time, the parameters and control the accuracy and stability of
the Newmark integration algorithm [46], t is the time integration
step, the subscript n indicates that the quantity is evaluated at the
discrete time tn. If the mass matrix and the damping matrix do not
depend on the sensitivity parameter, as is the case in the considered
study, then the solution of the sensitivity equation requires the

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

computation of:

Rn 



un

103





U t n1 E0 t n pl t n t E0 pl t n1 pl t n
25

0
1



Z
n ; pl;n ; 
T
B
C
A@ B
 de A

e1

nel

17

nel


13
2
0





t n1 pl t n 
A5:
V y0 y max;t y0 41 exp@
r;t

26

where A denotes the stiffness assembly operator, nel represents the

e1

number of elements in the FE model, and B is the strain displacement

transformation matrix. In fact, all other terms in the sensitivity equation are already determined during the solution of the response analysis problem at the considered time step.
3.2. Direct differentiation of the implicit time-discrete formulation of the
BRB constitutive model
Only the conditional stress sensitivity
j is required to compute

n

Eq. (17) and hence to solve Eq. (15) for the unknown dun/d. However,
in the following section the more general derivation of the unconditionis presented since the conditional stress sensitival stress sensitivities

n
ities can be obtained as a special case by setting
0. To compute the

stress sensitivities, the implicit formulation of the BRB constitutive

model is analytically differentiated with respect to the sensitivity parameter , which can be any of its ten constitutive parameters, i.e., E0,
y0, ymax,t, ymax,c, t, c, t, c, r,t, and r,c. The rst point to be solved
is the computation of the sensitivity of pl(tn) given in Eqs.
(14a)(14c) with respect to . Since pl(tn) is an implicit function of
, the derivative can be obtained through the steps described in the
equations from Eqs. (18) to (26).
If (tn) N 0 and (tn) N 0 then:

pl t n

1 H

18

H t jAj

t 2

h
i
2
At n E0 E0 t =V UF=V

 t
2

n
At n C=VUD=V
jA j t

ln jAjt n t

19

t 2

G t jAj

0:

27

Once the sensitivity of the plastic strain increment pl(tn) with respect to is obtained, the sensitivities of plastic strain, stress, cumulative
plastic strain, and current yield stress in tension and compression are
derived as:
pl t n

pl t n1

pl t n

28


t n t n1 E0

t n pl t n

!
t n pl t n

E0

 t
t n t n1
pl n

sgn pl t n

29

30

A U=V

22


13
0
!2





t n1 pl t n 
4
@
A5 y max;t y0

1 exp
D
r;t


1"
0
#



 1
t n1 pl t n 

t n1 1


r;t
A
exp@
t n1 pl t n  2
r;t
r;t

r;t
y max;t

y0

y max;t

y0

31

if (tn) b 0 and (tn) b 0

21


t n1 E0

t n pl t n t pl t n1 t pl t n
C


pl t n1
t n
t
E0
t E0
E0 pl t n1 pl t n

!"

!#
t n

1 exp

r;t

0
1
r;t
! t n




r;t
n
t n B
C
B
C
y max;t y0 exp
A
r;t @
2r;t

y;t t n

20

and

y0

pl t n

if (tn) N 0 and (tn) N 0

with:

jA j

Otherwise if (tn) b 0 and (tn) b 0, then in the above Eqs. (18) to

(26) ymax,t, t, t, and r,t are replaced by ymax,c, c, c, and r,c, respectively. If (tn) (tn) b 0, then:

y0

23

1
0







 1
t n1 pl t n 
Asgn pl t n
F y max;t y0 exp@
r;t
r;t
24

!"

!#
t n
r;c

0
1
r;c
! t n


r;c
n
t n B
C
B
C:
y max;c y0 exp
A
r;c @
2r;c

y;c t n

y0

y max;c

y0

1 exp

32

The conditional stress sensitivity

j can be also computed from
n
t n1
Eqs. (18) to (29) by simply setting t n
Eqs. (20), (22),
n
and (29). (22). Then R
j
can
be
computed
from
Eq. (17), and nally
un
the unknown dun/d is solved by Eq. (15). After dun/d is solved, the
strain sensitivity can be obtained by using the relationship between
nodal displacement and strain. Then, the above Eqs. (18) to (30) are
computed again to get the unconditional stress sensitivity, which is
required by the sensitivity analysis of the next time step. It is noteworthy that the above derivations benet from the smoothness of the
presented BRB model, a feature that gives many advantages both in
the computation of response sensitivities and in the convergence of
gradient-based methods, e.g., optimization method using the computed
sensitivities [47].

104

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

(a)

(b)

Table 2
Case study: BRB material properties.

htot = 4h = 13.6 m

Non dissipative
elastic segments
Dissipative
elastoplastic
segment

h = 3.4 m

Ld
Ly

b = 8.0 m
Fig. 2. Benchmark problem: (a) 4-storey model with bracing system and equivalent
column; (b) BRB element.

E0

y0

ymax,t

ymax,c

r,t

r,c

(GPa)

(MPa)

(MPa)

(MPa)

()

()

()

()

()

()

210

275

453.75

508.20

0.80

0.60

0.01

0.01

0.10

0.15

matrix, and 5% of the critical damping assigned to the rst and second
vibration modes.
The seismic input is given by a set of 28 natural ground motions from
the PEER Ground Motion Database [50], as described in [35] and scaled
so that the elastic response spectra for the records match the Eurocode
8 elastic spectrum at the rst natural period T1 of the structural system
(T1 = 0.746 s).
4.2. Sensitivity analysis for evaluating the inuence of the BRB constitutive
parameters

