You are on page 1of 9

Simulation of Spatially Correlated Nonstationary Response

Spectrum–Compatible Ground Motion Time Histories


M. D. Shields, A.M.ASCE1
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

Abstract: A new methodology is presented for the generation of spatially correlated nonstationary ground motion time histories that are com-
patible with a prescribed response spectrum using the spectral representation method. The method introduces two important improvements over
the current state of the art in that it preserves the coherence among the ground motion histories and enables the incorporation of both time and
frequency modulation by upgrading the evolutionary power spectral density function with random pulse-like perturbations. An example is
provided for the simulation of design spectrum–compatible, uniformly modulated nonstationary acceleration time histories at three locations
on the ground, and the results are compared directly with an existing state-of-the-art methodology. DOI: 10.1061/(ASCE)EM.1943-
7889.0000884. © 2014 American Society of Civil Engineers.
Author keywords: Response spectrum; Spectral representation method (SRM); Earthquake ground motion; Stochastic process; Non-
stationary; Simulation.

Introduction compatible with a prescribed response spectrum. These develop-


ments began in the 1970s with the work of Tsai (1972), Gasparini
The response spectrum has gained widespread acceptance as one of and Vanmarcke (1976), and Kaul (1978a) and have evolved for a
the fundamental descriptions of the effects of seismic ground motion wide variety of applications in the decades since. These develop-
on the behavior of structures (Chopra 2001). Although site-specific ments have generally progressed in two directions: (1) modification
strong ground motion records are available for many seismically of existing ground motion records for response-spectrum compat-
active regions worldwide, artificial ground motion records remain ibility and (2) generation of artificial ground motion records as
very useful for cases where multiple analyses are necessary (such as realizations of a nonstationary stochastic process. Methods based on
in Monte Carlo simulations) to evaluate, for example, the variability the modification of existing ground motion records are often con-
in the structural response, the maximum likely response, or some sidered preferable, and some notable developments along these lines
performance/reliability metric of the response. Artificial ground motion include the following. The WES RASCAL code developed by Silva
records are also useful, in certain cases, for singular analyses where and Lee (1987) iteratively modifies the Fourier amplitude spectrum
data are not available. In such cases, it is often desirable to produce of recorded histories to make them compatible with a prescribed
artificial records that are in some way compatible with a prescribed response spectrum. The RSPMATCH code (Abrahamson 1992) uses
response spectrum. Moreover, modern design codes (e.g., European the time-domain spectral matching methods proposed by Kaul
Committee for Standardization 2003; ASCE 2010) require artifi- (1978a) and Lilhanand and Tseng (1988) and has been updated in the
cially generated ground motions to be compatible with site-specific RSPMatch2005 code (Hancock et al. 2006), which uses wavelet-
response spectra for dynamic analysis. based time-domain adjustments on recorded accelerograms to make
Definitions of response-spectrum compatibility vary. In general, the records response-spectrum compatible. Similarly, Mukherjee
response-spectrum compatibility requires that the mean response and Gupta (2002) use a continuous wavelet transform of a record to
spectrum from a set of artificial records matches the target response modify it for response-spectrum compatibility. Recently, Cacciola
spectrum with some specified degree of accuracy over a given fre- (2010) developed a methodology based on the superposition of a
quency range. The defined target response spectrum varies with ground motion record and a correcting stochastic process intended
application as well and may correspond to the response spectrum to make the combined history compatible with a prescribed response
from a given record, the median or mean response spectrum from spectrum.
a set of records, a design spectrum, or a uniform hazard spectrum. It is not always possible to modify existing records to make them
The methodology presented herein considers the compatibility of response-spectrum compatible while preserving certain important
a single generated time history with any given target response features of the records such as coherence. This is particularly true in
spectrum (of the types previously listed or others) based on an the important case where multiple correlated records are necessary
overall error metric across a given frequency range. for analysis [see Zerva (2009) for a detailed discussion of this topic].
Numerous methods have been developed over the past several Such circumstances arise in the analysis of unsymmetrical buildings
decades to generate artificial ground motion histories that are or portfolios of buildings, long-span bridges with multiple support
excitations, dams (Sanchez Lizarraga and Lai 2014), pipelines
1 (Zerva 1994), and nuclear power plants (Jeremic et al. 2013), among
Assistant Professor, Dept. of Civil Engineering., Johns Hopkins
other analyses. In these instances, it has generally been the practice
Univ., Baltimore, MD 21218. E-mail: michael.shields@jhu.edu
Note. This manuscript was submitted on January 8, 2014; approved on to simulate ground motion as a multivariate (MV) nonstationary sto-
September 24, 2014; published online on October 21, 2014. Discussion chastic process (Shinozuka and Deodatis 1988; Vanmarcke and Fenton
period open until March 21, 2015; separate discussions must be submitted 1991; Zerva 1992; Zentner 2013). However, relatively few method-
for individual papers. This paper is part of the Journal of Engineering ologies have been proposed for generating artificial ground motion
Mechanics, © ASCE, ISSN 0733-9399/04014161(9)/$25.00. time histories that include spatial correlation and are compatible with

© ASCE 04014161-1 J. Eng. Mech.

