You are on page 1of 12

Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering


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

The internal substructure method for seismic wave input in 3D dynamic soil- T
structure interaction analysis
Jingbo Liua, Xin Baoa,∗, Dongyang Wanga, Hui Tanb, Shutao Lia
a
Department of Civil Engineering, Tsinghua University, Beijing, 100084, China
b
Energy Saving & Environmental Protection & Occupational Safety and Health Research Institute, China Academy of Railway Sciences, Beijing, 100081, China

A R T I C LE I N FO A B S T R A C T

Keywords: An internal substructure method for seismic wave input in 3D soil-structure systems is proposed in this paper. In
Seismic wave input this method, we convert seismic waves into equivalent input seismic loads through a dynamic analysis of the
Dynamic soil-structure interaction internal substructure intercepted from the original soil-structure model. Compared to the wave method and the
Internal substructure boundary substructure method for seismic wave input, this newly proposed method has a smaller substructure
Artificial boundary
size and requires less data management. Moreover, it avoids the participation of the artificial boundaries in the
Equivalent input seismic load
process of inputting seismic waves and thus overcomes the limitations of traditional methods on artificial
boundaries. We provide a detailed description of the steps required to implement the proposed method. Wave
propagation in homogeneous and layered half-space sites and the scattered wave problem of a hemispherical
cavity under incident plane waves are adopted as numerical examples to validate this new method.

1. Introduction and the seismic wave input.


According to the properties of the soil considered in the calculation,
The interactions between foundations and large-scale constructions, dynamic SSI analysis can be generally classified into two categories:
such as dams, bridges and nuclear power plants, and underground equivalent linear analysis and fully nonlinear analysis. The equivalent
structures, such as tunnels and subway stations, significantly impact the linear methods have been extensively used in geotechnical engineering
seismic responses of the structures themselves, resulting in the dynamic due to their simplicity and practicability [9–11]. With the rapid pro-
soil-structure interaction (SSI) problem [1–6]. Due to increasing in- gress of computers and the enhanced understanding of the mechanical
terest in performance-based seismic design, which is a technique for behavior of soils, the fully nonlinear analysis of SSI problems has re-
evaluating and designing seismic safety and risk assessments via per- ceived increasing attention over recent years. For example, Hashemi
formance-level prediction, there is a greater awareness of the im- et al. [12] studied the effects of structural nonlinearity and foundation
portance of dynamic SSI in the overall seismic capacity of the system, sliding on probabilistic response of a nuclear structure. Huang et al.
since it can significantly affect lateral displacements and inter-story [13] comprehensively considered the nonlinearity of the rock medium
drifts [7,8]. Research approaches for SSI problems mainly include and the tunnel structure and studied the seismic response of the long
theoretical analysis, model test and numerical simulation. Among these lined tunnels under different incident angles of seismic waves. Lokke
methods, numerical simulation has the natural advantage of dealing and Chopra [14] presented nonlinear seismic analysis of interacting
with complex structural forms, geological characteristics and boundary dam-water-foundation rock system. Tabatabaiefar et al. [7,15–17] and
conditions; therefore, this method has become the main practical re- Fatahi et al. [8] reported a series of shaking table tests and corre-
search scheme for the dynamic analysis of soil-structure systems. sponding numerical simulations, comprehensively compared the
Considerable efforts have been made in recent years regarding the equivalent linear method and the fully nonlinear method with the ex-
development of numerical procedures for determining the seismic re- perimental results and suggested the use of the nonlinear analysis
sponses of structures under the influence of SSI. Successful application method in practical engineering. Tomeo et al. [18] performed the
of these methods is fundamentally dependent on three key issues: the nonlinear seismic analysis of two-dimensional reinforced concrete
numerical simulation of the soils and the structures as well as their moment resisting frames by the substructures approach and the direct
interactions, the wave radiation effect of the semi-infinite foundation approach. Abate and Massimino [19,] and Massimino et al. [20] studied


Corresponding author.
E-mail address: bx15@mails.tsinghua.edu.cn (X. Bao).

https://doi.org/10.1016/j.soildyn.2019.105847
Received 15 March 2019; Received in revised form 7 August 2019; Accepted 3 September 2019
0267-7261/ © 2019 Elsevier Ltd. All rights reserved.
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