4. Application examples and sensitivity validation

Examples of application of the derived response sensitivities with respect to the BRB constitutive model parameters and relevant numerical
validation are illustrated using as benchmark problem a 4-storey steel
V-bracing system with BRBs as depicted in Fig. 2a. Such bracing system
is one of the four equal bracing systems that are the only seismic resistant components for each direction in a realistic 4-storey steel frame
described in [35,48] and designed using the elastic spectrum type 1 of
Eurocode 8 [49] for ground type B with design ground acceleration
ag = 0.35 g as seismic input. The considered benchmark bracing system
is connected to a side column representing the gravity-resisting steel
frame pertinent to the bracing system with equivalent stiffness, mass,
and loads. The columns of the bracing system and the side column
equivalent to the vertical-resisting steel frame are continuous with
pinned beam-to-column and column-to-base connections. The yield
length Ly of the BRBs is one third of their overall length Ld (Fig. 2b).
The design yield stress of the BRBs is 275 MPa (European steel grade
S275), columns and beams are made of steel with design yield stress
equal to 355 MPa (European steel grade S355). Geometric data are
given in Table 1 while other data not provided here can be found in [35].
A plane FE model of the testbed structure is dened in OpenSees
[39], with columns and beams modelled using linear elastic Euler
Bernoulli frame elements having one element for each tract between
two consecutive nodes. Each BRB is modelled using three connected
beam elements (Fig. 2b) with the non-dissipative parts assumed linear
elastic and the dissipative central tract having axial forcedeformation
relation described by the adopted BRB model with constitutive
parameters given in Table 2 and momentcurvature relationship linear
elastic. Nonlinear dynamic analyses are performed using the constant
average acceleration method, i.e., the Newmark method with = 0.5
and = 0.25 [46], with constant time step t = 0.01 s, in conjunction
to the NewtonRaphson iterative procedure. A global damping for the
structure was included using the Rayleigh model [46], with the damping
matrix proportional to the mass matrix and initial tangent stiffness
Table 1
Case study: geometric data of the 4-storey bracing system and equivalent column.

Storey
#
4
3
2
1

Bracing system columns

BRBs

Equivalent column

Area

Moment of inertia

Core area

Area

Moment of inertia

(cm2)

(cm4)

(cm2)

(cm2)

(cm4)

58
101
141
192

4099.3
12,733.4
35,151.8
45,760.0

15.55
24.47
30.44
35.17

406
587
787
1069

17,637.3
29,457.9
70,325.6
91,876.7

The monitored response quantities are the roof horizontal displacement uroof (top node of the left column of the bracing system) and the
plastic strain pl,BRB1 in the left BRB at ground oor, giving global information on the overall structural response and local information on the
behaviour of the considered BRB, respectively. Detailed results are presented for two accelerograms of the considered set [35], i.e., (i) 1989
Loma Prieta, Fremont Mission San Jose station, NGA# 762; (ii) 1999
Chi-Chi Taiwan, CHY028 station, NGA# 3268, that gave quite different
residual interstorey drifts and excursions of the BRBs in the plastic
range, as shown in Figs. 3 and 4. Results averaged over the entire set
of accelerograms are presented afterward.
In order to compare sensitivities of different response quantities
with respect to different sensitivity parameters, various normalized formats can be adopted, e.g., [38,44]. Two non-dimensional formats are
compared here, namely the normalized sensitivity sn which is dened
as:
sn r t ;

dr t
d r max

33

and the weighted normalized sensitivity swn which is dened as:

swn r t ;

drt r t
d r max r max

34

with
r max max jr t j

35

and r = global or local response quantity, e.g., r = uroof and r = pl,BRB1

in the considered cases. The normalized format dened in Eq. (33)

1.2

uroof / htot 100

4.1. Benchmark problem

0.8
0.4
0.0
-0.4
-0.8

Loma Prieta
0

10

Chi-Chi
20

30

40

time (s)
Fig. 3. Benchmark problem: roof horizontal displacements under the Loma Prieta and
Chi-Chi accelerograms.

1.0

1.0

0.5

0.5

pl,BRB1 100

BRB1 / y0

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

0.0
-0.5
-1.0
-1.5

105

0.0
-0.5
-1.0
-1.5

Loma Prieta

Loma Prieta

-2.0

0.01

1.0

1.0

0.5

0.5

pl,BRB1 100

BRB1 / y0

-2.0
-0.01

0.0
-0.5
-1.0

10

20

30

40

0.0
-0.5
-1.0
-1.5

-1.5
Chi-Chi
-2.0
-0.02

-0.01

Chi-Chi
-2.0

0.01

10

20

30

40

time (s)

Fig. 4. Stressstrain cycles (left-hand side gures) and plastic strain (right-hand side gures) of the left BRB at ground oor under the Loma Prieta and Chi-Chi accelerograms.

allows a simple quantication of the effect of the sensitivity parameter

on the considered response quantity r, e.g., sn b 1 means that a relative
variation d/ of results in a less than proportional relative
variation dr/rmax of the response quantity r. However, the largest
value of dr/rmax could be attained when the response quantity is rather
small and very different from a response peak. Thus, sn might highlight
large variations of very small response values, generally of less interest
as compared to the variations of larger values of the response quantities.
To avoid these issues, the weighted normalized sensitivity, as dened in
Eq. (34), is a possible alternative.

Examples of the evolution of normalized and weighted normalized

response sensitivities are given in Fig. 5 (sensitivity of uroof with respect
to E0,BRB1 and y0,BRB1 of the left BRB at ground oor) and in Fig. 6 (sensitivity of pl,BRB1 with respect to E0,BRB1 and y0,BRB1 of the same BRB).
These gures give detailed information on the inuence of the considered constitutive parameters during the seismic response of the structural system. For example, in Figs. 5 and 6 it is observed that the
elastic modulus has similar inuence in terms of absolute peak value
in both accelerograms while the initial yielding stress has a larger inuence in the accelerogram with less excursions in the post-elastic range

0.2

0.2

Loma Prieta

Loma Prieta
swn(uroof ,)

sn(uroof ,)

0.1
0.0
-0.1
-0.2

= E0,BRB1
= y0,BRB1

= y0,BRB1
10

0.0
-0.1

= E0,BRB1

0.1

20

30

-0.2

40

0.2

10

20

Chi-Chi

40

Chi-Chi
snw(uroof ,)

0.1

sn(uroof ,)

30

0.2

0.0
= E0,BRB1

-0.1

0.1
0.0
= E0,BRB1

-0.1

= y0,BRB1

= y0,BRB1
-0.2

-0.2
0

10

20

time (s)

30

40

10

20

30

40

time (s)

Fig. 5. Normalized (left-hand side gures) and weighted normalized (right-hand side gures) response sensitivities of the roof displacement to E0 and y0 of the left BRB at ground oor
under the Loma Prieta and Chi-Chi accelerograms.