J. Eng. Mech.
prescribed response spectra. Hao et al. (1989) proposed a method- where Ai ðv, tÞ 5 modulating functions describing the time variation
ology that iteratively updates the Fourier amplitude spectrum of a of the amplitude and frequency; Si ðvÞ 5 (stationary) PSDFs; and
uniformly modulated nonstationary stochastic process. Deodatis Gij ðvÞ 5 complex coherence functions.
(1996a) proposed a method that is conceptually similar to that pro- The SRM requires the decomposition of the evolutionary CSDM
posed by Hao et al. (1989), except the power spectral density function at every time instant as
(PSDF) is iteratively upgraded. More recently, Cacciola and Deodatis
(2011) proposed a methodology that is the MV extension of the meth- Sðv, tÞ ¼ Hðv, tÞHTp ðv, tÞ (4)
odology proposed by Cacciola (2010).
Perhaps the most widely used technique is the spectral representation– where superscript T p 5 conjugate transpose. The process is sim-
based technique developed by Deodatis (1996a) that iterates on the ulated by the following series as N → ‘
PSDFs at each location to produce spatially correlated time histories
P N 
m P pffiffiffiffiffiffiffi  
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

at several locations with specified separation. This methodology Hij ðvk , tÞ Dvcos vk t 2 uij ðvk Þ þ Fjk
fi ðtÞ ¼ 2
has proven quite effective, but the iterative process has one critical j51 k51
shortcoming—it causes the simulated ground motion histories to
deviate from the Gaussian distribution and causes a corresponding ði ¼ 1, 2, . . . , mÞ (5)
loss of coherence among the simulated histories. Additionally, the
methodology in Deodatis (1996a) cannot simulate frequency- where
modulated nonstationary histories; instead, it is limited to simu-
lating so-called uniformly modulated nonstationary ground motion vk ¼ kDv ðk ¼ 0, 1, . . . , N 2 1Þ (6)
histories. Cacciola and Zentner (2012) recently proposed a method-
ology for the generation of univariate response spectrum–compatible vu
Dv ¼ (7)
histories that includes frequency modulation. The intention of the N
current work is to develop a spectral representation–based method-
(  )
ology that produces fully (amplitude- and frequency-modulated) Im Hij ðvk , tÞ
21  
nonstationary ground motion histories that are response-spectrum uij ðvk Þ ¼ tan (8)
Re Hij ðvk , tÞ
compatible without losing coherence.
The paper begins with a review of the spectral representation
method (SRM) for the simulation of MV nonstationary stochastic in which uij ðvk Þ 5 phase angles resulting from wave passage, where
processes upon which the proposed methodology relies. A short re- Im½  and Re½  denote the real and imaginary components of Hij ðvk , tÞ,
view of the methodology proposed by Deodatis (1996a) is then respectively; vu 5 upper cut-off frequency beyond which elements
provided, along with a discussion of its general strengths and of the CSDM may be assumed to be zero; and Fjk 5 components of
weaknesses. Next, a new iterative methodology is proposed wherein the matrix F representing m-sequences of independent random
the evolutionary PSDF is updated through random functional per- phase angles uniformly distributed over (0, 2p).
turbations. A short discussion is provided that explores the benefits The uniformly modulated nonstationary stochastic processes
and drawbacks of the proposed methodology with a focus on im- that have only amplitude modulation and do not have frequency
proving computational efficiency. A numerical example follows in modulation comprise a special class of nonstationary processes.
which ground motion is simulated at three points on the ground as For uniformly modulated processes, the modulating functions in
a trivariate uniformly modulated stochastic process, and the results Eqs. (2)–(3) are not time dependent, and simulations are performed
are compared directly with the methodology proposed by Deodatis by
(1996a). Finally, some concluding remarks are given.
fi ðtÞ ¼ Ai ðtÞgi ðtÞ ði ¼ 1, 2, . . . , mÞ (9)

