Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Received: 29 May 2016 Revised: 13 March 2017 Accepted: 21 March 2017

DOI: 10.1002/tal.1377

SPECIAL ISSUE PAPER

A practical numerical substructure method for seismic nonlinear


analysis of tall building structures
Baoyin Sun1 | Quan Gu2 | Peizhou Zhang1 | Jinping Ou1,3

1
School of Civil Engineering, Faculty of
Infrastructure Engineering, Dalian University Summary
of Technology, Dalian, Liao Ning, China It is a challenging task to perform large‐scale nonlinear analysis for complicated tall building struc-
2
School of Architecture Civil Engineering, tures due to the trade‐off between computational efficiency and numerical accuracy. Existing
Xiamen University, Xiamen, Fujian, China methods cannot satisfactorily meet the demands of engineering practices because they require
3
School of Civil Engineering, Harbin Institute predetermining local yielded regions or adaptively refining models during the analysis. This paper
of Technology, Harbin, Hei Longjiang, China
presents a modified numerical substructure method (NSM) in which the difficult task is divided
Correspondence
Jinping Ou, School of Civil Engineering, Faculty
into seismic analysis of a linear elastic “master” structure and static nonlinear analyses of a limited
of Infrastructure Engineering, Dalian quantity of substructures for local yielded regions. A “nonlinear force corrector” is calculated in
University of Technology, Dalian, Liao Ning, the substructure and applied on the master structure to consider the contribution of the yielded
China.
region's nonlinearity. A 32‐story reinforced concrete frame shear‐wall building under earthquake
Email: oujinping@dlut.edu.cn
is analyzed to verify the NSM and investigate its accuracy and efficiency. In addition, a reinforced
Funding information
concrete frame shake‐table test is simulated using the modified NSM, and local damage behaviors
National Key Research and Development pro-
gram of China, Grant/Award Number: and improved accuracy can be obtained using the refined substructure model.
2016YFC0701106; National Natural Science
Foundation of China, Grant/Award Number: KEY W ORDS
51261120376, 5157847 and 91315301‐12
numerical substructure method, large‐scale computation, local nonlinearity, nonlinear force
corrector, CS technique, seismic analysis

1 | I N T RO D U CT I O N During the past decades, the substructure method has become the
an effective and widely used method for such problems.[1–6] The
It is a challenging task to perform large‐scale nonlinear dynamic finite essential process includes subdividing the total structural system into
element (FE) analysis for realistic tall building structures due to the two or more elastic and plastic subsystems, evaluating the stiffness
trade‐off between computational efficiency and numerical accuracy. properties of each subsystem and assembling these subsystems to
A distinctive feature of such problems is the “localized” nonlinearity obtain the stiffness of the complete structure. The internal degrees
exhibited during the earthquake process, that is, only a limited of freedom (DOFs) of each elastic substructure not needed for their
number of structural components concentrated in a few local regions interconnection to the main system can be condensed out (eliminated)
exhibit strong nonlinearities, whereas a large number of the prior to the substructure assembly, greatly reducing the DOFs involved
remaining components remain completely linearly elastic.[1] The local in the solution of the assembled system. In the framework of this
nonlinear regions, although their quantities are usually small, have “conventional” substructure method, extensive research has been
crucial effects on the dynamic behaviors of the entire structure conducted, and various improvements and applications have been
and require particularly careful treatment (e.g., using a very fine FE presented. Clough and Wilson[1] proposed a substructure method to
mesh or different simulation techniques to simulate the local perform dynamic analysis of large structural systems with local
damages). It is obviously unnecessary to perform a completely nonlinearities using mode superposition and substructuring
general nonlinear FE analysis, that is, including the entire structural techniques. Bathe and Gracewski[2] presented a detailed mathematical
system in a single nonlinear equation of motion and solving it using formulation to solve large‐scale structural systems with small isolated
a conventional iterative scheme (e.g., the Newton–Raphson areas of nonlinearities. Owen and Goncalves[3] applied substructuring
algorithm or the N‐R algorithm). The computational effort can be techniques to analyze quasistatic elastoplastic problems. Ali et al[4]
greatly reduced by taking advantage of the nonlinear localization. analyzed brick masonry subjected to in‐plane concentrated loads using

Struct Design Tall Spec Build. 2017;e1377. wileyonlinelibrary.com/journal/tal Copyright © 2017 John Wiley & Sons, Ltd. 1 of 13
https://doi.org/10.1002/tal.1377
2 of 13 SUN ET AL.