106

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

0.4

Loma Prieta

0.0

swn(pl,BRB1 ,)

sn(pl,BRB1 ,)

0.4

-0.4
= E0,BRB1

-0.8

Loma Prieta

0.0
-0.4
= E0,BRB1

-0.8

= y0,BRB1

= y0,BRB1
-1.2

-1.2
0

10

20

30

40

0.4

10

20

30

0.4

Chi-Chi

Chi-Chi

0.0

swn(pl,BRB1 ,)

sn(pl,BRB1 ,)

40

-0.4
-0.8

= E0,BRB1

= E0,BRB1

-0.4
= E0,BRB1

-0.8

= y0,BRB1

= y0,BRB1

-1.2

0.0

= y0,BRB1
-1.2

10

20

30

40

10

20

time (s)

30

40

time (s)

Fig. 6. Normalized (left-hand side gures) and weighted normalized (right-hand side gures) response sensitivities of the plastic strain of the left BRB at ground oor to E0 and y0 of the
same BRB under the Loma Prieta and Chi-Chi accelerograms.

(Loma Prieta) as compared to the case of larger damage (Chi-Chi). In

fact, if plastic strains are limited (as in the Loma Prieta case), then a
small increment (decrement) in the initial yielding stress can have a
non-negligible effect on the global and local response, reducing (increasing) the attained plastic deformations. On the other hand, if the
braces undergo signicant plastic strains (as in the Chi-Chi case), then
the structural response is not signicantly affected by small changes in
the initial yield stress.
A simpler way to present response sensitivity results is to compare
the extreme (positive and negative) values attained during the seismic
excitation. Examples of such a presentation of results are given in Figs. 7

to 12. The maximum and minimum weighted normalized sensitivities

of uroof with respect to the 10 constitutive parameters for each BRB,
i.e., 10 constitutive parameters for 8 BRBs give a total of 80 sensitivity
parameters for the considered case study, are presented in Figs. 7 and
8 for the Loma Prieta and Chi-Chi accelerograms, as well as averaged
over the set of accelerograms (Fig. 9). Similarly, the maximum and minimum weighted normalized sensitivities of pl,BRB1 with respect to the
same 80 sensitivity parameters, are presented in Figs. 10 and 11 for
the Loma Prieta and Chi-Chi accelerograms, and averaged over the set
of accelerograms (Fig. 12). The labels L and R in Figs. 7 to 12 identify
respectively the left BRB and the right BRB at each oor. Despite the

BRB #4 R

BRB #4 L

= ymax,c

= ymax,c

BRB #3 L

= ymax,t

= ymax,t
BRB #2 L

= y0
= E0
-0.6

-0.4

BRB #3 R

= E0

BRB #1 L

-0.2

0.2

0.4

0.6

-0.6

BRB #4 L

= c
= c
= c
= t
= t
= t

BRB #2 R

= y0

-0.4

BRB #1 R

-0.2

0.2

BRB #2 L

BRB #3 R

BRB #2 R
BRB #1 R

BRB #1 L

-0.6

-0.4

-0.2

swn(uroof ,)

0.2

0.4

0.6

BRB #4 R

= c
= c
= c
= t
= t
= t

BRB #3 L

0.4

0.6

-0.6

-0.4

-0.2

0.2

0.4

0.6

swn(uroof ,)

Fig. 7. Max and min weighted normalized sensitivities of the roof horizontal displacement to the 80 sensitivity parameter (Loma Prieta accelerogram).

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

107

BRB #4 R

BRB #4 L

= ymax,c

BRB #3 L

= ymax,t

= ymax,t
BRB #2 L

= y0
= E0
-0.6

-0.4

BRB #3 R

= ymax,c

= E0

BRB #1 L

-0.2

0.2

0.4

0.6

-0.6

BRB #4 L

= c
= c
= c
= t
= t
= t

BRB #2 R

= y0

-0.4

BRB #1 R

-0.2

0.2

BRB #2 L

BRB #3 R

BRB #2 R
BRB #1 R

BRB #1 L

-0.6

-0.4

-0.2

0.2

0.4

0.6

BRB #4 R

= c
= c
= c
= t
= t
= t

BRB #3 L

0.4

0.6

-0.6

-0.4

-0.2

swn(uroof ,)

0.2

0.4

0.6

swn(uroof ,)

Fig. 8. Max and min weighted normalized sensitivities of the roof horizontal displacement to the 80 sensitivity parameter (Chi-Chi accelerogram).

differences between the results, some general comments can be made.

When the weighted normalized sensitivities of uroof are considered
(Figs. 7, 8, and 9), the material properties related to stiffness and
strength (E0, y0, ymax,t, ymax,c) typically give the largest response sensitivities, with distributions over the oors that are similar between the
different accelerograms. Among the constitutive parameters specic of
the adopted model (t, c, t, c, r,t, r,c) the most important are the coefcients t and c controlling the hysteretic shape in tension and compression. On the other hand, r,t and r,c have generally minor effects and
t and c have always a very small inuence. These results follow from
the fact that E0, y0, ymax,t, and ymax,c have a major inuence on the
hysteretic response of BRBs, and hence on the seismic response of the
considered structure, given that E0 inuences the lateral stiffness of

the structure, y0 the entity of the lateral forces at which BRBs leave
the elastic range and start dissipating energy, ymax,t and ymax,c control
the maximum forces in tension and compression, respectively, and considerably affect the quantity of dissipated energy. Parameters t and c
have a noticeable effect on the structural response as they inuence the
quantity of energy dissipated in the BRBs by changing the area within
each deformation cycle, even if to a lesser extent as compared to the
strength parameters. In fact t and c can only give a slower of faster
transition from the elastic to the plastic range but do not change the
upper and lower stress boundaries of the plastic range. The considerations made for t and c can be repeated for r,t and r,c, although the
rate of the hardening ratio, controlled by the latter parameters, have a
smaller impact on the quantity of dissipated energy. Finally, t and c