SRM for MV Nonstationary Stochastic Processes where Ai ðtÞ 5 (time dependent only) modulating functions; and
gi ðtÞ 5 realizations of a MV zero-mean stationary process with
The SRM for the simulation of Gaussian stochastic processes was (stationary) CSDM SðvÞ simulated using the stationary formulation
introduced by Shinozuka and Jan (1972). The one-dimensional (1D), of the SRM (Shinozuka and Jan 1972).
MV nonstationary formulation of the SRM relies on Priestley’s
theory of evolutionary power (Priestley 1965) and produces sample
realizations of the process with a prescribed evolutionary cross- Review of Existing Methodology
spectral density matrix (CSDM)
2 3 The methodology proposed by Deodatis (1996a) produces uni-
S11 ðv, tÞ S12 ðv, tÞ / S1m ðv, tÞ formly modulated nonstationary response spectrum–compatible
6 S ðv, tÞ S ðv, tÞ / S ðv, tÞ 7
6 21 22 2m 7 time histories in an iterative fashion. The PSDFs Sii ðvÞ are ini-
Sðv, tÞ ¼ 6 7 (1)
4 « « ⋱ « 5 tialized and the (stationary) CSDM is evaluated as
Sm1 ðv, tÞ Sm2 ðv, tÞ / Smm ðv, tÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Sij ðvÞ ¼ Si ðvÞSj ðvÞGij ðvÞ (10)
The theory of evolutionary power enables the elements of the evo-
lutionary CSDM to be written as Ground motion time histories fi ðtÞ are generated using Eq. (9). The
ð jÞ
2
Sii ðv, tÞ ¼ jAi ðv, tÞj Si ðvÞ ði ¼ 1, 2, . . . , mÞ (2) response spectra of these time histories, Ri ðvÞ, are computed and
the PSDFs, Sii ðvÞ, are upgraded using the following relationship:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi " #2
Sij ðv, tÞ ¼ Ai ðv, tÞAj ðv, tÞ Si ðvÞSj ðvÞGij ðvÞ ð jþ1Þ ð jÞ RTi ðvÞ
Sii ðvÞ ¼ Sii ðvÞ ð jÞ (11)
ði, j ¼ 1, 2, . . . , m; i  jÞ (3) Ri ðvÞ

© ASCE 04014161-2 J. Eng. Mech.

J. Eng. Mech.
ð jÞ
where superscript (j) 5 current iteration. A new realization of the iteration j is denoted by Ri ðv, FÞ to show this weak dependence.
MV process is simulated with the upgraded CSDM, and the process Applying this notation to Eq. (11) yields
is repeated until the computed response spectra are sufficiently close " #2
to the target response spectra. ð jþ1Þ ð jÞ RTi ðvÞ
The methodology in Deodatis (1996a) converges quite rapidly Sii ðv, FÞ ¼ Sii ðv, FÞ ð jÞ
(12)
and produces time histories that are, for many purposes, sufficiently Ri ðv, FÞ
consistent with the prescribed response spectrum. Adequate con-
ð0Þ
vergence is typically achieved within approximately 10 iterations. where only Sii ðvÞ is independent of F. Eq. (12) demonstrates that
However, the methodology has certain limitations. Most notably, it as the iterations progress the dependence of Sii ðvÞ on F grows
has been shown to produce time histories that lose coherence as the stronger. The CSDM therefore becomes dependent on the selected
iterations progress. The iterations also result in a deviation from the phase angles, which has a corrupting effect on the ensemble prop-
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

Gaussian distribution (Deodatis and Micaletti 2001) that precludes erties of the histories generated using Eq. (5). Specifically, Shinozuka
the evaluation of the sample properties (i.e., the distributions of and Deodatis (1991) and Deodatis (1996b) have shown that the time
the extremes) and the use of these time histories with Grigoriu’s histories generated by Eq. (5) are Gaussian by the central limit
translation process theory (Grigoriu 1995) if the time histories are theorem. However, the central limit theorem is violated given the
desired to be non-Gaussian. dependence of Sii ðvÞ on F. More importantly for this work, Deodatis
Both the deviation from the Gaussian distribution and the loss of (1996b) has shown that the ensemble auto/cross-correlation functions
coherence are caused by the computed response spectrum’s de- of the histories generated by Eq. (5) match their targets owing to the
pendence on the specific selection of random phase angles, F, in the independence of the components of F. The dependence of Sii ðvÞ on
simulation. To represent this, the computed response spectrum at F corrupts these ensemble auto/cross-correlation functions. Because
coherence is the frequency domain analog of cross-correlation, the
loss of ensemble cross-correlation produces a corresponding loss in
coherence.
Additionally, the applicability of this methodology is limited to
the class of uniformly modulated nonstationary processes; temporal
variation in frequency content, which has been shown to be im-
portant for inelastic structural dynamics (Yeh and Wen 1990; Wang
et al. 2002), cannot be achieved. With this in mind, the intention of
this work is to develop a method for producing correlated, fully
nonstationary ground motion time histories that are consistent with
a prescribed response spectrum and maintain their desired coherence
(resulting from the preservation of the Gaussian properties of the
SRM). The following section is dedicated to outlining the proposed
methodology.

Fig 2. Configuration of three points for ground motion simulation with


direction of wave propagation

Fig 1. Proposed iterative method for simulation of MV response


spectrum–compatible ground motion time histories using SRM Fig 3. Modulating function for ground motion time history at Point 1

© ASCE 04014161-3 J. Eng. Mech.