multilevel substructuring techniques and a mesh grading scheme. number and locations of the plastic DOFs are not easy to predetermine
Huang and Wang[5] presented a method using substructuring before analysis, limiting the application of the force analogy method.
techniques for buckling analysis of large and complex structures. Chen This paper presents an accurate and efficient numerical substruc-
and Archer[6] presented a new domain decomposition method for ture method (NSM) for large‐scale seismic analysis of realistic high‐rise
nonlinear substructures that updates the nonlinear behavior of the building structures with local nonlinearities. In the NSM, the difficult
substructure by adding correct modes. problem of large‐scale nonlinear analysis is divided into two relatively
Although the conventional substructure methods can solve the easier subproblems, that is, a moderate‐scale linear elastic analysis of
problem by employing unrefined and refined meshes in linear and a “master” structure representing the entire building and nonlinear anal-
nonlinear regions, respectively, they require either predetermining yses of a limited quantity of “substructures” in local yielded regions.
the locations of local nonlinearity regions or refining the mesh during Each substructure is simulated using an isolated and possibly refined
the analysis. However, it is extremely difficult, if not completely nonlinear model, which can be analyzed in parallel with the master
impossible, to prejudge the local nonlinear regions before FE analysis structure on a most appropriate software platform. A “nonlinear force
due to various influencing factors, such as properties of structures, corrector” is presented which is calculated in the substructure and
foundations, soils, and earthquakes. Furthermore, once the FE mesh applied to the master structure to take into account the extra contribu-
is made, it is not easy to adaptively modify (e.g., refine) it during the tion of this substructure due to the nonlinearity. The data transfer
seismic analysis process. Therefore, it is difficult to simulate arbitrary between the master structure and the substructures is achieved by
plastic flows and damage propagation processes in the large‐scale using an efficient integration technique, that is, the CS technique.[15]
structural analysis by using the traditional substructure methods. In The novel NSM method has several distinguished features when com-
addition, although the refined and unrefined models have significantly pared with the existing methods. (a) The master structure is simulated
different accuracies (the simulation of local nonlinearity regions using an unrefined model and studied using a moderate‐scale linear
requires a more accurate model), they are solved by using a single elastic analysis. The computation is very efficient because a constant
equation of motion (with only noncondensed DOFs) with a unique structural stiffness matrix is assembled and decomposed (e.g., LU trian-
convergence criterion, which may cause a loss of accuracy in the gle decomposition) only once during the entire analysis. (b) The isolated
refined model of local yielded regions. Finally, the two models are substructures are analyzed efficiently and accurately due to various
typically analyzed on the same software platform instead of two merits of the method. First, each substructure model is created when
separate platforms because they are not physically uncoupled, which a new structural component yields (while the master structure model
discourages the use of the two most appropriate platforms for the remains unchanged), therefore, the number and locations of substruc-
two significantly different models. tures (i.e., the local yielded regions) do not need to be predetermined.
Beside substructure methods, there are other methods available in Furthermore, each substructure is isolated from the master structure
[7]
the literature. Wong and Yang proposed a force analogy method and may be modeled with a refined FE mesh or using other simulation
based on a concept of “force analogy” to study the inelastic structural techniques that can accurately simulate the local nonlinear behaviors
behaviors. In this method, a linear elastic model is used to simulate the (e.g., concrete cracking and buckling) on a most appropriate software
elastic behaviors of the structure, whereas an additional set of platform. In addition, the contribution of substructures due to the non-
independent plastic DOFs in possible yielded regions are added to linearity is taken into account by using a nonlinear force corrector that
the system for plastic behaviors. Correspondingly, the elastic force can be easily computed by a one‐step static analysis. Finally, the
(defined as the multiplication of the initial stiffness by the elastic number of local nonlinearities is usually small to moderate during the
displacement) and the analogy force (defined as the multiplication of entire seismic process for a typical tall building, therefore, the quantity
the initial stiffness by the plastic displacement) are included in an of substructures is limited, and the total computational cost is not
equation of motion that contains both the elastic and plastic DOFs. prohibitively high. (c) The models for the master structure and the
The modified equation of motion includes the elastic equations substructures are physically separated and can be analyzed on different
(i.e., between forces and displacements of elastic DOFs) and an addi- platforms, using an efficient and reliable CS technique for data transfer
tional relationship between forces and displacements of plastic DOFs. between them. Therefore, the master structure and the different
Wong and coworkers[8–12] further improved the force analogy method substructures can be analyzed using parallel and distributed
and applied it to various research areas, for example, analyzing real computation methods. (d) The method is potentially applicable to a wide
moment‐resisting frames using static condensation, investigating the range of large‐scale nonlinear problems with local nonlinearities. Two
potential energy, characterizing the energy and its transfer among application examples are employed to verify the feasibility, accuracy
various energy forms over the duration of an earthquake and proposing and efficiency of the NSM and its applicability to the large‐scale
a damage index for special moment‐resisting steel frames. Chao and nonlinear dynamic analysis of realistic tall building structures.
Loh[13] proposed a modified force analogy method to simulate the non-
linear response of reinforced concrete (RC) structures. Gang Li and
Hong‐Nan Li[14] presented an approach based on the force analogy 2 | N U M E R I C A L SU B S T R U C T U R E M E T H O D
method to analyze dynamic responses of structures with energy
dissipation devices. However, it is not easy to obtain the relation After spatial discretization using the FE method, the equation of
between forces and displacements of plastic DOFs for general motion of a nonlinear structural system (e.g., a high‐rise building) can
nonlinear structural systems under cyclic loading conditions, and the be expressed as
SUN ET AL. 3 of 13

e is
stiffness matrix). Furthermore, the nonlinear force corrector F
€ ðtÞ þ Cu_ ðtÞ þ RðuðtÞÞ ¼ PðtÞ;
Mu (1)
assembled from yielded components only, whereas the linear elastic
e
components have no contribution to F.
where M and C are mass and damping matrices, respectively; parame-
For the solution of Equation 7, the calculation for the displace-
ters u, u_ , and u€ are nodal displacement, velocity, and acceleration,
ment field of the whole structure system is performed in master
respectively; parameter R (u) denotes structural resisting force, which
structure system, whereas the nonlinear force corrector e
F is obtained
is a nonlinear function of u; parameter P represents external force
from substructure system, as depicted in Figure 1. Herein, The
applied on the structure; and t denotes time. Without loss of general-
governing equation (Equation 7) can be solved “exactly” with iterations
ity, it is assumed that the time continuous‐spatially discrete equation
or “approximately” without iterations which are called iterative and
(i.e., Equation 1) is integrated numerically in time using a general
noniterative NSMs, respectively. In the iterative NSM, a fixed‐point
one‐step integration algorithm, taking the well‐known Newmark‐β
iterative algorithm is used to compute the current displacement un+1,
method[16] as a special example herein, yielding
the nonlinear force corrector e
F, and the state variables (e.g., the plastic
" #
1 α displacement upnþ1), as shown in Table 1a, and the noniterative method
Mþ C unþ1 þ Rðunþ1 Þ ¼ Fnþ1 ; (2)
βðΔtÞ 2 βΔt is shown in Table 1b, where the subscripts i and n denote the ith
iteration and the nth time step, respectively; Tol denotes the convergence
where tolerance; and ‖⋅‖ denotes the L2‐norm. Given the displacement u, the
"  # resisting force R(u) can be computed using the substructure model, which
1 1 1
Fnþ1 ¼ Pnþ1 þ M un þ _
u n þ −1 €un
βðΔtÞ2 βΔt 2β
     
α α α
þC un þ −1 u_ n þ Δt −1 €un :
βΔt β 2β

The subscript n denotes the last converged time step, α and β are
constant parameters in the Newmark‐β method, and Δt represents the
integration time step size. Introducing a linear elastic resisting force
Kun + 1, where K is the initial elastic stiffness matrix of the structure,
Equation 2 can be modified as
" #
1 α
Mþ C þ K unþ1 ¼ Fnþ1 þ ½Kunþ1 −Rðunþ1 Þ: (3)
βðΔtÞ2 βΔt

The second term of right side in Equation 3 is defined as the “non-


linear force corrector”

e
Fnþ1 ¼ Kunþ1 −Rðunþ1 Þ (4)

because it calculates the difference between the linear elastic and


nonlinear resisting forces and remains zero for the linear elastic struc-
tural system. For a special type of nonlinear system where the elastic
moduli for the loading and unloading stages are the same (i.e., linear
elastic), the total displacement can be divided into elastic and plastic
displacements, that is, unþ1 ¼ uenþ1 þ upnþ1 , the resisting force R can
be expressed as Rðunþ1 Þ ¼ Kuenþ1 , and the nonlinear force corrector
enþ1 can be further simplified as
F

e
Fnþ1 ¼ Kupnþ1 : (5)

By defining an equivalent dynamic stiffness matrix,

1 α
Kdyn ¼ Mþ C þ K; (6)
βðΔtÞ2 βΔt

the governing Equation 3 for the NSM becomes

enþ1 :
Kdyn unþ1 ¼ Fnþ1 þ F (7)

It is worth mentioning that the equivalent dynamic stiffness Kdyn


remains constant if the mass and damping matrixes remain
unchanged during the analysis (e.g., Rayleigh damping using the initial FIGURE 1 Master structure and isolated substructures
4 of 13 SUN ET AL.