BRB #4 R

BRB #4 L

= ymax,c

BRB #3 L

= ymax,t
BRB #2 L

= y0
= E0
-0.6

-0.4

BRB #3 R

= ymax,c

= ymax,t

= E0

BRB #1 L

-0.2

0.2

0.4

0.6

-0.6

BRB #4 L

= c
= c
= c
= t
= t
= t

BRB #2 R

= y0

-0.4

BRB #1 R

-0.2

0.2

BRB #2 L

BRB #3 R

BRB #2 R
BRB #1 R

BRB #1 L

-0.6

-0.4

-0.2

swn(uroof ,)

0.2

0.4

0.6

BRB #4 R

= c
= c
= c
= t
= t
= t

BRB #3 L

0.4

0.6

-0.6

-0.4

-0.2

0.2

0.4

0.6

swn(uroof ,)

Fig. 9. Max and min weighted normalized sensitivities of the roof horizontal displacement to the 80 sensitivity parameter (averaged values over the accelerogram set).

108

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

= ymax,c
= y0

-0.8

BRB #4 R

= ymax,t
= y0

BRB #3 L

= E0

-1.2

= ymax,c

BRB #4 L

= ymax,t

BRB #3 R

= E0

-0.4

0.4

BRB #2 L

BRB #2 R

BRB #1 L

BRB #1 R

0.8

1.2

-1.2

-0.4

0.4

= c
= c
= c
= t
= t
= t

BRB #4 L

= c
= c
= c
= t
= t
= t

-0.8

BRB #3 L

BRB #2 L

-0.8

-0.4

0.4

0.8

1.2

BRB #4 R
BRB #3 R

BRB #2 R

BRB #1 L

-1.2

0.8

BRB #1 R

1.2

-1.2

-0.8

-0.4

swn(pl,BRB1 ,)

0.4

0.8

1.2

swn(pl,BRB1 ,)

Fig. 10. Max and min weighted normalized sensitivities of the plastic strain in the ground oor left BRB to the 80 sensitivity parameter (Loma Prieta accelerogram).

control a small fraction of the elastic response when BRBs are in the
plastic range; thus, the inuence of these parameters on the structural
seismic behaviour is negligible.
When the weighted normalized sensitivities of pl,BRB1 are observed
(Figs. 10, 11, and 12) similar considerations can be repeated, i.e., largest
response sensitivities to E0, y0, ymax,t, ymax,c, t and c, generally
minor inuence of r,t and r,c, very small sensitivities to t and c.
4.3. Comparison between DDM and FDM response sensitivity results
In order to validate the derivation and implementation in
OpenSees of the DDM sensitivities for the elastoplastic BRB model
considered in this study, a large number of comparisons were made

= ymax,c
= y0

BRB #4 R

= ymax,t
= y0

BRB #3 L

= E0

-0.8

= ymax,c

BRB #4 L

= ymax,t

-1.2

between the DDM results and the results obtained by the FDM
using increasingly small perturbations of the sensitivity parameters.
Due to space limitations, the comparison between DDM and FDM results is shown herein for only two cases (Figs. 13 and 14) under the
Loma Prieta seismic input. The rst case considered is the sensitivity
of the horizontal displacement uoor1 of the rst oor with respect to
ymax,t,BRB4 of the BRBs at the fourth oor (Fig. 13). The second case is
the sensitivity of the plastic strain pl,BRB3 in the left BRB of the third
oor with respect to t,BRB2 of the BRBs at the second oor (Fig. 14).
The relative perturbations of the sensitivity parameters used in the
FDM are / = 10 1 , 10 3, 10 5 , 10 7, and 10 9. FDM gives a
crude approximation of the DDM results when / = 10 1 and
converges asymptotically to the DDM results with the succession

BRB #3 R

= E0

-0.4

0.4

BRB #2 L

BRB #2 R

BRB #1 L

BRB #1 R

0.8

1.2

-1.2

-0.4

0.4

= c
= c
= c
= t
= t
= t

BRB #4 L

= c
= c
= c
= t
= t
= t

-0.8

BRB #3 L

BRB #2 L

-0.8

-0.4

0.4

swn(pl,BRB1 ,)

0.8

1.2

BRB #4 R
BRB #3 R

BRB #2 R

BRB #1 L

-1.2

0.8

BRB #1 R

1.2

-1.2

-0.8

-0.4

0.4

0.8

1.2

swn(pl,BRB1 ,)

Fig. 11. Max and min weighted normalized sensitivities of the plastic strain in the ground oor left BRB to the 80 sensitivity parameter (Chi-Chi accelerogram).

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

= ymax,c
= y0

-0.8

-0.4

0.4

= y0

BRB #3 R

= E0

BRB #2 L

BRB #2 R

BRB #1 L

BRB #1 R

0.8

= c
= c
= c
= t
= t
= t

BRB #4 R

= ymax,t

BRB #3 L

= E0

-1.2

= ymax,c

BRB #4 L

= ymax,t

109

1.2

-1.2

-0.8

-0.4

0.4

0.8

= c
= c
= c
= t
= t
= t

BRB #4 L
BRB #3 L

BRB #2 L

BRB #4 R
BRB #3 R

BRB #2 R
BRB #1 R

BRB #1 L

-1.2

-0.8

-0.4

0.4

0.8

1.2

1.2

-1.2

-0.8

-0.4

swn(pl,BRB1 ,)

0.4

0.8

1.2

swn(pl,BRB1 ,)

Fig. 12. Max and min weighted normalized sensitivities of the plastic strain in the ground oor left BRB to the 80 sensitivity parameter (averaged values over the accelerogram set).

/ = 10 3, 10 5, and 10 7. The FDM curves with / = 10 5

and 10 7 are superimposed to the DDM curve, even in the closeups in Figs. 13d and 14d. On the other hand, / = 10 9 is a too
small perturbation, resulting in a discrepancy of the response sensitivities between the FDM and the DDM. Such discrepancy is caused
by numerical instabilities in the FDM due to round-off errors and
truncation error, as commonly observed in nonlinear nite element
FDM-based response sensitivity analysis [51,52].