the dynamic SSI effect through experimental and numerical methods input seismic loads on the artificial boundaries can be obtained and are
and investigated the effects of soil nonlinearity on the dynamic re- then applied to the soil-structure model to simulate the input of
sponse of the soil-structure system. In addition, structure-soil-structure earthquake energy. These two methods are clear in theory and have
interaction (SSSI), also known as site-city interaction (SCI), which been proven to be accurate in application. However, the computational
means the interaction effects among the multistructure system through processes are complicated since the directions and the values of the
the soil [21], has been recognized as important for structural seismic equivalent input seismic loads on each boundary node are specific and
response and site ground motion, especially for densely populated cities need to be determined. Recently, based on the theory of the wave
where a large number of buildings are closely constructed. Many re- method, we proposed a seismic wave input method based on the sub-
lated studies can be found in the recent literature [22–25]. structure of artificial boundaries, called the boundary substructure
To study the dynamic response of a semi-infinite soil-structure method (BSM) [46]. This method converts the calculation of equivalent
system, a finite near-field domain, including the structures and the input seismic loads into a dynamic analysis of the substructure of ar-
surrounding soil, should be used to establish the numerical model, and tificial boundaries and is thus convenient for implementation. Similar
the cutoff boundaries should be manipulated to simulate the wave ra- to the wave method, the BSM is essentially a boundary source technique
diation effect of the semi-infinite foundation. At present, the most that brings the excitation sources to the boundary nodes to input the
practically applied method for the simulation of wave energy radiation seismic waves. However, a common defect in these two methods is the
toward infinity is to set artificial boundaries (i.e., nonreflecting participation of artificial boundaries in the calculation of equivalent
boundaries or absorbing boundaries) at the cutoff boundaries. input seismic loads. The applied artificial boundaries are limited to
According to the underlying mathematical or physical techniques, the stress-type artificial boundaries, whereas displacement-differential ar-
representative work on artificial boundaries can be classified into the tificial boundaries (such as the transmission boundary) and PML tech-
following general categories: (i) the infinite element method (IEM) niques, which have been widely applied in various branches of en-
[26,27], which is an infinite domain simulation method based on the gineering in recent years, are no longer applicable for these two
finite element method (FEM) that utilizes the attenuation solutions of methods. Moreover, in these two methods, the equivalent input seismic
outgoing waves as shape functions but is mainly applied to frequency loads should be calculated according to different artificial boundary
domain analysis; (ii) the boundary element method (BEM) [28,29], conditions and applied to all artificial boundary nodes. Since artificial
which has the advantage of dimensionality reduction but relies on the boundaries are usually approximately accurate for most practical cases,
basic solutions (Green's functions) satisfying the initial and boundary to reduce the calculation error caused by artificial boundaries, it is
conditions and is thus is difficult to apply to large-scale SSI systems necessary to set boundaries far from the region of interest. In addition,
with complex boundary conditions; (iii) the perfectly matched layer credible simulation of the dynamic SSI problem requires the con-
(PML) [30,31], which is an energy-consuming buffer (damping layer) sideration of a certain range of surrounding soil in the calculation.
outside the calculation domain with strong adaptability for the shape of Therefore, soil-structure models are usually large and utilize many
cutoff boundaries and is sensitive to changes in the buffer thickness and elements, especially for three-dimensional problems. Due to the tra-
the parameters of the damping functions; (iv) the transmitting versal calculation and application of equivalent input seismic loads on
boundary [32,33], which directly simulates the wave propagation on all boundary nodes, the processing techniques of current methods are
the cutoff boundaries through the recursion of nodal displacements but complicated, and the computational burdens are exceedingly heavy.
whose high-frequency stability needs to be further studied; and (v) the To further simplify the process and improve the efficiency of seismic
viscous boundary [34] and viscoelastic boundary [35–37], which con- wave input while overcoming the restrictions of traditional methods
vert the partial differential equations of unilateral wave motion into regarding the types of artificial boundaries, an internal substructure
stress boundary conditions, can be simulated by a series of basic me- method (ISM) for seismic wave input in three-dimensional dynamic
chanical components and are thus collectively referred to as stress-type soil-structure analysis is proposed in this study. In this method, seismic
artificial boundaries. Viscous boundaries are composed of dampers, and waves with arbitrary incident angles are transformed into equivalent
viscoelastic boundaries are simulated by spring-mass-damper systems. seismic loads and input only within the region of interest. In addition,
Due to their high precision, spatial decoupling and ease of assembling these equivalent input seismic loads are directly obtained through dy-
into general finite element software, stress-type artificial boundaries namic analysis of the internal substructure, which is composed only of
have been widely used and rapidly developed in recent years. two-layer elements and can be intercepted from the original soil-
According to the relative positions of the excitation source and the structure model. The size of the substructure model in the ISM is re-
region of interest, dynamic SSI problems can be divided into internal duced compared with those in the wave method and BSM, and the
and external source problems. For the former, which includes problems computational efficiency is increased. Moreover, this new method
related to structural vibrations and near-field explosions, the excitation avoids the participation of artificial boundaries in the process of in-
sources are within or very close to the region of interest; thus, the whole putting seismic waves. Therefore, the limitations on the artificial
analysis model can be established. No extra energy is input through the boundaries are completely overcome, and the universality of this new
cutoff boundaries; therefore, all the artificial boundaries mentioned method is improved. We provide the detailed implementation steps of
above can work efficiently in such an analysis. For external source the proposed method. Wave propagation in homogeneous and layered
problems, such as problems related to earthquakes and far-field ex- half-space sites and the scattered wave problem of a hemispherical
plosions, the problem becomes exceedingly large if the entire domain, cavity under incident plane waves are adopted as numerical examples
from the excitation source to the region of interest, is simulated; thus, to validate this new method.
only the local site model without excitation sources is generally es-
tablished. However, difficulties occur when inputting external excita- 2. Internal substructure method
tions into the near-field domain without affecting the absorption of
outgoing waves by artificial boundaries. Many studies have been con- 2.1. Theoretical derivation
ducted on this research topic [6,38–42], and two typical approaches are
the domain reduction method (DRM) proposed by Bielak et al. [43,44] Fig. 1 shows a typical schematic of the three-dimensional soil-
and the wave method proposed by Liu and Lu [45]. The DRM is a two- structure system. A near-field calculation domain is intercepted to es-
step method that converts the free field motions of unaltered soil cal- tablish the analysis model (see the dashed box in Fig. 1), and the arti-
culated by the large-scale global model into the equivalent seismic ficial boundaries need to be set on the cutoff boundaries to absorb the
loads of the small-scale local model. The wave method is derived based outgoing waves. The corresponding finite element model is illustrated
on viscoelastic artificial boundaries. Using this method, the equivalent in Fig. 2. For demonstration purposes, the quarter on the viewer's side is

2
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 1. Three-dimensional soil-structure system.