TABLE 1 Methods for solving the Equation 3 3.1 | Nonlinear force corrector of the unrefined
(a) Iterative NSM: substructure model
Step 1: Set i = 0; un+1 , i = un;
substructure  The DOFs of a substructure model are the same as those in the original
Step 2: unþ1;i → R unþ1;i (calculated in the substructure model);
 local nonlinearity region of the master structural model. The following
Step 3: eFnþ1;i ¼ Kunþ1;i −R unþ1;i ;
h i−1 h i flowchart (Table 2) is used to compute the nonlinear force corrector:
Step 4: unþ1;iþ1 ¼ Kdyn ⋅ Fnþ1 þ e Fnþ1;i ;
where uisolated
nþ1 is the nodal displacement in the isolated substructure,
Step 5: IF: ‖un+1 , i+1 − un + 1 , i‖ ≥ Tol
Set i=i+1; go to Step 2, which is obtained from the local yielded region of the master structure;
ELSE: this step converges; set n = n + 1; go to step 1 for the next step.
uElement
nþ1 and RElement
nþ1 denote the nodal displacement and resisting force
(b) Noniterative NSM:
substructure
of an element in the substructure, respectively; and ε and σ denote the
Step 1: un → Rðun Þ (calculated in the substructure model);
strain and stress in a material point of the element, respectively. The
en ¼ Kun −Rðun Þ;
Step 2: F
h i−1 h i enþ1 follows a similar
computation of the nonlinear force corrector F
en ;
Step 3: unþ1 ≈ Kdyn ⋅ Fnþ1 þ F
flowchart as that for computing the structural internal resisting force.
Step 4: set n = n + 1, go to Step 1 for the next time step.
First, the nodal displacement of the substructure uisolated
nþ1 is transferred
Note. NSM = numerical substructure method.
from the master structure, from which the nodal displacement of each
element uElement
nþ1 can be extracted, and then the strain in each material
point ε can be computed based on the assumed displacement field using
will be explained in detail later. In the noniterative method, the current dis-
a shape function interpolation. The stress σ in this material point can be
placement is approximated by that in the previous step, that is, un+1 ≈ un,
computed using a material constitutive model (material level), the
and the nonlinear force corrector by that in the previous step, that is,
element internal resisting force RElement can be integrated using the stress
enþ1 ≈F
F en ¼ Kun −Rðun Þ . Applying this assumption to Equation 7, the nþ1

σ of each material point (element level), and the structural internal force
current displacement can be computed as
Rn+1 can be computed by assembling RElement
nþ1 (structure level). Then, the

h i−1 h i nonlinear force corrector e


Fnþ1 can be computed using Equation 4.
unþ1 ≈ Kdyn ⋅ Fnþ1 þ e
Fn : (8)

3.2 | Nonlinear force corrector of the refined


substructure model
For refined substructures, it is important to set up the relation of
3 | N O N L I N E A R F O RC E C O R R E C T O R O F
responses (i.e., nodal forces and displacements) between the coarse
I S O LA T E D S UB S T R U C T U R E M O D E L
mesh in the master structure and the refined mesh in the substruc-

As mentioned above, a high‐rise building structure can be simulated tures. Figure 2 illustrates a local yielded region in the master structure

using a linear elastic FE model with coarse mesh, referred to as the model (Figure 2b) and an isolated refined substructure model (Figure 2c).

“master” structure model. During an earthquake, if a local As mentioned above, the nodal displacement uNL in the local yielded

nonlinearity occurs, a new separated or isolated (possibly refined) region of the master structure model is transferred to the isolated sub-

substructure model is built, and the connection between the master structure model, and then the corresponding nonlinear force corrector

structure and the substructures is set up (Figure 1). In each time step e
F (Equation 4) needs to be calculated by the substructure model and
of the iterative and noniterative NSMs (Table 1), the nodal displace- sent back to the master structure model. The nodal displacements in
ment in the yielded region of the master structure un+1 is sent to the the substructure usub can be divided into two groups: (a) the displace-
isolation substructure (refer to Step 2 of the iterative NSM and Step ment of the boundary nodes ub, including both the boundary nodes
1 of the noniterative NSM in Table 1). The substructure performs a on the physical edges of the substructure and the inside nodes at the

one‐step static analysis, calculates a nonlinear force corrector e


Fnþ1 same locations as those in the local yielded region of the master

(see Equation 4) and sends it back to the master structure (see structure (i.e., the central node in Figure 2b is treated as boundary node

Figure 1). Different types of nonlinear components can be treated as in substructure) and (b) the inner nodes ui, including all remaining nodes

substructures using the iterative and noniterative NSMs, including besides the boundary nodes. Therefore, the nodal displacement of the
 T
nonlinear frames, shear walls, buckling restrained braces (BRB), soils substructure can be expressed as usub ¼ uTb uTi . It is worth mention-
and foundations, and hybrid testing structures (Figure 1); however, ing that the boundary nodes defined herein are different from the
only the nonlinear frame and shear wall substructures are studied “conventional” boundary nodes connecting two FE meshes (or connec-
in this paper. The connection and data transfer between the master tion boundary) because they may include nodes not on the physical
structure and the substructures are achieved by using an efficient boundary of the substructure (e.g., the central node in Figure 2c).
and reliable data transfer technique, the CS technique. There are In the NSM presented herein, the boundary nodal displacement
two types of substructures that can be considered herein: unrefined ub can be assumed to be consistent with the displacement field of
and refined substructures. The computations of the nonlinear force the master structure to avoid displacement discrepancy on the
corrector for the two types of substructures are different, as connection boundary, that is, ub is interpolated from uNL using the
described later. shape function of the elements in the master structure, which is
SUN ET AL. 5 of 13