J. Eng. Mech.
Proposed Methodology 3. Apply random perturbations to each modulating function.
Perturbations to the modulating function are applied by
The following offers a step-by-step outline of the proposed pro-
cedure. The technique is also outlined schematically in Fig. 1. ð jþ1Þ ð jÞ
Ai ðv, tÞ ¼ Ai ðv, tÞ þ DAi ðv, tÞ (13)
1. Prescribe the target response spectra and complex coherences.
For each ground motion history, specify the target response
spectrum RTi ðvÞ and complex coherence functions Gij ðvÞ. A suggested perturbation is given as follows:
2. Initialize the CSDM. Initialize the (stationary) PSDFs for  
each ground motion history as white noise [Si ðvÞ 5 1, "v]. ðv 2 v0 Þ2 ðt 2 t0 Þ2
DAi ðv, tÞ ¼ B exp 2 × (14)
Initialize the modulating functions with arbitrarily selected c d
positive values [Ai ðv, tÞ . 0, "v, t]. The CSDM is computed
from the PSDFs, modulating functions, and complex coherence where B, (v0 , t0 ), and (c, d) 5 parameters specifying the
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

functions using Eqs. (2) and (3). magnitude, location, and spread (width), respectively, of the
pulse. The parameter B has a random sign and a magnitude
less than that of Ai ðv0 , t0 Þ [to avoid producing Sii ðv, tÞ # 0].
Parameters v0 and t0 are uniformly distributed random var-
iables over the ranges (0, vu ) and (0, T), respectively. Param-
eters c and d can be adjusted to produce pulses of the desired
width. Perturbations of this form are selected because they
influence Ai ðv, tÞ most strongly at (v0 , t0 ) and have diminish-
ing influence as separation from this point grows, with the
pffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffi negligible at points v  v0 6 3 c=2
perturbation becoming
and t  t0 6 3 d=2.
4. Simulate the ground motion time histories. Generate realiza-
tions fi ðtÞ of the MV nonstationary process with specified
CSDM using Eq. (5).
5. Compare the response spectra. Numerically evaluate the re-
ð jÞ
sponse spectra Ri ðvÞ for fi ðtÞ at each iteration ( j). Compute
Fig 4. Abrahamson model for (stationary) coherence functions between ð jÞ ð jÞ
Points 1, 2, and 3 the relative error, ɛ i , between Ri ðvÞ and the target response
spectra RTi ðvÞ. One suggested error metric is the following:

Fig 5. Simulated acceleration time histories and corresponding PSDFs at (a) Point 1; (b) Point 2; (c) Point 3

© ASCE 04014161-4 J. Eng. Mech.

J. Eng. Mech.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u h i Example
uPN21 Rð jÞ ðv Þ 2 RT ðv Þ 2
u k50 i k k
¼t
ð jÞ i
ɛi PN21  T 2 (15) In this section, a numerical example is provided for the generation
k50 Ri ðvk Þ of response spectrum–compatible ground motion at three points on
the ground surface as a trivariate uniformly modulated stochastic
process. Although the methodology proposed in the previous section
although others may certainly be used (alternatives may be con- affords the simulation of a nonseparable magnitude- and frequency-
sidered, for instance, when it is only required to match the re- modulated nonstationary stochastic ground motion, a uniformly
sponse spectrum for a limited range of frequencies). For each modulated process is studied here to perform a direct comparison
ground motion history, check to see if the error has improved with the earlier methodology in Deodatis (1996a). To accommodate
ð jÞ ? ð j21Þ ð jÞ ð j21Þ
from the previous iteration (ɛ i , ɛ i ). If ɛ i . ɛ i , the this, the following simplifications are made in the proposed meth-
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