excised from all the displayed models in this article. the excitations closer to the region of interest. By simplifying steps (i)
Under seismic excitation, the wave propagation process in the nu- and (ii), the seismic waves can be directly input around the region of
merical model shown in Fig. 2 can be decomposed into the following interest, thereby reducing the amount of calculations.
steps: (i) the seismic wave input from the artificial boundaries; (ii) the According to the theory of the wave method, the incident seismic
propagation of the incident wave through the soil to the vicinity of the waves can be transformed into equivalent input seismic loads, which
structures; (iii) the interactions between the structures and the soil, as are determined only by the free wave field and are unaffected by the
well as the generation of scattered waves with the structures as the internal structures. In addition, these equivalent input seismic loads
sources; and (iv) the wave radiation through the soil and the wave should ensure that the motions and stresses of the input boundaries are
absorption by the artificial boundaries. For common dynamic SSI pro- consistent with those of the free wave field. The three-dimensional free
blems caused by earthquake excitation, the interactions between the field finite element model corresponding to the soil-structure system is
structures and the surrounding soil, as described in steps (iii) and (iv), shown in Fig. 3. The nodes in this model are classified into five types
are the key issues of concern. The calculation domain where these in- according to their positions: internal nodes (denoted as I), external
teractions take place, as shown by the region within the red node box in nodes (denoted as E) and three-layer internal substructure nodes (re-
Fig. 2, is designated as the region of interest. The domain between the presented by red circles , green circles and blue circles respec-
region of interest and the artificial boundaries is denoted as the external tively, and denoted as A, B and C from the outside to the inside). The
region. As mentioned above, to ensure the accuracy of the simulation, domain within the box formed by the A nodes is the region of interest
the artificial boundaries are usually set far from the region of interest. where we wish the seismic waves to be input. It is important to em-
Thus, the seismic wave input from the artificial boundaries and the phasize that the mechanical behavior of the soils and the structures as
simulation of wave propagation in the external region are computa- well as their interactions within the region of interest can be linear or
tionally expensive. A feasible way to overcome this difficulty is to move nonlinear, but the soils outside this region should be linear elastic to

Fig. 2. Finite element model of the soil-structure system.

3
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 3. Finite element model of the free field.

satisfy the superposition principle. This hypothesis is reasonable be- namely:


cause of the amplitude decay of the scattered wave radiation [47].
Although the method proposed in this study is applicable to linear and uB = uB0, uC = uC0, uI = u0I , FC = 0, FI = 0 (3)
nonlinear problems, only the linear elastic case is considered during the
where the superscript 0 denotes the free wave field.
derivation for clarity of presentation.
By substituting Eqs. (2) and (3) into Eq. (1), the equivalent input
Notably, to realize the locational correspondences of the equivalent
seismic loads can be obtained:
input seismic loads between the free field model and the soil-structure
model, the mesh generation connected to the internal substructure
⎧ FE = 0
nodes A, B, and C, and the material properties in the region of the in- ¨ B0 + KAB⋅uB0
FA = MAB⋅u
ternal substructure should be identical in these two models. ⎨
¨ B0 + MBC⋅u
⎩ FB = MBB⋅u ¨ C0 + KBB⋅uB0 + KBC⋅uC0 (4)
Based on the above node classification, the motion equation of the
free field model shown in Fig. 3 can be expressed in the following form: The above derivation provides a direct method for determining the
equivalent input seismic loads based on the free field model. However,
⎡ ΜEE MEA 0 0 0 ⎤⎧u ¨E ⎫
MAB 0 0 ⎥⎪ u
¨ until now, this scheme offers little advantage over the wave method and
⎢ MAE MAA ⎪ A⎪ ⎪ DRM because the scheme requires to conduct a dynamic analysis of an
⎢ MBA MBB MBC 0 ⎥ u ¨B
⎢0 MCC MCI ⎥ ⎨ ⎬ additional free field finite element model, leading to tremendous
⎢0 0 MCB ⎥⎪u ¨ C⎪
MIC MII ⎥ ⎪ u ⎪ computational effort, especially for three-dimensional problems.
⎣0
⎢ 0 0 ⎦⎩ ¨I ⎭
Therefore, more manipulations should be taken to further simplify this
⎡ KEE KEA 0 0 0 ⎤ uE FE ⎫
KAB 0 0 ⎥⎧ ⎫ ⎧ F
method.
⎢ KAE KAA uA ⎪
⎪ ⎪ ⎪ A⎪ ⎪ Eqs. (3) and (4) show that the equivalent input seismic loads vanish
+ ⎢0 KBA KBB KBC 0 ⎥ uB = FB
⎢ K CC K CI ⎥ ⎨ uC ⎬ ⎨ FC ⎬ everywhere except on the A and B nodes. Moreover, their values in-
⎢0 0 K CB ⎥⎪ ⎪ ⎪ ⎪
KIC KII ⎥ ⎩ uI ⎭ ⎪ FI ⎪ volve only the free-wave-field motions of the B and C nodes. Therefore,
⎣0
⎢ 0 0 ⎦ ⎩ ⎭ (1)
we establish an internal substructure model that consists only of the
where M and K are the mass and stiffness matrices of the free field three-layer nodes A, B, C and the two-layer elements lying between
model, respectively, u and ü are the vectors of nodal displacement and them, as shown in Fig. 4.
acceleration, respectively, F is the nodal force vector, and the subscripts The motion equation of the internal substructure model can be ex-
E, A, B, C and I refer to the above nodal classification. pressed as follows:
The free field model is divided into the internal region (the region of
sub sub
interest) and the external region by the A nodes, as shown in Fig. 3. For ⎡ MAA MAB 0 ⎤⎧u ¨ Asub ⎫ ⎡ KAA
sub sub
KAB 0 ⎤ ⎧ uAsub ⎫ ⎧ F Asub ⎫
⎢ Msub Msub Msub ⎥ ⎪ u ⎪
sub + ⎢ sub sub sub ⎥
⎪ sub ⎪ ⎪ sub ⎪
=
the external region, since the seismic waves are not input in this area, ⎢ BA BB BC ⎥ ¨ B K KBB KBC ⎥ ⎨ uB ⎬ ⎨ F B ⎬
⎨ ⎬ ⎢ BA
and there is no scattered wave generated in the free field model, the A ⎢ 0 sub sub ⎥ ⎪ sub ⎪
MCB MCC ⎦ ⎩ u ⎢ sub sub ⎥ ⎪ sub ⎪
¨ C ⎭ ⎣ 0 K CB K CC ⎦ ⎩ uC ⎭ ⎪ sub ⎪
⎣ ⎩ FC ⎭
and E nodes are assumed to remain stationary during the calculation,
(5)
namely:
where the superscript sub denotes the internal substructure model.
uA = 0, uE = 0 (2)
According to the concept of finite element discretization, if the mesh
On that basis, according to the concept of the isolation body, the generation and the material properties of the internal substructure
motions in the internal region can be identical to those of the free wave model are consistent with those of the corresponding positions of the
field by applying only the appropriate nodal loads FB on the B nodes, free field model, the following equations can be obtained:

4
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 4. Finite element model of the internal substructure.