TABLE 2 Flowchart for computing the nonlinear force corrector for an unrefined model

ub ¼ NuNL ; (9) RNL ¼ NT Rsub;b : (13)

where N denotes “continuous boundary transfer matrix” that guaran- e of this yielded region is calculated
The nonlinear force corrector F
tees the displacement continuity on the boundary between the by substituting Equation 13 into Equation 4, yielding
master structure and the substructures. In each time step, the
substructure model can be analyzed using a one‐step static analysis e ¼ KuNL −RNL ¼ KuNL −NT Rsub;b :
F (14)
by constraining the boundary nodes to have the specified displace-
ment ub. From Equation 9, it is clear that where K is the initial stiffness of the yielded region in the master
structure.
δub ¼ NδuNL : (10)

On the basis of the principle of virtual work, that is, the local
4 | VERIFICATION AND APPLICATION
yielded region of the master structure and the isolated substructure
O F NSM
have the same virtual work as the following expression:
n o
δub In this section, the iterative and noniterative NSMs mentioned above
RTNL δuNL ¼ RTsub δusub ¼ RTsub;b RTsub;i ; (11)
δui are implemented, and their accuracy and computational efficiency are
studied in detail by comparing with the existing modified Newton–
where RNL and Rsub are resisting forces of the local yielded region of
Raphson method using initial stiffness (i.e., named as modified NR or
the master structure and the substructure, respectively; Rsub,b and
NR in the following sections). In the NSM, the resisting forces of the
Rsub,i represent resisting forces corresponding to the boundary node
isolated substructures (i.e., the Step 2 in the iterative NSM or the Step
and inner node of the substructure, respectively.
1 in the noniterative NSM as illustrated in Table 1) are computed using
In the NSM, the external force is applied on the master structure
the open‐source software framework OpenSees,[17–19] an open sys-
instead of the substructure. The corresponding displacements of the
tem for earthquake engineering simulation, developed under the aus-
master structure are applied to the boundary nodes of the substruc-
pice of the Pacific Earthquake Engineering Research Center (http://
ture for calculating the responses of substructures. The inner nodes
peer.berkeley.edu); the other calculations, including the formation
of the substructure do not subject to the external force. Therefore,
and LU decomposition of the whole structure stiffness matrix, the
the resisting force Rsub,i of the inner node is zero according to the
elastoplastic state judgment of element, and the iteration and conver-
equivalence between the resisting force and external force. Then
gence determination of the whole structural system, are all performed
Equation 11 can be modified as
in the master structure system using another OpenSees. The different
RTNL δuNL ¼ RTsub;b δub : (12) OpenSees platforms are integrated by using an efficient and simple CS
technique. In addition, a seismic analysis of the structure using the
Substitute Equation 10 into Equation 12, the resisting force RNL of modified NR method is also conducted on the OpenSees, to make their
the local yielded region in the master structure is obtained, as computational times comparable.

FIGURE 2 Calculation of nonlinear force corrector using refined substructure model


6 of 13 SUN ET AL.

Two different‐scale application examples are presented to verify of one‐dimensional steel and concrete bars. The details for this model
the NSM. The first example studies a 32‐story RC tall building are described in the literature.[23] Whereas for simplicity, an elastic
subjected to earthquake base excitation, and computational results model is adopted to simulate the out‐of‐plate behavior of the shell
obtained by using the iterative NSM, noniterative NSM and NR are element.
compared to study the efficiency and accuracy of the NSM. After static application of the gravity loads, the structure is sub-
Furthermore, the other example performs a nonlinear seismic analysis jected to a uniform base excitation, in which the input ground motion
of a two‐dimensional one‐story three‐bay RC frame and compares the is taken as the 40‐s El‐Centro (1994) earthquake acceleration scaled to
structural responses calculated by using the NSM with the shake table a peek ground acceleration (PGA) of 0.4 g (see Figure 4). Herein, the
test results. Both refined and unrefined models are modeled in the Newmark‐β integration scheme is employed with the parameters
isolated substructures during the process of the NSM, and the results α = 0.5 and β = 0.25, and Rayleigh damping is assumed with damping
are compared to demonstrate the effect of model refinement on the ratio of 0.025[24,25] corresponding to the first and third frequencies
numerical accuracy. (0.419 and 2.71 Hz) in the seismic action direction (i.e., X direction as
shown in the Figure 3b). A tolerance of 1.0e‐5 mm on the norm of
the incremental displacement is used as the convergence criterion of
4.1 | Example I the modified NR and iterative NSM. During the seismic analysis, when-
A 32‐story RC frame‐shear wall tall building (as shown in Figure 3) ever local nonlinearity occurs, refined or unrefined substructure
under an earthquake is presented to verify the accuracy and efficiency models were created and presented in the following section 4.1 and
of the NSM. The building is 129 m high with the first story height of 4.2, respectively, from which the nonlinear force corrector was calcu-
5.0 m and others of 4.0 m (see Figure 3a). The structure system lated and applied to the master structure model.
consists of RC shear walls, outer steel‐RC mega columns, outer steel There are two analysis cases that are illustrated in Table 3. In
beam, and steel braces (see Figure 3b). The detail dimensions Case 1 (section 4.1), the numerical accuracy and computational effi-
properties, steel and concrete material parameters, reinforcing detail- ciency of the NSM are studied, where each shear wall and frame
ing of the building are referred to the literature.[20] beam‐column component of all stories are simulated by using one
During the seismic analysis of the building under cyclic earthquake four‐node shell element and one displacement‐based fiber element in
load, displacement‐based fiber Euler‐Bernoulli beam‐column elements the master structure, respectively. Each yield element (or component)
with five Gauss points are used to simulate the nonlinear behavior of in the master structure system is isolated and modeled using a shell
the frame beams and columns, where uniaxial behaviors of concrete or displacement‐based fiber element in a substructure system. With-
and steel fibers in each fiber section of the frame element are out generality, yield regions, some structural components, buckling
[21]
simulated by using a modified Kent‐Scott‐Park model and Giuffro‐ restrained braces, soils and foundation, or hybrid testing structures
[22]
Menegotto‐Pinto model, respectively. The shear wall are modeled can be regarded as substructures (see Figure 1), and in this example
using a four‐node shell element, in which in‐plate nonlinear behavior herein a single shell or frame component is taken as a substructure.
of each Gauss point is simulated using a plane stress material At the same time, in order to verify the numerical accuracy and compu-
consisting of two‐dimensional steel and concrete stresses, and both tational efficiency, a comparative analysis using the modified NR is
the steel and concrete layers' stresses are calculated from the stresses performed, where the FE mesh is the same as the master structure.