perturbation is rejected, the modulating function corresponding odology. The time- and frequency-dependent components of the
to history i is returned to its previous form, and a new iteration evolutionary PSDF for a uniformly modulated process are separable
ð jÞ ð j21Þ
begins by returning to Step 3. If ɛ i , ɛ i , the perturbation is such that the modulating functions in Eqs. (2)–(3) are independent of
accepted, and iterations proceed to Step 6. frequency [i.e., Ai ðv, tÞ 5 Ai ðtÞ], and simulation may be performed
6. Check the compatibility. Compare the errors computed for using Eq. (9). Consequently, the perturbations are applied to the
each time history in Step 5 with a specified threshold error PSDF of the amplitude-modulated stationary process gi ðtÞ in Eq. (9)
ð jÞ
ɛ th . If ɛ i , ɛ th "i, then the simulation is complete. Otherwise, as follows:
another iteration is performed by returning to Step 3.  
The proposed methodology preserves the important Gaussian ðv 2 v0 Þ2
character of the generated time histories, because the evolutionary DSi ðv, tÞ ¼ B exp 2 (16)
c
PSDF is being randomly updated. This is important for two reasons.
First, the prescribed coherences among histories are preserved. For this example, adaptive perturbations are applied such that c 5 6
Second, it enables the generation of non-Gaussian time histories ð jÞ ð jÞ
when ɛ i . 0:06 and c 5 0:6 when ɛ i # 0:06 [using Eq. (15)]. All
with a specific distribution when coupled with Grigoriu’s translation other steps remain unchanged.
process theory (Grigoriu 1995) and a method such as that proposed in
Shields et al. (2011) and Shields and Deodatis (2013a, b). However,
these benefits come at an appreciable computational cost. The pro-
posed technique may require a very large number of iterations to
converge (in some cases up to 20,000), compared with tens of iter-
ations for the methodology proposed in Deodatis (1996a). Moreover,
depending on the nature of the stochastic process, each iteration may
come at a considerable computational cost. During each iteration, time-
history generation is necessary. For fully nonstationary processes with
both time and frequency modulation, this time-history generation can
become quite expensive, because computationally efficient routines
using the fast Fourier transform (FFT), for instance, cannot be used.
This computational cost can be alleviated to a large degree. Us-
age of adaptive perturbations in which the perturbation width is
progressively narrowed or multigrid methods such as those used in
Benowitz et al. (2013) and B. Benowitz, M. Shields, and G. Deodatis
(“Determining evolutionary spectra from nonstationary autocorre-
lation functions,” submitted, Probabilistic Engineering Mechanics)
have been shown to reduce the number of iterations by 30% or more.
This may be further enhanced by introducing some logic or memory
into the random assignment of perturbation locations such that
perturbations will not continue in regions where they have been
repeatedly rejected and will be concentrated in regions with a high
probability of acceptance. Perhaps the means to most significantly
reduce computational cost is to begin with an initial evolutionary
CSDM that is close to the expected final one. This can be done by
leveraging prior work to estimate a PSDF directly from the target
response spectrum, such as by using the methods proposed in
Gasparini and Vanmarcke (1976), Kaul (1978b), Unruh and Kana
(1981), Pfaffinger (1983), Spanos and Vargas Loli (1985), Christian
(1989), Park (1995), Gupta and Trifunac (1998), and Kjell (2002). This
has been shown to reduce the initial computed error by 70% or more.
Lastly, the proposed methodology represents a specific variety of
stochastic search algorithms sharing many similarities with simulated
annealing (Kirkpatrick et al. 1983) in particular. Although simulated
annealing is not likely to improve the computational efficiency of the Fig 6. Target response spectra (dashed line) and response spectra
proposed methodology given the high-dimensional continuous search (solid line) for simulated time histories of ground motion at (a) Point 1;
space, other stochastic search methods may prove beneficial in lieu of (b) Point 2; (c) Point 3
the random functional perturbation method used herein.

© ASCE 04014161-5 J. Eng. Mech.

J. Eng. Mech.
Problem Definition where ji1 5 distance from Point 1 to Point i; and v 5 2,000 m=s
5 velocity of wave propagation. Here, j21 5 50, and j31 5 100.
Consider the acceleration time histories at three points on the ground
For demonstration purposes, the stationary coherence func-
surface to be a trivariate, uniformly modulated nonstationary sto-
tions, gij ðvÞ, where i, j 5 1, 2, 3 and i  j (plotted in Fig. 4), de-
chastic process as shown in Fig. 2 with the direction of the seismic
fining the correlation between the acceleration histories fi ðtÞ
wave propagation as shown. The three points correspond to different
and fj ðtÞ follow the model proposed by Abrahamson (1993) as
local soil conditions as follows:

follows:
Point 1: ASCE/SEI 7-10 Site Class A (hard rock); 8
• Point 2: ASCE/SEI 7-10 Site Class C (very dense soil); and >
>


< c3 jij
Point 3: ASCE/SEI 7-10 Site Class E (soft soil) 1
g ij ðvÞ ¼ " #6  tanh
>


:1 þ v c4 jij þ v c7 jij
Design spectra for these sites are constructed in accordance with 2
v >
ASCE/SEI 7-10 (ASCE 2010) and the 2012 International Building 1þ
2p 4p 2
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

Code (IBC) (International Code Council 2012). 2pc8 jij


The modulating functions of the time histories (shown in Fig. 3 9
>
>
for Point 1) are assumed to follow the form proposed by Jennings 
 h
v i =
et al. (1968) þ 4:8 2 c3 jij exp c6 jij þ 0:35
2p >
>
80 ;
1
>
> j 2
> t 2 i1
> j ði, j ¼ 1, 2, 3; i  jÞ
>
> B vC
>
> @ A t 2 i1 # 2
>
> 2 v (18)
>
>
<
Ai ðtÞ ¼ ji1 (17) where jij 5 distance between Points i and j; and
>
> # 2 #
>
> 1 2 t 9
>
> v
3:95

>
>    c3 jij ¼ þ 0:85 exp 20:00013jij
>
> 1 þ 0:0077jij þ 0:000023jij
2
>
> j j
: exp 20:4 t 2 i1 2 9 t 2 i1 $ 9
v v (19)

Fig 7. Gaussian probability plots for 20 generated time histories using methodology in Deodatis (1996a) at (a) Point 1; (b) Point 2; (c) Point 3; Gaussian
probability plots for three generated time histories using proposed methodology at (d) Point 1; (e) Point 2; (f) Point 3

© ASCE 04014161-6 J. Eng. Mech.