MΦsub = MΦ , K Φsub = K Φ (6) which is much smaller than the original soil-structure model. Therefore,
a new seismic wave input method based on the internal substructure for
where the subscript Φ can denote “AB”, “BA”, “BB”, “BC” or “CB”. three-dimensional SSI problems is proposed in this study, and this
By substituting Eq. (6) into Eq. (5), fixing the A nodes (uAsub
= 0 ),
method is called the internal substructure method (ISM). In addition,
and applying the free field motions on the B and C nodes (uBsub = uB0 and
the derivation process of ISM does not involve the artificial boundaries,
uCsub = uC0 ), the following equations can be obtained:
and the input of seismic waves and the absorption of scattered waves
sub
¨ B0 + KAB⋅uB0 are spatially separated. The artificial boundaries do not participate in
⎧ F A = MAB⋅u
⎨ F Bsub = MBB⋅u the calculation of equivalent seismic loads, and thus the restrictions on
⎩ ¨ B0 + MBC⋅u
¨ C0 + KBB⋅uB0 + KBC⋅uC0 (7) the applied artificial boundaries are completely released. The common
A comparison of Eq. (4) and Eq. (7) shows that the nodal forces on artificial boundaries supported by the finite element programs, such as
the A and B nodes calculated by the free field model are identical to the transmitting boundary, the viscoelastic boundary and the PML, can
those of the internal substructure model. Therefore, the equivalent all be applied in this new seismic wave input method. Notably, we give
input seismic loads can be obtained through the dynamic analysis of the the “structures” here a broad definition including general super-
internal substructure model, which is much smaller than the original structures and underground structures, the nonlinear subsystem formed
free field model. by the structures and the surrounding nonlinear soils, and even the
Then, we apply the equivalent input seismic loads FA and FB on the local topographies, as shown in Fig. 6. It is obvious that the sizes of the
corresponding nodes of the soil-structure model shown in Fig. 2, substructures used in ISM are much smaller than those of the boundary
thereby inputting the incident seismic waves u0 into the region of in- substructure method (BSM). This feature of size reduction effectively
terest through the internal substructure. The incident waves propagate reduces the computational effort, especially the tremendous amounts of
forward and interact with the structures, generating the scattered waves data storage and transmission on the input boundaries.
uS. According to the concept of wave field decomposition, the wave
field inside the internal region consisting of the incident free wave field 2.2. Implementation steps
and the scattered wave field caused by the soil-structure interaction is
the total wave field, namely: The implementation steps of the internal substructure method are
provided as follows:
uT = u0 + uS (8)
T
When the outgoing waves u from the internal region reach the (1) The three-dimensional near-field soil-structure finite element model
boundary consisting of the A nodes, the free field motions u0 are offset is established, as shown in Fig. 2. The spatial range of the internal
by the nodal forces FA, while the scattered waves uS pass through and region where the seismic waves are input can be determined ac-
propagate outward until they are absorbed by the artificial boundaries. cording to the position of the structures (see the red node box in
Because of this “filtering” effect of the internal substructure on the Fig. 2). Then, by deleting the external E nodes and the internal I
outgoing free field waves, the wave field outside the internal sub- nodes as well as all the elements connected to them, the corre-
structure contains only the scattered wave field uS. The above process sponding internal substructure model can be established (see
of wave propagation in the soil-structure model is shown in Fig. 5. Fig. 6(a)), which consists of only the three-layer nodes (A, B, and C)
According to the above derivation, the seismic wave input problem and the two-layer elements between them.
of the soil-structure system can be transformed into the internal source (2) According to the incident seismic waves, the free wave field can be
problem by applying the appropriate equivalent input seismic loads obtained by the one-dimensional method or the one-dimensional
around the region of interest. These equivalent input seismic waves can time domain algorithm [48,49]. Then, by fixing the A nodes, im-
be obtained through dynamic analysis of the internal substructure, posing the free field motions on the B and C nodes, as shown in

5
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 5. Wave propagation in the soil-structure model.

Fig. 7(a), and conducting a dynamic analysis of the internal sub- The ISM is convenient in application and can be easily integrated in
structure, the reaction forces on the A and B nodes can be obtained; general finite element software. Furthermore, there are few restrictions
these forces are exactly equal to the equivalent input seismic loads on the spatial range of the internal substructure in this new method. In
FA and FB. an extreme case, the internal substructures can be adjacently inter-
(3) The equivalent input seismic loads FA and FB are applied to the cepted outside the structures, and these substructures are far smaller
corresponding nodes of the soil-structure finite element model, as than the artificial boundary substructures in the BSM. This feature ef-
shown in Fig. 7(b). Then, the seismic response of the soil-structure fectively reduces the computational effort, especially the tremendous
model can be obtained through dynamic analysis. amounts of data storage and transmission on the input boundaries. In
addition, the internal substructure is adopted only to determine the
It is important to note that although Eq. (4) involves two sets of equivalent input seismic loads, on the only requirement that the ma-
variables, i.e., displacement u and acceleration ü , only one of them is terial properties of the internal substructure be the linear elastic. The
required to be input in implementation, and the other set of variables materials inside the internal region, however, can be arbitrary, noticing
can be automatically obtained through the dynamic analysis of the that their properties do not participate in the calculation of equivalent
internal substructure model. loads FA and FB. Therefore, the nonlinear SSI effect can be considered

Fig. 6. Application prospects of the ISM in earthquake engineering. a. Nuclear power plant. b. Bridge. c. Local topography. d. Underground structure.