FIGURE 3 The 32‐story RC frame‐shear wall building (unit: millimeter)


SUN ET AL. 7 of 13

calculated by using various time step sizes, that is, 0.02, 0.01, and
0.005 s, as shown in Figure 6. The discrepancy between the analytical
results using the modified NR and noniterative NSM decreases as the
time step size of the noniterative NSM becomes small. The root‐mean‐
square errors obtained by using Equation 15 are 18.0%, 11.6%, and
7.5% when the time step size is taken as 0.02, 0.01, and 0.005 s,
respectively. An acceptable discrepancy (i.e., 7.5%) can be obtained
FIGURE 4 The El‐Centro earthquake accelerations (1940)
when the time step size is taken as 0.005 s in this application example.
Furthermore, in order to quantify the computational efficiency, all
Then an application example of the noniterative NSM is studied in the these analyses are performed on a Dell workstation with two IntelR
Case 2, in which each shear wall, frame beam‐column component of all XeonR CPU 2.60 GHz 6‐Core processor. The detail computational
stories is simulated by using one four‐node shell element, one displace- information, including number of load steps, time step size, number
ment‐based fiber element in the master structure, respectively, and of iterations, number of formations and LU decompositions for the
each yield element in the master structure system is isolated and whole structural stiffness matrix, computational time, and numerical
modeled using a shell or displacement‐based fiber element in a errors, using the three methods with various time step sizes is summa-
substructure system. Whereas the yield shell elements at the bottom rized in Table 4. From the table, it is evident that the iterative NSM and
and 11th stories of the master structure are modeled using 16 shell the modified NR have the same computational information, that is, the
elements in the substructure system, considering that these shear number of load steps, the time step size, the number of iterations, for-
walls may be the key components as the largest drift ratio of the shear mations, and LU decompositions for the whole structural stiffness
wall at the bottom and 11th story (see Figure 8). matrix; however, the iterative NSM has a higher efficiency than the
modified NR, where the absolute computational time of the iterative
Case 1. Accuracy and efficiency investigation
NSM and modified NR are 7.47 and 9.64 hr. When the time step size
In this section, each yield element is isolated and modeled using is taken as 0.02 s, the noniterative NSM and the modified NR have
unrefined substructure model (i.e., 1 shell or frame element in the the same number of load steps, formations, and LU decompositions
substructure) and the numerical accuracy and computational efficiency for the whole structural stiffness matrix, but the number of iterations
of the NSM are studied. The 32‐story RC frame‐shear wall tall building using the noniterative NSM is much less than that using the modified
mentioned above is analyzed by using the iterative NSM, noniterative NR, and as a result of that, the computational efficiency using the
NSM, and NR, respectively. noniterative NSM is 13 times of that using the modified NR. The com-
The time histories of top displacements obtained by using the putational time (efficiency) when time step size reduces to 0.01 and
three methods are compared in Figure 5. Results from the Figure 5 0.005 s is twice (half) and 4 times (a quarter) of that when the time step
show that the responses obtained by using the iterative NSM and size equals to 0.02 s.
the modified NR are almost the same, while the noniterative NSM Figure 7 gives the time histories of isolation element ratio for
gives slightly different results. A root‐mean‐square error (eRMS) is used outer‐frame, core‐tube, and whole structure elements. Results from
to quantify the difference between two response, that is, the figure show that around 73% of the total elements, including
49% of the outer‐frame elements and 82% of the core‐tube elements,
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 N yield and then are isolated and analyzed in the substructure systems.
eRMS ¼ × ∑ ðX i −Y i Þ2 = maxðjXi jÞ×100%; (15)
N i¼1 That is to say, 27% of the whole structure elements remain elastic
and thus the determination of states (i.e., calculation for the strain
where N represents total number of data instances (i.e., time steps), Xi and stress resultant of integrator points and the resisting force of
and Yi are two responses to be compared, and subscript i denotes ith element) for them need not to perform, which may be the main factor
time step. The discrepancies, quantified by the eRMS, of the modified leading to the high efficiency of the iterative NSM.
NR versus the iterative NSM and the modified NR versus the From above analyses, it is concluded that (a) the iterative NSM and
noniterative NSM are 3.4% and 18.0%, respectively. the modified NR have the same analytical results and computational
The relatively large error of the noniterative NSM may be caused information, that is, the time histories of top displacement, the number
by the unbalance between the internal and external forces of the of iterations, formations, and LU compositions for the whole structural
structure system at each time step. In order to study the effect of time stiffness matrix; however, the iterative NSM has a higher efficiency
step size on the numerical accuracy using the noniterative NSM, than the modified NR; (b) the discrepancy between the noniterative
comparative studies are performed on the structural responses NSM and the modified NR decreases as the time step size becomes

TABLE 3 Two analysis cases for verification of the numerical substructure method

Shear wall component Beam‐column component


Analysis case Story levels Master structure Substructure Master structure Substructure

Case I 1–32 1 shell element 1 shell element 1 frame element 1 frame element
Case II 1st, 11th 1 shell element 16 shell elements 1 frame element 1 frame element
Others 1 shell element 1 shell element 1 frame element 1 frame element
8 of 13 SUN ET AL.