J. Eng. Mech.
2 3
methods. The proposed method produces histories with lower peak
6 7 accelerations, because the method produces a much more refined
6 1 7
0:461 2  3 7 PSDF, whereas Deodatis’ method (Deodatis 1996a) produces a
4 jij 5
1þ comparatively smooth PSDF resulting in significantly more total

5 power (i.e., the stochastic process has larger variance, and the
c4 jij ¼ "  #"  3 # (20)
jij 8
jij resulting ground motion has more energy).
1þ 1þ More importantly for the purposes here, time histories simulated
190 180
using the proposed method possess the prescribed coherence in
   ensemble. Fig. 8 shows the mean computed coherence from 100

jij time-history simulations generated using the methodology proposed
c6 jij ¼ 3 exp 2 2 1 2 0:0018jij (21) by Deodatis (1996a). A clear loss of coherence is observed using this
20
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

methodology—particularly in the low-frequency range. Note also





that the mean computed coherence is not smooth. This is a result
c7 jij ¼ 20:598 þ 0:106 ln jij þ 325 2 0:0151 exp 20:6jij of the complex dependence of the CSDM on F, as discussed
(22) previously. Shields (2010) has shown that, in addition to the co-
herence loss growing with the number of iterations, the variability of




c8 jij ¼ exp 8:54 2 1:07 ln jij þ 200 þ 100 exp 2jij (23) the coherence grows with the number of iterations. It is known that
the variability of the coherence function around its mean value is
heteroscedastic (i.e., frequency dependent). In general, when using
the SRM this variability increases with frequency. However, the
The preceding model gives plane-wave coherency functions
iterations have the effect of producing higher variability at lower
describing the spatial variation in strong ground motion in the fre-
frequencies. The precise reason for this reversal of heteroscedasticity
quency domain under the assumption that the seismic wave can be
is unclear, because the complex dependency of the CSDM on the
described by a plane wave. A detailed explanation of plane-wave
phase angles cannot be explored analytically. Meanwhile, Fig. 9
spatial ground motion coherency models of this form is provided
shows that the mean computed coherence from 100 time-history
by Abrahamson et al. (1991) where the coherency model is de-
veloped empirically from seismic array data. This model does not
include wave propagation. From the (stationary) coherence functions,
the complex coherence functions accounting for wave propagation
are given by

vjij
Gij ðvÞ ¼ g ij ðvÞexp 2I ði, j ¼ 1, 2, 3; i  jÞ (24)
v
pffiffiffiffiffiffiffi
where I 5 21.

Results
For each location, an initial white noise PSDF with Sii ðvÞ
5 100 cm2 =s3 is prescribed. The simulation is performed using
the proposed methodology (Fig. 1). Sample realizations of the
acceleration time histories and the associated PSDFs after 10,000
adaptive iterations are shown in Fig. 5.
The response spectra for these generated acceleration histories
are shown with their respective target response spectra in Fig. 6.
The corresponding errors computed from Eq. (15) for the computed
response spectra relative to the target response spectra are 4.1, 4.8,
and 5.3%, respectively.

Discussion
The response spectra computed from the time histories simulated
using the proposed methodology accurately match the target re-
sponse spectra. The histories also possess a Gaussian distribution.
Fig. 7 provides Gaussian probability plots for 20 time histories
using the methodology proposed in Deodatis (1996a) along with
Gaussian probability plots for three time histories using the meth-
odology proposed herein. Note that the histories are almost perfectly
Gaussian with the proposed methodology even when considering an
ensemble of only three histories. Meanwhile, the large number of Fig 8. Computed (plus marks) and prescribed (solid line) coherence
realizations using the methodology proposed in Deodatis (1996a) between time histories generated using simulation methodology pro-
shows a strong deviation from the Gaussian distribution. Note also posed in Deodatis (1996a) for (a) Point 1; (b) Point 2; (c) Point 3
the significant difference in the accelerations between the two

© ASCE 04014161-7 J. Eng. Mech.

J. Eng. Mech.
References

Abrahamson, N. (1992). “Non-stationary spectral matching.” Seismol. Res.


Lett., 63(1), 30.
Abrahamson, N. (1993). “Spatial variation of multiple support excitations.”
Proc., 1st U.S. Symp. on Seismic Evaluation and Retrofit, National
Information Service for Earthquake Engineering (NISEE), Univ. of
California, Berkeley, CA.
Abrahamson, N. A., Schneider, J. F., and Stepp, J. C. (1991). “Empirical
spatial coherency functions for application to soil-structure interaction
analyses.” Earthquake Spectra, 7(1), 1–27.
ASCE. (2010). “Minimum design loads for buildings and other structures.”
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

ASCE/SEI 7-10, Reston, VA.