6
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

of the ISM without the impact of the SSI effect, in this section, the
seismic waves are input into the engineering sites without structures,
and the numerical results are compared to the corresponding theore-
tical solutions. Without loss of generality, we consider the three types of
engineering sites, i.e., the homogeneous half-space site, the layered
half-space site and the homogeneous half-space site with a hemi-
spherical cavity, and three types of incident seismic waves, i.e., the
plane P, SH and SV waves, incident at different angles.

3.1. Homogeneous half-space site

First, we consider the seismic wave input in the flat homogeneous


half-space site. The computational model is shown in Fig. 8. L, W and H
are the length, width and height of the model, respectively, θ is the
angle between the wave propagation direction and z-axis, and ϕ is the
angle between the projection of the wave propagation direction in the
horizontal plane and the x-axis. The green and red boxes in Fig. 8
correspond to the two internal substructures used in this example,
which are denoted as IS1 and IS2, respectively. In addition, the three-
dimensional viscoelastic artificial boundaries are applied on all cutoff
boundaries [50].
The material properties of the homogenous half-space site are as
follows: mass density ρ = 2000 kg/m3, velocity of shear wave
cS = 200 m/s and Poisson's ratio ν = 0.25. The dimensions of the nu-
merical model are L = W = 100 m, H = 50 m. The dimensions of the
internal substructures IS1 and IS2 are L1 = W1 = 40 m, H1 = 20 m and
L2 = W2 = 60 m, H2 = 30 m, respectively. The coordinates of observa-
tion points A and B are (50, 50, 50) and (50, 50, 34), respectively. A
Fig. 7. The implementation steps of the internal substructure method. a. Nodal Dirac-δ pulse wave with a pulse duration of 0.2 s is selected as the in-
load calculation in the internal substructure model. b. Seismic wave input in the cident plane P wave, whose displacement time history and Fourier
soil-structure model. amplitude spectrum are plotted in Fig. 9. The incident angles are
θ = 30° and ϕ = 45°.
The incident wave is input by three means: the internal substructure
inside the internal region.
method based on substructures IS1 and IS2, which are denoted as ISM-
IS1 and ISM-IS2, respectively, and the BSM. The displacements of ob-
3. Numerical examples servation points A and B calculated by the above numerical approaches
are plotted in Fig. 10. The results show that the dynamic response
As mentioned above, the process of seismic wave input through ISM calculated by the ISM is consistent with the theoretical solution [13],
is independent of the SSI effect and the wave absorption effect of the indicating that the present method can effectively input seismic waves
artificial boundaries; it is related only to the site condition, the incident into the homogeneous half-space model. In addition, the size of the
wave and the incident angle. To validate the effectiveness and accuracy internal substructure has no effect on the calculation accuracy.

Fig. 8. Homogeneous half-space model.

7
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 9. Time history and Fourier amplitude spectrum of the displacement of the incident pulse wave.

Fig. 10. The displacements of observation points on the homogeneous half-space model under an incident plane P wave. a. Observation point A. b. Observation point
B.

Table 1 Table 1 gives the numbers of elements and nodes in different sub-
Case 1: The numbers of elements and nodes in different substructures. structures when using the ISM and the BSM. Obviously, the model of
Quantity Internal substructure method Boundary substructure method
the internal substructure is much smaller than the artificial boundary
substructure. To be specific, the quantities of elements and nodes in the
IS1 IS2 ISM-IS1 and ISM-IS2 are reduced by 84% and 64%, respectively,
compared to those in the artificial boundary substructure. Since one of
Element 2408 5408 15008
Node 3747 8307 22827
the most time-consuming procedures in seismic wave input is the data
storage and transfer of nodal displacements and forces, which is pro-
portional to the number of nodes in the substructure model, this size
reduction feature of the ISM significantly reduces the number of nodes,
thus increasing the computational efficiency.

Fig. 11. Snapshots of the homogeneous half-space model under an incident P wave. a. t = 0.224 s. b. t = 0.332 s. c. t = 0.440 s.

8
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

dotted box calculated by the ISM are consistent with those of the BSM,
indicating that the proposed method can efficiently input the seismic
waves into the internal region. In addition, due to the “filtering” effect
of the ISM on the outgoing free field waves, only the scattered waves
can pass through the internal substructures and propagate outward. For
the homogeneous half-space model with no structures inside, no scat-
tered waves are generated, and thus, the wave field vanishes in the
external region.

3.2. Layered half-space site

The layered half-space model, which considers the local site effect,
is more practical than the homogeneous model. Wave propagation
theory denotes that the soil layers modify the amplitude and frequency
content of the input seismic waves while passing though the soil layers,
which can exert a crucial influence on the structural seismic response
and the ground motion field [51]. Due to the wave transmissions and
reflections at the interfaces of different media, the distributions of the
wave field are complicated, and seismic wave input in such engineering
sites is challenging. Here, we consider a two-layer system containing
the elastic half-space and the covering layer, as shown in Fig. 12. The
dimensions of the model are L = W]H=100 m, and the thickness of
Fig. 12. Layered half-space model. the covering layer is Hc = 50 m. The three-dimensional viscoelastic
artificial boundaries are applied on all cutoff boundaries. The green and
Table 2 red boxes in Fig. 12 correspond to the two internal substructures IS1
Case 2: The numbers of elements and nodes in different substructures. and IS2, respectively. The dimensions of IS1 and IS2 are L1 = W1]
H1=40 m and L2 = W2]H2=60 m, respectively.
Quantity Internal substructure method Boundary substructure method
The material parameters of the computational model are as follows:
IS1 IS2 mass density ρ1 = ρ2 = 2000 kg/m3, velocity of shear wave
cS1 = 200 m/s, cS2 = 100 m/s and Poisson's ratio ν1 = ν2 = 0.25, where
Element 4008 9008 25008
the subscripts 1 and 2 denote the half-space and the covering layer,
Node 6147 13707 37827
respectively. A pulse plane SH wave with amplitude shown in Fig. 9 is
inputted into the model, and the incident angles are θ = 45° and
Fig. 11 illustrates snapshots of the displacement wave field under ϕ = 60°. The coordinates of observation points A, B and C are (50, 50,
the obliquely incident plane P wave. The dotted boxes denote the in- 100), (50, 50, 90) and (50, 50, 50), respectively. Table 2 provides the
ternal substructures, of which the red one is adopted for input of the numbers of elements and nodes in different substructures. The model
seismic waves. The wave fronts of the incident P wave, the reflected P size of the internal substructure is still much smaller than that of the
wave and the reflected SV wave can be clearly observed in the snap- artificial boundary substructure, which guarantees the high computa-
shots. It is important to note that the wave field distributions in the red tional efficiency of the ISM.
Fig. 13 shows the displacements in the x and y directions calculated

Fig. 13. The displacements of observation points on the layered half-space model under an incident plane SH wave. a. Observation point A. b. Observation point B. c.
Observation point C.