It needs to stress that the L‐shaped wall, including Q2 and Q3, is


the main lateral force resisting component in the core tube shear wall,
and it is simulated using two four‐node shell elements for two plane
shear walls in the two perpendicular directions (i.e., Q2 and Q3 in
the Figure 3b) and a displacement‐based fiber Euler‐Bernoulli beam‐
column element with five Gauss points for the connected concealed
column. In this paper, the building structure is subjected to unidirec-
tional input ground motion; therefore, the shear hysteretic behavior
of in‐plane for the shear walls along the direction of the seismic motion
FIGURE 5 Comparisons of top displacement time histories using
is studied, whereas the out‐of‐plane shear capacity for the shear walls
modified NR, iterative NSM and noniterative NSM. iter = iterative;
perpendicular to ground motion direction is not investigated. Herein
NR = Newton–Raphson; NSM = numerical substructure method
gives a local simulation of the shear wall located in the Q3 of the bot-
smaller. In the example presented above, the analytical results tom story using the refined substructure, as illustrated in Figure 9. It
between using the noniterative NSM with the time step size of shows that the hysteretic curve between the shear force and the drift
0.005 s and the modified NR with the time step size taken as 0.02 s ratio of the shear wall. In addition, two points (i.e., initial crack and
have an acceptable numerical error (i.e., the root‐mean‐square error maximum crack) are marked to represent initial crack state and maxi-
is below 10%); at the same time, the time cost using the noniterative mum crack state under the cyclic earthquake load in the Figure 9. It
NSM is only around 23% of that using the modified NR. Overall, the is worth mentioning that strains along diagonal directions (i.e., at 45°
noniterative NSM taken a small time step size is prohibitively efficient or −45° angle from the horizontal direction) in each Gauss point of
and may have an acceptable accuracy, and thus it is adopted to per- the shell element are studied herein, and the shear wall cracks when
form the analysis in the following section. the strain in any of the two directions is larger than 80 με.
Figure 10 gives the initial and maximum crack strain states, respec-
Case 2. Use of the refined substructure
tively. Results from Figures 9 and 10 show that the initial crack occurs
In this section, the same problem as illustrated above is reanalyzed when the drift ratio is 0.01% and a maximum crack width of 2,168 με is
by using refined substructure models for some yielded regions to study obtained when the drift ratio is −0.23% in the bottom story. From the
the detailed local nonlinearity behaviors. Shear walls in the seismic Figure 10a,b, it can be concluded that the bottom regions of the shear
action direction of the bottom and 11th stories are isolated and simu- wall have a lager tensile strain; that is to say, the crack concentrates on
lated by using refined substructures, in which relatively refined mesh, the bottom regions of the shear walls in the bottom story.
that is, mesh 4 × 4 and total 16 four‐node shell elements are used to In addition, Figure 11 gives the simulated result of the shear wall
simulate the nonlinear behavior of the wall components (see Figure 8 Q3 in the 11th story. It shows the maximum drift ratio in the positive
). And each shell element is modeled as the same as that in the master direction is 0.46% whereas a larger drift ratio (i.e., 0.825%) is obtained
structure, that is, using the nonlinear plate stress material[22] and the in the negative direction. The initial and maximum crack states are
elastic models to simulate the in‐plate and out‐of‐plate behaviors in marked in the Figure 11, but it is worth mentioning that the maximum
the four Gauss points, respectively. Comparison of maximum drift crack state occurs corresponds to the maximum drift ratio (i.e., 0.46%)
envelopes between using refined substructure and unrefined substruc- in the positive direction rather than the negative direction with larger
ture is illustrated in Figure 8; result from the figure shows that drift ratio (i.e., 0.825%).
responses of the whole structure using the refined substructure and Then Figure 12 depicts the initial and maximum crack states of the
unrefined substructure have a little difference; the maximum drift shear wall in the 11th story, respectively. The crack occurs when the
value using the refined substructure is slightly larger than that using drift ratio reaches 0.033% in the positive direction, and the top right
the unrefined substructure. The refined model can provide more infor- region of the shear wall cracks first; the crack width of the shear wall
mation about the crack development (as shown later in Figures 9–12) nearly reaches 2,000 με when the drift ratio is 0.46% in the positive
and give more detailed results. direction in the 11th story.

FIGURE 6 Comparisons of structural


responses using the modified NR method and
the noniterative NSM with various time step
sizes (the time step sizes are 0.02, 0.01, and
0.005 s). iter = iterative; NR = Newton–
Raphson; NSM = numerical substructure
method
SUN ET AL. 9 of 13

TABLE 4 Comparison of computational accuracy and efficiency using various methods


No. of Numerical errors/%
formations
Load Time No. of and LU Comput.
Methods steps step/s iterations decompositions Time/efficiency Top Disp. Top Vel. Bottom resistance

Modified NR 2000 0.02 60419 1 9.64 h/1 ‐‐ ‐‐ ‐‐


Iter. NSM 2000 0.02 70946 1 7.47 h/1.3 3.4 2.0 0.9
Non‐iter. NSM 2000 0.02 2000 1 0.59 h/13.0 18.0 31.6 5.9
Non‐iter. NSM 4000 0.01 4000 1 1.08 h/8.9 11.6 10.3 4.3
Non‐iter. NSM 8000 0.005 8000 1 2.21 h/4.4 7.5 6.8 3.1

Note. iter = iterative; NSM = numerical substructure method.

The responses of the whole structure using the refined substruc- columns are modeled using an unrefined nonlinear frame element
ture and unrefined substructure are slightly different. The stiffness (Figure 13b) satisfying a specified sectional moment‐curvature rela-
degradation, the process of cracking, widths of crack, and the region tionship with the following material parameters: bending stiffness
of crack for the shear walls can be well captured in the refined sub- EI = 160.9574 kN.mm2; yield moment My = 0.4088 kN.mm; and ratio
structure model. The refined model can provide more accurate simula- of the post yield to the initial stiffness b = 0.05. Alternately, the second
tion in some cases as shown in the section 4.2. analysis employs a refined substructure model. The two ductile col-
umns (i.e., C3 and C4) are modeled using a fiber beam column model,
whereas the other two non‐ductile columns (i.e., C1 and C2) are
4.2 | Example application II modeled using a refined truss model[27] (Figure 13e). In the refined
The second example is used to illustrate the effect of model refine- fiber beam column model for C3 and C4, the constitutive behavior of
ment on the simulation accuracy of the isolated substructure and dem- the reinforcement steel is represented using a uniaxial Giuffre‐
onstrate the improved accuracy using the refined over the unrefined Menegotto‐Pinto constitutive model with linear kinematic harden-
model. An example of a RC frame[26] subjected to a unidirectional ing[22] with the following material parameters: Young's modulus
shake table excitation is studied. The frame consists of two ductile col- E = 2.06e5 MPa; yield stresses fy = 232 MPa and 471 MPa for No. 2
umns (i.e., C3 and C4), two non‐ductile columns (i.e., C1 and C2), rigid and No. 3 steels, respectively (see Figure 13a); and ratio of the post
foots, and rigid top beams. Figure 13 shows the geometry, cross‐ yield to the initial stiffness b = 0.01. The concrete material is simulated
sectional properties, and FE model of the RC frame. The EW compo- using a uniaxial modified Kent‐Scott‐Park model.[21] Different material
nent of the TCU082 accelerogram from the 1999 Chi‐Chi Taiwan parameters are used for the confined “core” and unconfined “cover”
Earthquake scaled by 7.0 is employed as the shake table input motion concrete in the columns. For the confined concrete model, the peak
(peek ground acceleration = 1.55 g). The details of the experiments strength fc = −45.08 MPa; the strain at peak strength εc = −0.006;
and test results can be found in the literature.[26]
the residual strength fcu = −6.44 MPa; and the strain at which the resid-
Analyses are performed in OpenSees for both the master structure ual strength is reached εcu = −0.008. For the unconfined cover con-
(Figure 13a) and the isolated column substructures modeled at differ- crete model, fc = −32.2 MPa; εc = −0.002; fcu = −6.44 MPa; and
ent levels of refinement (Figure 13b,c). The first analysis employs an εcu = −0.008. For both the core and cover concrete, the tensile strength
unrefined substructure model. The top beams in the master structure ft = 1.885 MPa; the ratio between the unloading slope at crushing
are modeled using a rigid beam column element, whereas the four strength and the initial slope b = 0.1; and the tension softening stiff-
ness Et = −942.5 MPa.