Benowitz, B. A., Deodatis, G., and Shields, M. D. (2013). “Simulation of
non-Gaussian, non-stationary stochastic processes with evolutionary
power.” Proc., 2013 Int. Conf. on Structural Safety and Reliability
(ICOSSAR), International Association for Structural Safety and Re-
liability (IASSAR), New York.
Cacciola, P. (2010). “A stochastic approach for generating spectrum com-
patible fully nonstationary earthquakes.” Comput. Struct., 88(15–16),
889–901.
Cacciola, P., and Deodatis, G. (2011). “A method for generating fully non-
stationary and spectrum-compatible ground motion vector processes.”
Soil. Dyn. Earthquake Eng., 31(3), 351–360.
Cacciola, P., and Zentner, I. (2012). “Generation of response-spectrum-
compatible artificial earthquake accelerograms with random joint time–
frequency distributions.” Probab. Eng. Mech., 28(Apr), 52–58.
Chopra, A. K. (2001). Dynamics of structures: Theory and applications to
earthquake engineering, Prentice Hall, Upper Saddle River, NJ.
Christian, J. (1989). “Generating seismic design power spectral density
functions.” Earthquake Spectra, 5(2), 351–368.
Deodatis, G. (1996a). “Non-stationary stochastic vector processes: Seismic
ground motion applications.” Probab. Eng. Mech., 11(3), 149–168.
Deodatis, G. (1996b). “Simulation of ergodic multivariate stochastic pro-
cesses.” J. Eng. Mech., 10.1061/(ASCE)0733-9399(1996)122:8(778),
Fig 9. Using proposed simulation methodology, computed (solid line) 778–787.
and prescribed (dashed line) coherence between time histories generated Deodatis, G., and Micaletti, R. C. (2001). “Simulation of highly skewed non-
Gaussian stochastic processes.” J. Eng. Mech., 10.1061/(ASCE)0733-
for (a) Point 1; (b) Point 2; (c) Point 3
9399(2001)127:12(1284), 1284–1295.
European Committee for Standardization. (2003). “Design of structures for
earthquake resistance—Part 1: General rules, seismic actions and rules
for buildings.” Eurocode 8, Brussels, Belgium.
simulations generated using the proposed methodology matches the Gasparini, D., and Vanmarcke, E. (1976). “Simulated earthquake motions
target coherence quite well. compatible with prescribed response spectra.” Rep. No. R76-4, Mas-
sachusetts Institute of Technology (MIT), Cambridge, MA.
Grigoriu, M. (1995). Applied non-Gaussian processes, Prentice Hall, Upper
Conclusions Saddle River, NJ.
Gupta, I. D. and Trifunac, M. D. (1998). “Defining equivalent stationary
An improved spectral representation–based method for simulating PSDF to account for nonstationarity of earthquake ground motion.” Soil.
spatially correlated fully nonstationary response spectrum–compatible Dyn. Earthquake Eng., 17(2), 89–99.
ground motion time histories has been proposed that alleviates the Hancock, J., et al. (2006). “An improved method of matching response
problems of coherence loss and deviation from the Gaussian distri- spectra of recorded earthquake ground motion using wavelets.” J. Earth-
bution, which were caused by the iterations used in the preceding quake Eng., 10(S1), 67–89.
methodology by Deodatis (1996a). The new methodology additionally Hao, H., Oliveira, C., and Penzien, J. (1989). “Multiple-station ground
allows for the inclusion of both time and frequency modulation in the motion processing and simulation based on smart-1 array data.” Nucl.
nonstationary histories. An example has been presented for the gen- Eng. Des., 111(3), 293–310.
International Code Council. (2012). 2012 International building code (IBC),
eration of uniformly modulated nonstationary ground motion accel-
Washington, DC.
erations compatible with prescribed design spectra at three locations on Jennings, P., Housner, G., and Tsai, N. (1968). Simulated earthquake
the ground. The results show accurate compatibility of the time histories motions, Earthquake Engineering Research Laboratory, California In-
with the prescribed response spectra and are compared directly with the stitute of Technology, Pasadena, CA.
existing methodology to highlight the preservation of the Gaussian Jeremic, B., Tafazzoli, N., Ancheta, T., Orbovic, N., and Blahoianu, A.
distribution and coherence. (2013). “Seismic behavior of NPP structures subjected to realistic 3D,
inclined seismic motions, in variable layered soil/rock, on surface or
embedded foundations.” Nucl. Eng. Des., 265(Dec), 85–94.
Acknowledgments Kaul, M. K. (1978a). “Spectrum-consistent time-history generation.” J. Engrg.
Mech. Div., 104(4), 781–788.
The author is grateful to Prof. George Deodatis for his insight and Kaul, M. K. (1978b). “Stochastic characterization of earthquakes through
feedback on this work. their response spectrum.” Earthquake Eng. Struct. Dyn., 6(5), 497–509.

© ASCE 04014161-8 J. Eng. Mech.