9
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 14. The displacements of observation points on the layered half-space model under an incident plane SH wave. a. t = 0.44 s. b. t = 0.732 s. c. t = 0.94 s.

wave input for layered half-space sites can be verified.

3.3. Homogeneous half-space site with a hemispherical cavity

To illustrate the applicability of the ISM for analyzing topographic


effects, we examine, as a third example, the case of a homogeneous
half-space site with a hemispherical cavity. The geometry and material
parameters, as well as the internal substructures adopted in this nu-
merical model, are all identical to those of the first example, except that
there is a hemispherical cavity in the middle of the free surface, as
shown in Fig. 15. The cavity radius is R = 6 m. A pulse plane SV wave
with amplitude shown in Fig. 9 is selected as the incident wave, and the
incident angles are θ = 20° and ϕ = 45°.
The displacements of observation points A (50, 50, 44) and B (50,
50, 34) are plotted in Fig. 16 and agree well with the results of the BSM.
Fig. 15. The homogeneous half-space model with a hemispherical cavity. Fig. 17 shows snapshots of the displacement wave fields under the
obliquely incident plane SV wave. Within the internal region, free field
waves and scattered waves together compose the total wave field. When
by the ISM, and these displacements show very good consistency with
the outgoing total waves reach the internal substructure, the free field
the theoretical solutions. Snapshots of the displacement wave field
motions are completely absorbed, while the scattered waves pass
under the obliquely incident plane SH wave are illustrated in Fig. 14.
through and propagate outward. Namely, the ISM realizes effective
The wave field in the layered model is composed of the incident, re-
seismic wave input inside the region of interest without affecting the
flected and transmitted SH waves, whose wave fronts are clearly dis-
generation and propagation of scattered waves. Therefore, the total
played in the snapshots. In addition, the wave field distributions within
wave field within the internal region and the scattered wave field in the
the internal regions (see the red dotted boxes in Fig. 14) calculated by
entire model calculated by the present method are both consistent with
the ISM agree well with those of the BSM, as expected. Consequently,
those of the BSM.
the accuracy and effectiveness of the present method for the seismic

10
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

Fig. 16. The displacements of observation points on the homogeneous half-space model with a hemispherical cavity. a. Observation point A. b. Observation point B.

Fig. 17. Snapshots of the homogeneous half-space model with a hemispherical cavity. a. t = 0.33 s. b. t = 0.46 s. c. t = 0.65 s.

4. Summary and conclusions reduces the computational effort, especially the tremendous amounts of
data storage and transmission on the input boundaries. In addition, the
Based on the theory of the wave method and the concept of the proposed method avoids the participation of the artificial boundaries in
substructure method, an internal substructure method (ISM) for seismic the process of inputting seismic waves. Therefore, the restrictions of
wave input in three-dimensional dynamic soil-structure interaction traditional methods on the types of artificial boundaries are completely
(SSI) analysis is proposed in this study. In this method, seismic waves overcome, and this new method can be more universally adopted.
with arbitrary incident angles are converted into equivalent input
seismic loads around the region of interest in the numerical model. The Acknowledgments
equivalent input seismic loads are obtained through dynamic analysis
of the internal substructure, without complex formula calculations. In This study is supported by the National Natural Science Foundation
addition, the internal substructure model is directly intercepted from of China (no.51878384) and the National Key Research and
the original soil-structure model; thus, the locational correspondences Development Program of China (no.2018YFC1504305). Financial sup-
of the loading nodes can be established, and this new method can be port from these organizations is gratefully acknowledged.
conveniently implemented in general finite element software. The de-
tailed implementation steps of the ISM are provided. The accuracy and References
effectiveness of this new method are verified by numerical examples of
wave propagation in homogeneous and layered half-space sites, as well [1] Amorosi A, Boldini D. Numerical modelling of the transverse dynamic behaviour of
as the scattered wave problem of a hemispherical cavity. circular tunnels in clayey soils. Soil Dyn Earthq Eng 2009;29(6):1059–72https://
doi.org/10.1016/j.soildyn.2008.12.004.
The sizes of the substructures used in this new method are much [2] Chopra AK, Wang JT. Earthquake response of arch dams to spatially varying ground
smaller than those of the wave method and the boundary substructure motion. Earthq Eng Struct Dyn 2010;39(8):887–906https://doi.org/10.1002/eqe.
method (BSM). For the numerical examples conducted in this study, the 974.
[3] Vlado G, Mihailo DT, Nebojša O. Two-dimensional translation, rocking, and waves
quantities of elements and nodes in the ISM are reduced up to 84% in a building during soil-structure interaction excited by a plane earthquake SV-
compared to those of the BSM. This feature of size reduction effectively wave pulse. Soil Dyn Earthq Eng 2016;88:76–91https://doi.org/10.1016/j.soildyn.
2016.05.008.