FIGURE 7 Time histories of isolation element ratio for outer frame, FIGURE 8 Comparison of maximum drift responses using refined
inner core tube, and total element substructure and unrefined substructure
10 of 13 SUN ET AL.

FIGURE 9 Hysteretic curve between shear force and drift of the


FIGURE 12 Initial and maximum crack states of the shear wall Q3 in
bottom story shear wall simulated using the refined substructure
the 11th story (a) initial crack state and (b) maximum crack state

(Figure 13d). The concrete and steel constitutive models in the truss
pair are the same as those in the literature.[27] The cross section areas
of the truss pairs are set to 10714 and 17.230 mm2 for the horizontal
concrete and steel trusses, respectively; 7500.0 and 212.65 mm2 for
the vertical boundary concrete and steel trusses, respectively; and
7500.0 and 141.76 mm2 for the vertical center concrete and steel
trusses, respectively. In contrast to the horizontal and vertical truss
components, the diagonal truss component only includes one concrete
truss element with an area of 6144.2 mm2. However, it should be
emphasized that the compressive strength of the truss along one
diagonal direction is reduced by using a softening coefficient β, which
is a function of the tensile strain normal to this compressive direction,
FIGURE 10 Initial crack and maximum crack states of the shear wall as explained in detail in the literature.[27]
Q3 at the bottom story (a) initial crack state and (b) maximum crack
The global or macro responses using the two models are studied
state
by comparing them with the experimental results. The drift ratio
histories obtained by FE computations and experiments are compared
in Figure 14. It is clear that the analytical results using refined fiber and
truss models have much better agreement with the experimental
results than those using the unrefined model. The peak drift ratios
can be simulated more precisely when using the refined model than
the unrefined model.
The shear force‐displacement responses of C1 computed using
unrefined and refined models are compared to the experimental
results in Figure 15a,b, respectively. It is observed that the refined
model can precisely simulate the complicated nonlinear hysteric
behavior of the column during the base excitation process (e.g., the
pinch effect, reductions of stiffness, and strength), whereas the unre-
fined model cannot simulate these phenomena accurately. Further-
more, the refined truss model can provide extra information
FIGURE 11 Hysteretic curve between shear force and drift of the regarding the local damages, such as the generation and propagation
11th story shear wall simulated using the refined substructure of local cracks in the column (see Figure 15c) simulated by using the
tensile strain in the concrete trusses of the refined truss model. The
trusses with tensile strains larger than 0.1 at a few representative time
In the refined truss model for C1 and C2, the bending and shearing steps indicate that cracks occur at a few local regions at these time
behaviors of the column are modeled by a set of horizontal, vertical, steps.
and diagonal truss components. For each of the horizontal and vertical From Example 2, it can be concluded that a refined isolated
truss components, a truss pair, including a concrete truss and a steel substructure model based on the refined fiber and truss models is able
truss, is used to model the bending behavior of the column to simulate the nonlinear bending and shear behaviors of the column
SUN ET AL. 11 of 13

FIGURE 13 Geometry, cross‐sectional properties and finite element model of the RC frame. (a) Geometry and section properties; (b) unrefined
moment‐curvature model; (c) refined fiber model; and (d) refined truss model

FIGURE 14 Comparisons of drift ratio


histories obtained from experiments and finite
element analyses

FIGURE 15 Comparisons of the shear force versus the displacement responses of C1 between experimental results and computed results using (a)
an unrefined model and (b) a refined model. (c) Simulations of the local cracks in the C1 column at three representative times by using the refined
truss model

more precisely than the unrefined model. Furthermore, the refined appropriate software platform. A nonlinear force corrector is calcu-
model (e.g., the refined truss model) is able to simulate the generation lated by the substructure and applied to the master structure to take
and propagation of local cracks and therefore can be used to model the into account the extra contribution of this substructure due to the
local damage behaviors in the framework of the NSM. nonlinearity. Both iterative and noniterative NSMs are developed and
compared with the existing modified NR method. A 32‐story high‐rise
frame‐shear wall building is employed to verify the accuracy and
5 | CONCLUSIONS AND SUMMARY efficiency of the newly developed iterative and noniterative NSMs.
Besides, a RC frame subjected to a unidirectional shake table excitation
This paper presents an efficient and accurate method, referred to as is investigated by comparing with experimental results, the advantage
the NSM, for large‐scale nonlinear static or dynamic analysis of realistic of using refined versus unrefined substructure models are demon-
high‐rise buildings. The difficult problem of large‐scale nonlinear anal- strated by comparing with experimental results, the advantage of using
ysis is divided into two simpler uncoupled subproblems, that is, the refined versus unrefined substructure models are demonstrated.
moderate‐scale linear elastic analysis of a master structure Several conclusions obtained from these studies can be summa-
representing the entire building and nonlinear analyses of a limited rized as follows:
quantity of substructures in local yielded regions. Each substructure
is simulated using an isolated refined or unrefined nonlinear model (a) The iterative and noniterative NSMs are practically useful. The
and analyzed in parallel with the master structure on a most iterative NSM gives the same results as modified NR, while
12 of 13 SUN ET AL.

noniterative NSM gives slightly different results (i.e., the root‐ judge the elastoplastic states of the elements in the master
mean‐square error between them is 18.0% when time step size structure.
is 0.02 s), the root‐mean‐square error decreases with reducing
time step (i.e., 11.6% and 7.5% when time step size taken as ACKNOWLEDGEMENT
0.01 and 0.005 s, respectively). Financial support by the National Key Research and Development pro-
(b) The iterative and noniterative NSMs are more efficient than gram of China under Grant No. 2016YFC0701106 and the National
modified NR. In the examples presented above, the time costs Natural Science Foundation of China under grant No. 51261120376,
using the iterative and noniterative NSMs are 7.47 and 0.59 hr, 51578473 and 91315301‐12 are gratefully acknowledged.
whose computational efficiencies are 1.3 and 13.0 times of that
using the modified NR (i.e., 9.64 hr), respectively. And the compu- RE FE RE NC ES
tational cost using the noniterative NSM increases as more [1] R. W. Clough, E. L. Wilson, Comput. Methods Appl. Mech. Eng. 1979, 17,
107.
number of iterations when its time step size becomes smaller
[2] K. Bathe, S. Gracewski, Comput. Struct. 1981, 13(5–6), 699.
(i.e., the time costs are 1.08 and 2.21 hr when the time step
[3] D. R. J. Owen, O. J. A. Goncalves, Comput. Struct. 1982, 15(3), 205.
sizes are 0.01 and 0.005 s, respectively) but is still much less
than that using the modified NR. In addition, the noniterative [4] S. Ali, I. D. Moore, A. W. Page, Comput. Struct. 1987, 27(3), 417.