(b)
DDM
FDM 10e-1
FDM 10e-3
FDM 10e-5
FDM 10e-7
FDM 10e-9

0.12
0.08
0.04
0.00

see close-up in (b)

0

swn(uufloor1, ymax,t,BRB4)

swn(uufloor1, ymax,t,BRB4)

0.16

-0.04

(c)

The same simple steel structure with BRBs in Fig. 2a is also taken
here as benchmark problem involving a simulated FE model updating
problem to illustrate another potential application of the presented sensitivity derivations. Given that E0, y0, and ymax,t can be obtained from
tensile monotonic tests on the core material of the BRB, attention is focused here to evaluate if it is possible to identify the other constitutive

10

20

30

time (s)
see close-up in (d)
0.03

0.02

21.1

21.2

21.3

time (s)

21.4

21.5

0.05
0.04
0.03
0.02
0.01
see close-up in (c)

0.00
18

(d)

0.04

0.01
21.0

0.06

-0.01
40

swn(uufloor1, ymax,t,BRB4)

swn(uufloor1, ymax,t,BRB4)

(a)

4.4. Sensitivity-based FE model updating

21.6

19

20

21

22

23

24

time (s)
0.0334
0.0333
0.0332
0.0331
0.0330
0.0329
0.0328
21.12

21.13

21.14

21.15

time (s)

Fig. 13. Sensitivity of the horizontal displacement of the rst oor with respect to ymax,t of the BRBs at the fourth oor (Loma Prieta accelerogram): comparisons between DDM and FDM.

110

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

(b)

0.02
DDM
FDM 10e-1
FDM 10e-3
FDM 10e-5
FDM 10e-7
FDM 10e-9

0.01
0.00

swn(pl,BRB3 ,t,BRB2 )

swn(pl,BRB3 ,t,BRB2 )

(a)

-0.01
see close-up in (b)
0

10

20

30

see close-up in (c)

0.005
0.000

time (s)

(d)

0.015

0.010

40

swn(pl,BRB3 ,t,BRB2 )

swn(pl,BRB3 ,t,BRB2 )

0.010

-0.005

-0.02

(c)

0.015

see close-up in (d)

0.005

0.000

10

11

12

time (s)
0.0139

0.0138

0.0137

0.0136
9.2

9.3

9.4

9.5

9.6

9.47

9.48

time (s)

9.49

time (s)

Fig. 14. Sensitivity of the plastic strain in the left BRB of the third oor with respect to t of the BRBs at the second oor (Loma Prieta accelerogram): comparisons between DDM and FDM.

parameters based on the recorded structural responses (e.g., during

shaking table tests) using a sensitivity-based model updating procedure
through the combined use of OpenSees and SNOPT [53,54]. Since
response sensitivities are very small for t and c, and these two parameters show very small variability [27], the model updating design variables are limited to parameters ymax,c, t, c, r,t, and r,c. The steps of
the adopted model updating procedure are: 1) upper and lower bounds
are assumed for ymax,c, t, c, r,t, and r,c (Table 3) based on the results
in [27]; 2) the initial values of the unknown parameters at the rst step
of the model updating procedure are taken as their average values between the upper and lower bounds for t, c, r,t, and r,c (Table 3)
while ymax,c = ymax,t; 3) SNOPT is employed to minimize the differences between the target time histories of the roof horizontal displacement (global response quantity signicantly sensitive to the parameters
of the BRBs of all oors) of the original FE model and the time histories
obtained by the current FE model with the unknown parameters ymax,c,
t, c, r,t, and r,c (having the same values for the 8 BRBs in the considered example). The objective function F to be minimized in the model
updating algorithm and its gradient are dened as:
steps


X
1
S
R 2
F
ui ui

2
R
2 umax i1

36

Table 3
Case study: boundary and rst trial values of the BRB material parameters in the model
updating problem.

Upper value
Lower value
First trial

ymax,c

r,t

r,c

(MPa)

()

()

()

()

535.43
453.75
453.75

0.90
0.60
0.75

0.90
0.40
0.65

0.80
0.10
0.45

0.80
0.10
0.45

steps


X
F
1
S
R ut i
ui ui

uRmax i1

37

where ui is the roof displacement at the i-th time step, apex R refers to
the target time history, apex S refers to the current model updating iteration, nsteps are the time steps, and uRmax = max{uRi } (i = 1, , nsteps). The
gradient uSi / is computed using the DDM algorithm previously illustrated. For comparison purposes, the FDM-base model updating is also
adopted, which in this case computes the gradient from the rst-order
nite difference approximation. The described model updating steps
are repeated for three accelerograms (1989 Loma Prieta and 1999 ChiChi already mentioned as well as 1994 Northridge, LA W 15th St station,
NGA# 1008 from the same set of accelerograms [35]) to evaluate if and
how convergence is inuenced by three different seismic inputs.
4.5. Sensitivity-based model updating results
Fig. 15a shows the convergence rate comparison in terms of the objective function values at each major iteration of the model updating
procedure. It is observed that the achieved convergence rate is inuenced by the seismic input and that DDM-based sensitivities allow a
faster convergence rate than FDM-based ones. Table 4 reports the number of iterations required to achieve a good identication of the assumed unknown BRB parameters, the resulting values of the objective
function F and the number of evaluations of F as required in SNOPT.
Here the convergence tolerance of F is set to 1.0E 8, reached in all
cases except for the DDM-based model updating under the Chi-Chi
earthquake where a local minimum is obtained. The number of evaluations of F in Table 4 gives a measure of the required computational effort, although it must be remarked that one response plus DDM-based
response sensitivity analysis is more expensive than one simple response analysis as used in FDM-based response sensitivity analysis
[31,37,38,40]. Thus, the number of evaluations of F in DDM-based
model updating cannot be directly compared to the number in FDMbased model updating. Still DDM-based model updating in the

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

(b)

1.0E_01
1.0E_02

DDM LP
DDM N
DDM CC
FDM LP
FDM N
FDM CC

1.0E_03
1.0E_04
1.0E_05
1.0E_06
1.0E_07
1.0E_08

0.8
see close-up in (c)

uroof/ htot 100

objective function F

(a)