11
J. Liu, et al. Soil Dynamics and Earthquake Engineering 127 (2019) 105847

[4] Carbonari S, Morici M, Dezi F, Gara F, Leoni G. Soil-structure interaction effects in :7 < 951::aid-nme989 > 3.0.co;2-t.
single bridge piers founded on inclined pile groups. Soil Dyn Earthq Eng [28] Álvarez-Rubio S, Sánchez-Sesma FJ, Benito JJ, Alarcon E. The direct boundary
2017;92:52–67https://doi.org/10.1016/j.soildyn.2016.10.005. element method: 2D site effects assessment on laterally varying layered media
[5] Mirzaie F, Mahsuli M, Ghannad MA. Probabilistic analysis of soil-structure inter- (methodology). Soil Dyn Earthq Eng 2004;24(2):167–80https://doi.org/10.1016/j.
action effects on the seismic performance of structures. Earthq Eng Struct Dyn soildyn.2003.09.003.
2017;46:641–60https://doi.org/10.1002/eqe.2807. [29] Karabalis DL, Beskos DE. Dynamic response of 3-D rigid surface foundations by time
[6] Huang JQ, Zhao M, Du XL. Non-linear seismic responses of tunnels within normal domain boundary element method. Earthq Eng Struct Dyn
fault ground under obliquely incident P waves. Tunn Undergr Space Technol 2010;12(1):73–93https://doi.org/10.1002/eqe.4290120106.
2017;61:26–39https://doi.org/10.1016/j.tust.2016.09.006. [30] Francis C, Chrysoula T. Application of the perfectly matched absorbing layer model
[7] Tabatabaiefar HR, Fatahi B, Ghabraie K, Zhou W. Evaluation of numerical proce- to the linear elastodynamic problem in anisotropic heterogeneous media. Geophys J
dures to determine seismic response of structures under influence of soil-structure Int 2006;167(1):294–307https://doi.org/10.1190/1.1444908.
interaction. Struct Eng Mech 2015;56(1):27–47https://doi.org/10.12989/sem. [31] Papadopoulos M, Beeumen RV, François S, Degrande G, Lombaert G. Modal char-
2015.56.1.027. acteristics of structures considering dynamic soil-structure interaction effects. Soil
[8] Fatahi B, Tabatabaiefar HR, Samali B. Performance based assessment of dynamic Dyn Earthq Eng 2018;105:114–8https://doi.org/10.1016/j.soildyn.2017.11.012.
soil-structure interaction effects on seismic response of building frames. Proceedings [32] Liao ZP. A transmitting boundary for the numerical simulation of elastic wave
of georisk 2011 - geotechnical risk assessment & management (geotechnical special propagation. Int J Soil Dyn Earthq Eng 1984;3(4):174–83https://doi.org/10.1016/
publication No. 224). American Society of Civil Engineers (ASCE); 2011. p. 0261-7277(84)90033-0.
344–51https://doi.org/10.1061/41183(418)29. [33] Du XL, Zhao M. A local time-domain transmitting boundary for simulating cy-
[9] Yoshida N. Equivalent linear method considering frequency dependent character- lindrical elastic wave propagation in infinite media. Soil Dyn Earthq Eng
istics of stiffness and damping. Soil Dyn Earthq Eng 2002;22(3):205–22https://doi. 2010;30(10):937–46https://doi.org/10.1016/j.soildyn.2010.04.004.
org/10.1016/S0267-7261(02)00011-8. [34] Georges K, Olivier V, Calogero C. Finite-dynamic model for infinite media: cor-
[10] Lin Y, Miranda E. Evaluation of equivalent linear methods for estimating target rected solution of viscous boundary efficiency. J Eng Mech
displacements of existing structures. Eng Struct 2009;31(12):3080–9https://doi. 2011;137(7):509–11https://doi.org/10.1061/(ASCE)EM.1943-7889.0000250.
org/10.1016/j.engstruct.2009.08.009. [35] Liu JB, Li B. A unified viscous-spring artificial boundary for 3-D static and dynamic
[11] Kontoe S, Avgerinos V, Potts DM. Numerical validation of analytical solutions and applications. Sci China Ser E Technol Sci 2005;48(5):570–84https://doi.org/10.
their use for equivalent-linear seismic analysis of circular tunnels. Soil Dyn Earthq 1360/04ye0362.
Eng 2014;66:206–19https://doi.org/10.1016/j.soildyn.2014.07.004. [36] Wang Z, Zhao C, Dong L. An approximate spring-dashpot artificial boundary for
[12] Hashemi A, Elkhoraibi T, Ostadan F. Effects of structural nonlinearity and foun- transient wave analysis of fluid-saturated porous media. Comput Geotech
dation sliding on probabilistic response of a nuclear structure. Nucl Eng Des 2009;36(1–2):199–210https://doi.org/10.1016/j.compgeo.2008.01.008.
2015;295:887–900https://doi.org/10.1016/j.nucengdes.2015.08.028. [37] Liu JB, Du YX, Du XL, Wang ZY, Wu J. 3D viscous-spring artificial boundary in time
[13] Huang JQ, Du XL, Jin L, Zhao M. Impact of incident angles of P waves on the domain. Earthq Eng Eng Vib 2015;5(1):93–101https://doi.org/10.1007/s11803-
dynamic responses of long lined tunnels. Earthq Eng Struct Dyn 006-0585-2.
2016;45(15):2435–54https://doi.org/10.1002/eqe.2772. [38] Wolf JP. Soil-structure-interaction analysis in time domain. Prentice Hall,
[14] Lokke A, Chopra AK. Direct finite element method for nonlinear analysis of semi- Englewood Cliffs, NJ.
unbounded dam-water-foundation rock systems. Earthq Eng Struct Dyn [39] Pan JW, Zhang CH, Wang JT, Xu YJ. Seismic damage-cracking analysis of arch dams
2017;46(8):1267–85https://doi.org/10.1002/eqe.2855. using different earthquake input mechanisms. Sci China Ser E Technol Sci
[15] Tabatabaifar SHR, Fatahi B, Samali B. Finite difference modelling of soil-structure 2009;52(2):518–29https://doi.org/10.1007/s11431-008-0303-6.