NSM with small time step size can be adopted to perform [5] J. Huang, T. Wang, Comput. Struct. 1993, 46(5), 845.

seismic analysis of structure. [6] H. Chen, G. C. Archer, J. Comput. Civ. Eng. 2005, 19(2), 148.
[7] K. K. Wong, R. Yang, J. Eng. Mech. 1999, 125(10), 1190.
(c) Analytical and experimental results shows that a refined isolated
[8] K. K. Wong, R. Yang, J. Eng. Mech. 2002, 128(3), 308.
substructure model based on the refined fiber and truss models is
[9] K. K. Wong, R. Yang, Earthquake Eng. Struct. Dynam. 2003, 32(14),
able to simulate the nonlinear bending and shear behaviors of the
2179.
column more precisely than the unrefined model. Furthermore,
[10] K. K. Wong, Y. Wang, Struct. Design Tall Spec. Build. 2003, 12(5), 393.
the refined model (e.g., the refined truss model) is able to simu-
[11] K. K. Wong, D. F. Zhao, J. Eng. Mech. 2007, 133(10), 1061.
late the generation and propagation of local cracks and therefore
[12] X. Zhang, K. K. Wong, Y. Wang, Eng. Struct. 2007, 29(10), 2792.
can be used to model the local damage behaviors in the frame-
[13] S. H. Chao, C. H. Loh, Earthquake Eng. Struct. Dynam. 2007, 36(12),
work of the NSM.
1659.
On the basis of the study presented in this paper, some [14] G. Li, H. N. Li, Struct. Design Tall Spec. Build. 2011, 20(3), 291.
advantages of the NSM are demonstrated: (a) the master structure [15] Q. Gu, O. Ozcelik, Struct. Eng. Mech. 2011, 40(1), 85.
can be simulated using a coarse‐mesh or even elastic model, which is [16] A. K. Chopra, Dynamics of structures, Prentice Hall, New Jersey 1995.
not changed during the entire analysis. The computation effort for [17] F. Mckenna, Object‐oriented finite element programming: Frame-
the master structure is not prohibitively large. The analysis for the works for analysis, algorithms and parallel computing. University of
California, Berkeley, 1997.
linear elastic master structure model is efficient because the
[18] F. Mckenna, M. H. Scott, G. L. Fenves, J. Comput. Civ. Eng. 2009,
constant structural stiffness matrix can be assembled and
24(1), 95.
decomposed (e.g., LU triangle decomposition) only once during a
[19] G. L. Fenves, F. McKenna, M. H. Scott, Y. Takahashi, An
seismic analysis; (b) each substructure model is built when a new object‐oriented software environment for collaborative network
structural component yields; therefore, the number and locations simulation. Proceedings of 13th World Conference on earthquake
engineering, Vancouver, Canada, 2004; 1492. DVD‐ROM.
of the substructures (i.e., the local nonlinearity regions) do not need
to be predetermined, and each substructure is isolated from the [20] H. Goman, C. Hongchao, C. Rohin, Build. Struct. 2013, 46(16), 54
(in Chinese).
master structure and may be modeled with a refined FE mesh or
[21] R. Park, M. J. N. Priestley, W. D. Gill, J. Struct. Div. 1982, 108(4), 929.
other simulation techniques that can accurately simulate the local
[22] M. Menegotto, P. E. Pinto, Method of analysis for cyclically loaded
nonlinear behaviors (e.g., concrete cracking) on a most appropriate
R.C. Plane frames including changes in geometry and non‐elastic
software platform; the number of local nonlinearity regions is usually behavior of elements under combined normal force and bending.
small to moderate during an earthquake; therefore, the quantity of Proc., IABSE Symp. on Resistance and Ultimate Deformability of
Structures Acted on by Well‐Defined Repeated Loads, Final Rep.,
substructures is usually acceptably small and the total computational
Lisbon, Portugal, 1973.
cost for the substructures is not prohibitively high; (c) the models for
[23] B. Sun, Q. Gu, P. Zhang, J. Ou, Study and application of a novel multi‐
the master structure and the substructures are physically separated; cross‐line reinforced concrete shear wall model. Proceedings of the 25th
therefore, they can be analyzed on different platforms and National Conference on Structural Engineering, Neimenggu, China,
integrated by using an efficient and reliable CS technique. The 2016. (in Chinese).

master structure and the substructures can be analyzed using parallel [24] TBI Guidelines Working Group. Guidelines for performance‐based
seismic design of tall buildings. PEER Rep. No. 2010/05, Pacific
and distributed computation methods.
Earthquake Engineering Research Center, Berkeley, CA, 2010.
Limitations include the following: (a) the method is not efficient
[25] C. Cruz, E. Miranda. J. Struct. Eng., 2016; 143(1), 04016144.
when applied to nonlinear elastic structure models because the
[26] C. L. Wu, W. W. Kuo, Y. S. Yang, S. J. Hwang, K. J. Elwood, C. H. Loh,
nonlinear force correctors for all structural components are nonzero; J. P. Moehle, Earthquake Eng. Struc. Dynam. 2009, 38(2), 205.
(b) the geometric nonlinearity has not been considered in the NSM [27] M. Mohammadreza, K. Ioannis, P. Marios, C. G. Sadik, Earthquake Eng.
presented herein; and (c) additional work is needed in each step to Struct. Dynam. 2015, 44(4), 677.
SUN ET AL. 13 of 13

Mr. Baoyin Sun is a PhD candidate at Dalian University of Technol-


How to cite this article: Sun B, Gu Q, Zhang P, Ou J. A prac-
ogy, China. His current research interests include nonlinear finite
tical numerical substructure method for seismic nonlinear anal-
element analysis, simulation, and evaluation of reinforced concrete
ysis of tall building structures. Struct Design Tall Spec Build.
structures.
2017;e1377. http://dx.doi.org/10.1002/tal.1377
Dr. Quan Gu is a professor at the Xiamen University, China. His
research interests include structural analysis and dynamics, earth-
quake engineering, structural reliability, and risk analysis.

Mr. Peizhou Zhang is a PhD at Dalian University of Technology,


China. His current research interests include nonlinear analysis,
simulation, and evaluation of reinforced concrete structures.

Dr. Jinping Ou is a professor at Dalian University of Technology,


China. His research interests include structural monitoring, control
and disaster prevention, and mitigation.

You might also like