1.0E_09

0.4
0.0
-0.4
Target

-0.8
0

20

40

111

60

1st trial

10

20

model updating iteration steps

(d)

0.6

uroof/ htot 100

0.4

30

40

time (s)

see close-up in (d)

uroof/ htot 100

(c)

Updated

0.2
0.0
-0.2
-0.4

0.0

-0.2

-0.4

-0.6
Target

-0.8
14

1st trial
16

Updated
18

20

-0.6
16.5

Target
16.6

1st trial
16.7

time (s)

16.8

Updated
16.9

17

time (s)

Fig. 15. Model updating: (a) convergence rate for the three considered accelerograms; (b,c,d) comparisons between roof displacements of the original target model, rst trial model, and
updated model for the Loma Prieta accelerogram.

considered case study allowed a reduction of up to 60% of the required

computation time, as compared to FDM-based model updating. Table 5
compares the values of the BRB material parameters as obtained from
model updating using DDM-based and FDM-based sensitivity computations. It is observed that the distribution of the differences between
target and identied values is different between the DDM-based and
FDM-based approaches even when a similar value of the objective function is attained, as in the case of the Loma Prieta seismic input. Differences are also observed within the same sensitivity computation
approach when the three different seismic inputs are compared.
Fig. 15b,c,d compares the target roof displacement, its rst trial, and the
one obtained from the DDM-based model updating procedure after 20 iterations for the Loma Prieta accelerogram. It is observed that the updated
model gives a very good approximation of the target displacements with
signicantly improved accuracy as compared to the rst trial.
Although this example is an ideal case of model updating, in real
world model updating problems the recorded horizontal displacements
are generally effected by signal noises and measurement errors that
might degrade the quality of the target function, making convergence
more difcult, it is interesting to observe and compare how the DDMbased and FDM-based model updating procedures behave under such
controlled conditions. With regard to this, and in the considered case
study, it is observed that the DDM algorithm generally improves the
convergence rate of SNOPT (as shown in Fig. 15a) and reduces the required computation time (up to 60% less in the examples considered
in this case study). Based on above remarks, it is clear that DDM based

response sensitivity computation presented in this work and implemented into OpenSees might be used in real world sensitivity-based optimization problems, both to improve convergence and reduce required
computational times.
5. Conclusions
In this paper an elastoplastic model specically developed for the
description of the behaviour of BRBs was reviewed, and its consistent
response sensitivities with respect to its ten constitutive parameters
(elastic modulus, initial yield stress, ultimate stresses in tension and
compression, hardening moduli in tension and compression, hardening
rates in tension and compression, elastic-to-plastic transition shapes in
tension and compression) were derived based on the direct differentiation method (DDM). Both the BRB model and the derived response sensitivities were implemented into the open source FE software OpenSees,
taking advantage of the OpenSees' capabilities for DDM-based response
sensitivity analysis. Results for a steel frame with BRBs were presented
to show examples of the use of the adopted sensitivity-based approach
to evaluate the inuence of the BRB constitutive parameters on the
global and local structural response, making possible an efcient analysis of the effect of a large number of parameters (80 in the considered
case study), without requiring extensive parametric analyses. In this
way it was shown that the most important effects on global and local response quantities, regardless of the damage level induced by the considered accelerogram, were given by the elastic modulus, the initial yield

Table 4
Case study: convergence in model updating.

DDM-based model updating

FDM-based model updating

Seismic input

Model updating iterations

Objective function F

Number of evaluations of F

LP
N
CC
LP
N
CC

20
20
28
42
26
56

6.64E09
7.15E09
6.30E07
6.41E09
2.43E09
9.67E09

45
29
37
346
215
535

112

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

Table 5
Case study: values of the BRB material parameters indentied using model updating.
Seismic input

DDM based model updating

FDM based model updating

LP
N
CC
LP
N
CC

Identied values

Error in the identied values

ymax,c

r,t

r,c

ymax,c

(MPa)

()

()

()

()

506.4852
508.9568
513.5679
508.5906
510.9159
509.7923

0.8010
0.7995
0.7921
0.7984
0.7994
0.7996

0.5994
0.6003
0.6044
0.6014
0.6004
0.6002

0.1003
0.1000
0.1000
0.1004
0.1000
0.1000

0.1481
0.1506
0.1538
0.1499
0.1522
0.1511

stress, the ultimate stresses in tension and compression, and the parameters dening the elastic-to-plastic transition shapes in tension and
compression. On the other hand, the hardening moduli in tension and
compression as well as the hardening rates in tension and compression
typically have a very small inuence. In addition, in order to provide a
practical example of application of response sensitivities within
gradient-based methods, results were presented for a simulated nite
element model updating problem, showing under controlled situations
that the use of DDM-based response sensitivities of the considered BRB
model generally provides more efcient model updating results than
those obtained from the nite difference method (FDM). The work presented opens the way to many applications and potentialities for the
growing community of users of OpenSees, such as sensitivity analysis
of complex BRB design solutions, both in new structures or in the rehabilitation of existing buildings, performance-based selection of optimal
BRB properties, and the development and use of optimization-based design procedures. In addition, enhancements of the adopted formulation
might be developed, for example involving the inclusion of damage
models to assess the BRB residual capacity after a seismic event.
Acknowledgments
The rst author gratefully acknowledges invaluable advices from Dr.
Frank McKenna, University of California at Berkeley, and the supports
for this research from (1) NSFC with grant Nos. 51261120376 and
51281220267 (2) Open Research Fund Program of State key Laboratory
of Hydroscience and Engineering in China with No. sklhse-2013-C-02.
Any opinions, ndings, conclusions or recommendations expressed in
this publication are those of the writers and do not necessarily reect
the views of the sponsoring agencies.
References
[1] Uang CM, Nakashima M. Steel buckling-restrained braced frames. In: Bertero VV,
Bozorgnia Y, editors. Boca Raton: CRC Press; 2004 [Chapter 16].
[2] Uang CM, Nakashima M, Tsai KC. Research and application of buckling restrained
braced frames. Int J Steel Struct 2004;4(4):30113.
[3] Black CJ, Makris N, Aiken ID. Component testing, seismic evaluation and characterization of buckling-restrained braces. J Struct Eng 2004;130(6):88094. http://
dx.doi.org/10.1061/(ASCE)0733-9445(2004)130:6(880).
[4] Kim J, Choi H. Behavior and design of structures with buckling-restrained braces.
Eng Struct 2004;26(6):693706. http://dx.doi.org/10.1016/j.engstruct.2003.09.010.
[5] Xie Q. State of the art of buckling-restrained braces in Asia. J Constr Steel Res
2005;61(6):72748. http://dx.doi.org/10.1016/j.jcsr.2004.11.005.
[6] Iwata M, Murai M. Buckling-restrained brace using steel mortar planks; performance
evaluation as a hysteretic damper. Earthq Eng Struct Dyn 2006;35(14):180726.
http://dx.doi.org/10.1002/eqe.608.
[7] Tremblay R, Bolduc P, Neville R, Devall R. Seismic testing and performance of
buckling-restrained bracing systems. Can J Civil Eng 2006;33(2):18398. http://
dx.doi.org/10.1139/l05-103.
[8] Fahnestok LA, Ricles JM, Sause R. Experimental evaluation of a large-scale bucklingrestrained braced frames. J Struct Eng 2007;133(9):120514. http://dx.doi.org/
10.1061/(ASCE)0733-9445(2007)133:9(1205).
[9] Tsai KC, Hsiao PC, Wang KJ, Weng YT, Lin ML, Lin KC, et al. Pseudo-dynamic test of a
full-scale CFT/BRB frame part I: specimen design, experiment and analysis. Earthq
Eng Struct Dyn 2008;37(7):1081198. http://dx.doi.org/10.1002/eqe.804.
[10] Tsai KC, Hsiao PC. Pseudo-dynamic test of a full-scale CFT/BRB frame part II: seismic performance of buckling-restrained braces and connections. Earthq Eng Struct
Dyn 2008;37(7):1099115. http://dx.doi.org/10.1002/eqe.803.