interaction for seismic design of moment resisting building frames. Aust Geomech J [40] Jeremić Boris, Jie G, Preisig M, Nima Tafazzoli. Time domain simulation of soil-
2012;47(3):113–9. foundation-structure interaction in non-uniform soils. Earthq Eng Struct Dyn
[16] Tabatabaiefar HR, Fatahi B. Idealisation of soil–structure system to determine in- 2010;38(5):699–718https://doi.org/10.1002/eqe.896.
elastic seismic response of mid-rise building frames. Soil Dyn Earthq Eng [41] Poursartip B, Fathi A, Kallivokas LF. Seismic wave amplification by topographic
2014;66:339–51https://doi.org/10.1016/j.soildyn.2014.08.007. features: a parametric study. Soil Dyn Earthq Eng 2017;92:503–27https://doi.org/
[17] Tabatabaiefar HR, Mansoury B. Detail design, building and commissioning of tall 10.1016/j.soildyn.2016.10.031.
building structural models for experimental shaking table tests. Struct Des Tall [42] Li Y, Zhao M, Xu CS, Du XL, Zheng L. Earthquake input for finite element analysis of
Special Build 2016;25(8):357–74https://doi.org/10.1002/tal.1262. soil-structure interaction on rigid bedrock. Tunn Undergr Space Technol
[18] Tomeo R, Bilotta A, Pitilakis D, Nigro E. Soil-structure interaction effects on the 2018;79:250–62https://doi.org/10.1016/j.tust.2018.05.008.
seismic performances of reinforced concrete moment resisting frames. Procedia Eng [43] Bielak J, Loukakis K, Hisada Y, Yoshimura C. Domain reduction method for three-
2017;199:230–5https://doi.org/10.1016/j.proeng.2017.09.006. dimensional earthquake modeling in localized regions, part I: theory. Bull Seismol
[19] Abate G, Massimino MR. Dynamic soil-structure interaction analysis by experi- Soc Am 2003;93(2):817–24https://doi.org/10.1785/0120010251.
mental and numerical modelling. Riv Ital Geotec 2016;50(2):44–70. [44] Yoshimura C, Bielak J, Hisada Y, Fernández A. Domain reduction method for three-
[20] Massimino MR, Abate G, Corsico S, Louarn R. Comparison between two approaches dimensional earthquake modeling in localized regions, part II: verification and
for non-linear FEM modelling of the seismic behaviour of a coupled soil-structure applications. Bull Seismol Soc Am 2003;93(2):825–41https://doi.org/10.1785/
system. Geotech Geol Eng 2019;37(3):1957–75https://doi.org/10.1007/s10706- 0120010252.
018-0737-y. [45] Liu JB, Lu Y. A direct method for analysis of dynamic soil-structure interaction
[21] Lou M, Wang H, Chen X, Zhai Y. Structure-soil-structure interaction: literature re- based on interface idea. Dev Geotech Eng 1998;83(3):261–76https://doi.org/10.
view. Soil Dyn Earthq Eng 2011;31(12):1724–31https://doi.org/10.1016/j.soildyn. 1016/S0165-1250(98)80018-7.
2011.07.008. [46] Liu JB, Tan H, Bao X, Wang DY, Li ST. The seismic wave input method for soil-
[22] Semblat JF, Kham M, Bard PY. Seismic-wave propagation in alluvial basins and structure dynamic interaction analysis based on the substructure of artificial
influence of site-city interaction. Bull Seismol Soc Am 2008;98(6):2665–78https:// boundaries. Chin J Theor Appl Mech 2018;50(1):32–43. [in Chinese)]. https://doi.
doi.org/10.1785/0120080093. org/10.6052/0459-1879-17-336.
[23] Isbiliroglu Y, Taborda R, Bielak J. Coupled soil-structure interaction effects of [47] Kim JM, Lee EH, Lee SH. Boundary reaction method for nonlinear analysis of soil-
building clusters during earthquakes. Earthq Spectra 2015;31(1):463–500https:// structure interaction under earthquake loads. Soil Dyn Earthq Eng
doi.org/10.1193/102412eqs315m. 2016;89:85–90https://doi.org/10.1016/j.soildyn.2016.07.020.
[24] Sahar D, Narayan JP, Kumar N. Study of role of basin shape in the site-city inter- [48] Liu JB, Wang Y. A 1D time-domain method for in-plane wave motions in a layered
action effects on the ground motion characteristics. Nat Hazards half-space. Acta Mech Sin 2007;23(6):673–80https://doi.org/10.1007/s10409-
2015;75(2):1167–86https://doi.org/10.1007/s11069-014-1366-2. 007-0114-1.
[25] Lu X, Tian Y, Wang G, Huang D. A numerical coupling scheme for nonlinear time [49] Zhao M, Gao Z, Wang L, Du X, Huang J, Li Y. Obliquely incident earthquake input
history analysis of buildings on a regional scale considering site-city interaction for soil-structure interaction in layered half space. Earthq Struct
effects. Earthq Eng Struct Dyn 2018;47:2708–25https://doi.org/10.1002/eqe.3108. 2017;13(6):573–88https://doi.org/10.12989/eas.2017.13.6.573.
[26] Zhao C. Applications of infinite elements to dynamic soil-structure interaction [50] Liu JB, Gu Y, Du YX. Consistent viscous-spring artificial boundaries and viscous-
problems. Dev Geotech Eng 1998;83:153–60https://doi.org/10.1016/S0165- spring boundary elements. Chin J Geotech Eng 2006;28(9):1070–5. [in Chinese)].
1250(98)80010-2. https://doi.org/10.3321/j.issn:1000-4548.2006.09.004.
[27] Astley RJ. Infinite elements for wave problems: a review of current formulations [51] Fatahi B, Tabatabaifar SHR, Samali B. Soil-structure interaction vs site effect for
and an assessment of accuracy. Int J Numer Methods Eng seismic design of tall buildings on soft soil. Geomech Eng
2000;49(7):951–76.https://doi.org/10.1002/1097-0207(20001110)49. 2014;6(3):293–320https://doi.org/10.12989/gae.2014.6.3.293.

12

You might also like