EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL.

2, 249-267 (1974)

SIMULATION OF ARTIFICIAL EARTHQUAKES

G . RODOLFO SARAGONI* AND GARY c. HART?

University of California, Los Angela, U.S.A.

SUMMARY
A procedure is developed for the simulation of artificial earthquake accelerograms. The time variation of amplitude
and frequency content is preserved in the simulation procedure. Sixteen artificial earthquake accelerograms are
simulated and compared with a target acceierogram. The time variation of amplitude and frequency content for
26 historical earthquake accelerograms is characterized.

INTRODUCTION
Earthquake engineering has progressed to the stage where it is now computationally practical and desirable
to perform a dynamic analysis of most civil engineering structures. Such an analysis requires the engineer to
create an accurate analytical model of the structure as well as prescribe an earthquake input excitation. The
design earthquake input excitation for the site under consideration is usually prescribed in the form of
response spectra or an ensemble of artificial earthquake acceleration time histories. A response spectrum
provides a convenient way of obtaining a preliminary design solution to a structural dynamics problem and is
certainly an important analysis tool. The prescribing of an input time history for use in a detailed linear
and/or non-linear analysis has ushered into earthquake engineering artificial earthquake accelerograms.
The basic concept of generating and using artificial earthquakes in seismic analyses was proposed by
Housner and Jenningsl and later by Amin and Ang.2 Their common goal was to generate an ensemble of
artificial accelerograms which possess an average ensemble response spectrum similar to that proposed by
Hou~ner.~
The simulation of artificial earthquakes based upon a stationary stochastic process assumption has been
done by Housner? Thompson,5 Bycroft? Tajimi,' Barstein! Rosenblueth and Bustamante: Housner and
Jennings: Cornell,lo Arias and Petit-Laurent?' Brady12 and Ruiz and Penzien.13 In these simulation pro-
cedures the amplitude and frequency content of the underlying stochastic process is stationary-i.e. time
invariant.
Non-stationary models in mean square amplitude have been considered by Bolotin,I* Amin and Ang?
Shinozuka and Sato,15 Jennings and co-workersla and Iyengar and 1yengar.l' These procedures involve the
use of an amplitude time modulation function and a stationary frequency content.
Non-stationary models in both amplitude and frequency content have been considered by BeaudetI8 for
the synthesis of seismograms of nuclear explosions using analogue computers. Trifunac19 has proposed a
deterministic method of simulation that includes the time variation in frequency content due to the dispersion
of surface seismic waves.
The objectives of the research described here were :
1. The development of a method for the simulation of artificial earthquake accelerograms which incor-
porates the time variation of mean square amplitude and frequency content.
2. Proposal of a method for earthquake ground motion characterization which describes the time variation
of mean square amplitude and frequency content.
* Former Graduate Student, now Professor, Departmento de Obres Chiles, Universidad de Chile.
t Assistant Professor.
Received 22 June 1973
0 1974 by John Wiley & Sons, Ltd.
249
250 G. RODOLFO SARAGONI AND GARY C. HART

The generation of artificial earthquake accelerograms which have a time variation in frequency content
is an attempt to bring the analysis process closer to reality. It has been recognized in the past and once again
observed in the 9 February 1971 San Fernando earthquake that the frequency content of ground motion
generally changes with time and is highly dependent upon epicentral distance.
The method proposed for the simulation of artificial earthquakes can be visualized as a direct extension of
previously developed time modulated non-stationary simulation procedures. A time modulating envelope
function is used to control the process amplitude level. The time variation of frequency content is accom-
plished by modelling the underlying stochastic process as a segmented nonstationary stochastic process.20
In such a model the time axis is subdivided into contiguous time regions each having a unique, but stationary,
power spectral density function. Prior simulation procedures can be visualized as having only one time region
and therefore only one power spectral density function. It is emphasized that the number of time regions is
arbitrary, but in the present research three time regions are used.
The establishment of numerical values of the amplitude and frequency characterization parameters for
simulating artificial earthquakes at a particular site is a difficult task. To assist in this estimation process,
this paper gives values of the characterization parameters obtained from the analyses of 26 recorded accelero-
grams. A method is proposed for establishing values of these characterization parameters where the earth-
quake's expected energy function is used to define the time modulating envelope function parameters. Then
each region's uni-modal power spectral density function is estimated using a non-stationary Poisson counting
process of zero crossings and maxima.
It is the fundamental premise of the approach presented here that the selection of the simulation para-
meters should be made by a team of scientists, and that the dissemination of standardized earthquake
accelerograms in the form of coded computer cards for general misuse is undesirable. In other words, in the
design of special structures a set of artificial earthquakes should be simulated which are the most appropriate
for that particular site. The data presented in this paper must be integrated with past professional experience
to obtain the simulation characterization parameters for any given case. Also, the scaling up or down of the
magnitude of the characterization parameters should be recognized as a necessary decision operation, and
should be based upon such considerations as
1. The type of structure being designed.
2. The social or economic significance of the structure's failure.
3. The appropriateness of past historical information for the site under consideration.