r,t

r,c

(%)

(%)

0.34%
0.15%
1.06%
0.08%
0.53%
0.31%

0.13%
0.06%
0.99%
0.20%
0.07%
0.05%

(%)

(%)

(%)

0.10%
0.05%
0.73%
0.24%
0.07%
0.04%

0.30%
0.00%
0.00%
0.39%
0.00%
0.00%

1.26%
0.40%
2.51%
0.09%
1.45%
0.74%

[11] Di Sarno L, Elnashai AS. Bracing systems for seismic retrotting of steel
frames. J Construct Steel Res 2009;65(2):45265. http://dx.doi.org/10.1016/
j.jcsr.2008.02.013.
[12] Chou CC, Chen PJ. Compressive behavior of central gusset plate connections for a
buckling-restrained braced frame. J Construct Steel Res 2009;65(5):113848.
http://dx.doi.org/10.1016/j.jcsr.2008.11.004.
[13] Wigle VR, Fahnestock LA. Buckling-restrained braced frame connection performance. J Construct Steel Res 2010;66(1):6574. http://dx.doi.org/10.1016/
j.jcsr.2009.07.014.
[14] Eryaar ME, Topkaya C. An experimental study on steel-encased buckling-restrained
brace hysteretic dampers. Earthq Eng Struct Dyn 2010;39(5):56181. http://
dx.doi.org/10.1002/eqe.959.
[15] Oviedo JA, Midorikawa M, Asari T. Earthquake response of ten-story story-driftcontrolled reinforced concrete frames with hysteretic dampers. Eng Struct
2010;32(6):173546. http://dx.doi.org/10.1016/j.engstruct.2010.02.025.
[16] Takeuchi T, Hajjar JF, Matsui R, Nishimoto K, Aiken ID. Local buckling restraint condition for core plates in buckling restrained braces. J Construct Steel Res
2010;66(2):13949. http://dx.doi.org/10.1016/j.jcsr.2009.09.002.
[17] Chou CC, Chen SY. Subassemblage tests and nite element analyses of sandwiched
buckling-restrained braces. Eng Struct 2010;32(8):210821. http://dx.doi.org/
10.1016/j.engstruct.2010.03.014.
[18] Mirtaheri M, Gheidi A, Zandi AP, Alanjari P, Samani HR. Experimental optimization
studies on steel core lengths in buckling restrained braces. J Construct Steel Res
2011;67(8):124453. http://dx.doi.org/10.1016/j.jcsr.2011.03.004.
[19] Wang CL, Usami T, Funayama J. Evaluating the inuence of stoppers on the low-cycle
fatigue properties of high-performance buckling-restrained braces. Eng Struct
2012;41(1):16776. http://dx.doi.org/10.1016/j.engstruct.2012.03.040.
[20] Gneyisi EM. Seismic reliability of steel moment resisting framed buildings retrotted
with buckling restrained braces. Earthq Eng Struct Dyn 2012;41(5):85374. http://
dx.doi.org/10.1002/eqe.1161.
[21] Chou CC, Liu JH, Pham DH. Steel buckling-restrained braced frames with single and
dual corner gusset connections: seismic tests and analyses. Earthq Eng Struct Dyn
2012;41(7):113756. http://dx.doi.org/10.1002/eqe.1176.
[22] Chou CC, Liou GS, Yu JC. Compressive behavior of dual-gusset-plate connections for
buckling-restrained braced frames. J Construct Steel Res 2012;76(9):5467. http://
dx.doi.org/10.1016/j.jcsr.2012.03.003.
[23] Freddi F, Tubaldi E, Ragni L, Dall'Asta A. Probabilistic performance assessment of
low-ductility reinforced concrete frames retrotted with dissipative braces.
Earthq Eng Struct Dyn 2013;42(7):9931011. http://dx.doi.org/10.1002/
eqe.2255.
[24] Fahnestock LA, Sause R, Ricles JM, Lu LW. Ductility demands on buckling-restrained
braced frames under earthquake loading. Earthq Eng Eng Vib 2003;2(2):25568.
http://dx.doi.org/10.1007/s11803-003-0009-5.
[25] Takeuchi T, Ida M, Yamada S, Suzuki K. Estimation of cumulative deformation capacity of buckling restrained braces. J Struct Eng 2008;134(5):82231.
[26] Andrews BM, Fahnestock LA, Song J. Ductility capacity models for bucklingrestrained braces. J Construct Steel Res 2009;65(89):171220.
[27] Zona A, Dall'Asta A. Elastoplastic model for steel buckling-restrained braces. J Construct Steel Res 2012;68(1):11825. http://dx.doi.org/10.1016/j.jcsr.2011.07.017.
[28] Sabelli R, Mahin S, Chang C. Seismic demands on steel braced frame building with
buckling-restrained braces. Eng Struct 2003;25(5):65566. http://dx.doi.org/
10.1016/S0141-0296(02)00175-X.
[29] Kiggins S, Uang CM. Reducing residual drift of buckling-restrained braced frames as
a dual system. Eng Struct 2006;28(11):152532. http://dx.doi.org/10.1016/
j.engstruct.2005.10.023.
[30] Ariyaratana C, Fahnestock LA. Evaluation of buckling-restrained braced frame seismic performance considering reserve strength. Eng Struct 2011;33(1):7789.
http://dx.doi.org/10.1016/j.engstruct.2010.09.020.
[31] Kleiber M, Antunez H, Hien TD, Kowalczyk P. Parameter sensitivity in nonlinear mechanics: theory and nite element computations. New York: Wiley; 1997.
[32] Haftka RT, Grdal Z. Elements of structural optimization. Dordrecht, The
Netherlands: Kluwer Academic Publishers; 1992.
[33] Ditlevsen O, Madsen HO. Structural reliability methods. New York: Wiley; 1996.
[34] Friswell MI, Mottershead JE. Finite element model updating in structural dynamics.
Dordrecht, The Netherlands: Kluwer Academic Publishers; 1995.
[35] Zona A, Ragni L, Dall'Asta A. Sensitivity-based study of the inuence of brace
over-strength distributions on the seismic response of steel frames with BRBs.
Eng Struct 2012;37(1):17992. http://dx.doi.org/10.1016/j.engstruct.2011.
12.026.