J. Eng. Mech.
Kirkpatrick, S., Gelatt, C. D., Jr., and Vecchi, M. P. (1983). “Optimization Shinozuka, M., and Deodatis, G. (1988). “Stochastic process models for
by simulated annealing.” Science, 220(4598), 671–680. earthquake ground motion.” Probab. Eng. Mech., 3(3), 114–123.
Kjell, G. (2002). “Predicting response spectra for earthquake signals gen- Shinozuka, M., and Deodatis, G. (1991). “Simulation of stochastic processes
erated as filtered noise.” Probab. Eng. Mech., 17(3), 241–252. by spectral representation.” Appl. Mech. Rev., 44(4), 191–204.
Lilhanand, K., and Tseng, W. (1988). “Development and application of Shinozuka, M., and Jan, C.-M. (1972). “Digital simulation of random
realistic earthquake time histories compatible with multiple-damping processes and its applications.” J. Sound Vib., 25(1), 111–128.
design spectra.” Proc., 9th World Conf. on Earthquake Engineering, Silva, W. J., and Lee, K. (1987). “State-of-the-art for assessing earthquake
Indian Institute of Technology, Kanpur, India, 819–824. hazards in the United States.” WES RASCAL code for synthesizing
Mukherjee, S., and Gupta, V. K. (2002). “Wavelet-based generation of earthquake ground motions, U.S. Army COE, Washington, DC.
spectrum-compatible time-histories.” Soil. Dyn. Earthquake Eng., Spanos, P. D., and Vargas Loli, L. M. (1985). “A statistical approach to
22(9–12), 799–804. generation of design spectrum compatible earthquake time histories.”
Park, Y. J. (1995). “New conversion method from response spectrum to Int. J. Soil. Dyn. Earthquake Eng., 4(1), 2–8.
Tsai, N.-C. (1972). “Spectrum-compatible motions for design purposes.”
Downloaded from ascelibrary.org by Indian Institute of Technology, Kanpur on 11/03/14. Copyright ASCE. For personal use only; all rights reserved.

PSD functions.” J. Eng. Mech., 10.1061/(ASCE)0733-9399(1995)121:


12(1391), 1391–1392. J. Engrg. Mech. Div., 98(2), 345–356.
Pfaffinger, D. D. (1983). “Calculation of power spectra from response Unruh, J. F., and Kana, D. D. (1981). “An iterative procedure for generation
spectra.” J. Eng. Mech., 10.1061/(ASCE)0733-9399(1983)109:1(357), of consistent power/response spectrum.” Nucl. Eng. Des., 66(3), 427–
357–372. 435.
Priestley, M. (1965). “Evolutionary spectra and non-stationary processes.” Vanmarcke, E. H., and Fenton, G. A. (1991). “Conditioned simulation of local
J. R. Stat. Soc. B, 27(2), 204–237. fields of earthquake ground motion.” Struct. Saf., 10(1–3), 247–264.
Sanchez Lizarraga, H., and Lai, C. G. (2014). “Effects of spatial variability Wang, J., Fan, L., Qian, S., and Zhou, J. (2002). “Simulations of non-
of soil properties on the seismic response of an embankment dam.” stationary frequency content and its importance to seismic assessment
Soil. Dyn. Earthquake Eng., 64(Sep), 113–128. of structures.” Earthquake Eng. Struct. Dyn., 31(4), 993–1005.
Shields, M. D. (2010). “Simulation of stochastic processes: Applications in Yeh, C.-H., and Wen, Y. K. (1990). “Modeling of nonstationary ground
civil engineering.” Ph.D. thesis, Columbia Univ., New York. motion and analysis of inelastic structural response.” Struct. Saf., 8(1–4),
Shields, M. D., and Deodatis, G. (2013a). “A simple and efficient 281–298.
methodology to approximate a general non-Gaussian stationary sto- Zentner, I. (2013). “Simulation of non-stationary conditional ground motion
chastic vector process by a translation process with applications in wind fields in the time domain.” Georisk: Assess. Manage. Risk Eng. Syst.
velocity simulation.” Probab. Eng. Mech., 31(Jan), 19–29. Geohazards, 7(1), 37–48.
Shields, M. D., and Deodatis, G. (2013b). “Estimation of evolutionary Zerva, A. (1992). “Seismic ground motion simulations from a class of
spectra for simulation of non-stationary and non-Gaussian stochastic spatial variability models.” Earthquake Eng. Struct. Dyn., 21(4), 351–
processes.” Comput. Struct., 126, 149–163. 361.
Shields, M. D., Deodatis, G., and Bocchini, P. (2011). “A simple and Zerva, A. (1994). “On the spatial variation of seismic ground motions and
efficient methodology to approximate a general non-Gaussian sto- its effects on lifelines.” Eng. Struct., 16(7), 534–546.
chastic process by a translation process.” Probab. Eng. Mech., 26(4), Zerva, A. (2009). Spatial variation of seismic ground motions: Modeling
511–519. and engineering applications, Taylor & Francis, London.

© ASCE 04014161-9 J. Eng. Mech.

J. Eng. Mech.

You might also like