SIMULATION OF ARTIFICIAL EARTHQUAKE ACCELEROGRAMS

The ground motion amplitude variation is modelled herein by using a deterministic time modulating envelope
function. This concept is not new and has been used by various authors; see Reference 13 for a summary of
this research. The time variation in frequency is a new concept and is accomplished by dividing the time
domain into three contiguous time regions and requiring that the simulated ground motion in each region
possesses a unique uni-modal power spectral density function.
Figure 1 is a schematic diagram of the proposed artificial earthquake simulation procedure. The remainder
of this section describes in mathematical terms the necessary simulation operations.
The proposed method simulates sample functions of a stochastic earthquake process, {a(r)}, obtained by
modulating contiguous regions of filtered Gaussian white noise with a deterministic time envelope function.
Mathematically each sample function of the process (i.e. each artificial earthquake) is
3
t ) # ( t ) C A,(t)("'si(t)
( k ) ~ (=
i=l
where
(k)a(t)= kth artificial earthquake of the ensemble
(k)Si(t)= kth-filtered white noise sample function for the ith time region
A,(t) = deterministic rectangular time window
= U(t - ti-1) - V(t - ti)
 - --
SIMULATION OF ARTIFICIAL EARTHQUAKES 25 1

FILTER WHITE
NOISE SAMPLE FORM TIME SUM SEPARATE TtME MODULATE
FUNCTIONS REGIONS TIME RECORDS SAMPLE FUNCTION
‘-,-’
v v v v

M’“ ( 1 )

GENERATE
*U
--------------
.REGION 3
‘K’S,It)

FILTERED
A,lt)‘K’S,lt)

TIME UNMOOULATED
EARTHQUAKE
TIME MODULATED
ARTIFICIAL
WHITE WHITE REGION
NOISE NOISE DEFINITION RECORD EARTHOUAKE
SAMPLE ACCELEROGRAM
FUHCTIONS

Figure 1. Schematic diagram of the simulation procedure

and
#(t) = deterministic envelope or shaping function

The time variation in frequency content of {a(t)} follows from the fact that each contiguous stochastic
process {Si(t)}is allowed to possess a unique power spectral density function. Each sample function (k)Si(t)
of the stochastic process {Si(t)}is obtained by filtering sample functions %(t) of a zero mean Gaussian white
noise stochastic process, {n(t)}. The development of these component parts is now described.
Consider a white noise process {n(t)} which has a zero mean value function, that is
E(n(t)) = 0
and has an auto-correlation function
R,,(tl, t z ) = E(n(t1) n(tz)>
= R,(T) = w,8(7) (3)
where T = tz- tl, W, is a positive real constant and S(T) is the Kronecker delta. It then follows that samples
(k)Si(t)of the filtered Gaussian white noise process are given by

and where hi(r) is the impulse response function of a yet to be discussed linear and time invariant filter.
It then follows from the definition of the expected mean square acceleration and equations (1)-(4) that

and since by definition

the mean square value function can be reduced to

Therefore, using the fact that

E ( g ( t ) ) = W, .I
Pcm

-‘X
’ h:(t-~~)d~~
252 G . RODOLFO SARAGONI AND GARY C. HART

it then follows that if we normalize h , ( ~ such

) that
E(S,2(t)) = 1
we obtain
E(a2(t)) = IC12(t)
Since the ground acceleration is a stochastic process it is recognized that a recorded acceleration versus
time history represents only one possible sample function of the stochastic process. Therefore, past researchers
have found it convenient to postulate a mathematical expression for the mean square acceleration. Herein
we use the expression
E(a2(t)) = (10)
It can be shown that this expression represents exactly the expected mean square acceleration resulting from
the superposition of a large number of non-stationary wave pulses.2l It will soon be apparent that /3 is an
intensity parameter whereas 01 and y are envelope shape characterization parameters. It now follows that the
deterministic envelope function $(t) of equation (9) can be evaluated using equation (10) and is
+(t) = t @ 5e~4 5 d JS (1 1)
Each of the time regions of the ground acceleration stochastic process is assumed to have a uni-modal
power spectral density function which satisfies equation (8) and is of the analytical form
S,(w) = So/wJPe-'wlQ; -co<w<co (12)
Note that So, P and Q assume, in general, unique values for each time region, even though this is not explicitly
indicated above. Therefore, the filtered white noise processes {Si(t)}with this power spectral density function
must be simulated. Such a simulation follows directly once the parameters of the impulse response function,
hi(.), corresponding to equation (1 2), have been determined.
The function hi(7) can alternately be expressed as

where Hi(w) is the complex frequency response function relating the input and output of a time invariant
filter or system. The function H,(w) can be represented in the polar form
H,(O)= IH,(w) I e-jel(w) (14)
where I H,(w) I and $,(w) are the modulus and phase of Hi(@). Therefore, equations (13) and (14) indicate
that the determination of hi(7) can be accomplished through the intermediate determination of the parameters
I I
which describe Hi(w) and $,(w).
The modulus IH,(w)I is obtained by using the fact that the power spectral density function of {S,(t)} is