Q. Gu et al. / Journal of Constructional Steel Research 98 (2014) 100113

[36] Tsay JJ, Arora JS. Nonlinear structural design sensitivity analysis for path dependent problems. Part 1: general theory. Comput Methods Appl Mech Eng
1990;81(2):183208.
[37] Zhang Y, Der Kiureghian A. Dynamic response sensitivity of inelastic structures.
Comput Methods Appl Mech Eng 1993;108(1):2336. http://dx.doi.org/10.1016/
0045-7825(93)90151-M.
[38] Conte JP, Vijalapura PK, Meghella M. Consistent nite-element response sensitivity
analysis. J Eng Mech 2003;129(12):138093. http://dx.doi.org/10.1061/(ASCE)
0733-9399(2003) 129:12(1380).
[39] OpenSees. Open system for earthquake engineering simulation. http://opensees.
berkeley.edu/; May 24 2012.
[40] Haukaas T, Der Kiureghian A. Parameter sensitivity and importance measures in
nonlinear nite element reliability analysis. J Eng Mech 2005;131(10):101326.
http://dx.doi.org/10.1061/(ASCE)0733-9399(2005)131:10(1013).
[41] Gu Q, Conte JP, Barbato M. OpenSees command language manual response sensitivity analysis based on the direct differentiation method. http://jaguar.ucsd.edu/
OpenSees.html; May 24 2012.
[42] Simo JC, Hughes TJR. Computational inelasticity. New York: Springer-Verlag; 1998.
[43] Sloan SW, Abbo AJ, Sheng D. Rened explicit integration of elastoplastic models with
automatic error control. Eng Comput 2001;18(1/2):12154.
[44] Conte JP, Barbato M, Spacone E. Finite element response sensitivity analysis using
force-based frame models. Int J Numer Methods Eng 2004;59(13):1781820.
http://dx.doi.org/10.1002/nme.994.
[45] Barbato M, Zona A, Conte JP. Finite element response sensitivity analysis using threeeld mixed formulation: general theory and application to frame structures. Int J
Numer Methods Eng 2007;69(1):11461. http://dx.doi.org/10.1002/nme.1759.

113

[46] Chopra AK. Dynamics of structures: theory and applications to earthquake engineering. 2nd ed. Upper Saddle River, New Jersey: Prentice-Hall; 2001.
[47] Barbato M, Conte JP. Finite element structural response sensitivity and reliability
analyses using smooth versus non-smooth material constitutive models. Int J Reliab
Saf 2006;1(12):339. http://dx.doi.org/10.1504/IJRS.2006.010688.
[48] Ragni L, Zona A, Dall'Asta A. Analytical expressions for preliminary design of dissipative braces in steel frames. J Constr Steel Res 2011;67(1):10213. http://dx.doi.org/
10.1016/j.jcsr.2010.07.006.
[49] Eurocode 8: design provisions for earthquake resistance of structures, Part 1.1: general rules, seismic actions and rules for buildings, EN 1998-1; Bruxelles; 2004.
[50] PEER Ground Motion Database. http://peer.berkeley.edu/peer_ground_motion_database; March 26 2012.
[51] Gu Q, Conte JP. Convergence studies in nonlinear nite element response sensitivity
analysis. Proceedings of the 9th Int. Conf. on Applications of Statistics and Probability
in Civil Engineering, San Francisco; 2003. p. 297304.
[52] Zona A, Barbato M, Conte JP. Finite element response sensitivity analysis
of steelconcrete composite beams with deformable shear connection.
J Eng Mech 2005;131(11):112639. http://dx.doi.org/10.1061/(ASCE)07339399(2005)131:11(1126).
[53] Gill PE, Murray W, Saunders MA. SNOPT: an SQP algorithm for large-scale
constrained optimization. SIAM Rev 2005;47(1):99131. http://dx.doi.org/
10.1137/S0036144504446096.
[54] Gu Q, Barbato M, Conte JP, Gill PE, McKenna F. OpenSees-SNOPT framework for
nite-element-based optimization of structural and geotechnical systems. J Struct
Eng 2012;138(6):82234. http://dx.doi.org/10.1061/(ASCE)ST.1943-541X.0000511.