It follows directly from equations (12) and (15) that for each time region
IH , ( ~I )= J ( ~ S JI 1~/2~-051wl~ - co < <+ (16)
Since there exists no unique condition for the determination of $Xw), we may use any system or filter with a
IHi(w)l that satisfies equation (16). Hence, an arbitrary ~&(w)can be used to perform the simulation. It is
always possible to give a complete description of the filter by introducing some additional arbitrary con-
) a causal function (i.e. h&T) = 0 for t > 0),
straint. For example, one may impose the condition that h , ( ~ be
I I
which can be accomplished since Hi(w) is square-integrable and satisfies the Paley-WienerZ2condition.
However, it proves to be more convenient computationally to impose the arbitrary condition that
4kJ)= 0 (17)
If this condition is imposed then hi(.) is an even function Of T and equations (13) and (16) lead to a filter whose
impulse response function is given by
 SIMULATION OF ARTIFICIAL EARTHQUAKES 253
where

o = tan-'
and r(-)is the gamma function.
Notice that this impulse response function is completely established once So, P and Q are defined. Also
observe that the above impulse response function is not causal since h , ( ~ ) #0 for T < 0. Non-causal filters are
not physically realizable because they imply that if the input is zero for 1 c 0 the output may in general be
non-zero for t < 0. However, the fact that a non-realizable filter is used in the artificial simulation of earth-
quakes is irrelevant. On the other hand, the use of such a filter is computationally efficient, and as we shall
see later, it produces excellent results.
This section has shown that in order to simulate artificial earthquake accelerograms as described herein,
we must select numerical values for several characterization parameters. In particular, the values of three
time modulating envelope function parameters must be selected-i.e. a, /3 and y. Then, for each contiguous
time region the numerical values of P and Q must also be chosen. Note also that the number of time regions
to be used and the time duration of each region must be specified. The following section presents methods for
the selection of numerical values of these parameters.

AMPLITUDE AND FREQUENCY CHARACTERIZATION

This section presents one method for the non-stationary amplitude and frequency characterization of an
earthquake ground motion process. The method is particularly appropriate when only one or a few earth-
quake records appropriate to the region are available. The characterization provides numerical values of the
parameters defined in the previous section, and is carried out by using: (a) an expected energy function
(b) an intensity function of zero crossings and (c) an intensity function of maxima.
The expected energy function of the ground acceleration, denoted a(t), is defined to be

Upon substitution of the mean square acceleration from equation (10) we obtain

where

is the incomplete gamma function, and r(u) is the gamma function.

Figure 2 shows the epicentrally directed radial ground motion recorded at 8244 Orion Blvd. during the
9 February 1971 San Fernando earthquake. This record will be used to demonstrate the characterization
procedures discussed in this section. Figure 3 shows a plot of the expected energy function obtained using the
ground motion shown in Figure 2. The irregular line in the figure shows a plot of the expected energy function
using a numerical integration of equation (2). The smooth curve represents a least square best fit obtained
with a,/?and y as parameters and using the mathematical expression given in equation (21). Therefore, with
a knowledge of a, and y it is possible, using equation (lo), to plot the expected mean square acceleration
of the process, as shown in Figure 4. For future reference, Table I gives the values of a, /3 and y obtained
using the above procedure for processes corresponding to several past earthquake accelerograms. The reader
is referred to Reference 21 for plots similar to Figure 3 for all of the records noted in Table I.
Since the mean square acceleration function statistically characterizes only the amplitude characteristics of
an earthquake accelerogram, it is necessary to introduce other functions which characterize the time variation
of frequency content. The characterization of the frequency content in each contiguous time region is
achieved by considering the statistics of zero and maxima crossings plus a uni-modal power spectral density
function.
254 G. RODOLFO SARAGONI AND GARY C. HART

0 5 10 15 20 25 30
TIME (SEC)
Figure 2. Radial ground motion at 8244 Orion during the 9 February 1971 earthquake
Table I. Earthquake amplitude characterization
Expected
Time (T) total energy B OL
Station Date Component duration g2sec) g2sec) y (sec-*)
El Centro, 20-1 2-1 934 N-S 25.259 3.6353 0.1851 1a272 0.286
California E-W 24.956 2-9967 0.0106 2.991 0.380
Helena, Montana 21-10-1935 N-S 9-490 0.4436 2.2585 6.6085 3.434
E-W 9.150 0.6234 6.1175 7.3443 3.972
El Centro, 18-5-1 940 N-S 53.732 12-7569 1.2848 0.125g 0.123
California E-W 53.469 9.0947 0.3448 0.755 0.148
Olympia, 13-4-1949 S1O"E 30.272 5.2705 0.0123 3.020 0-348
Washington S8O"W 30.273 8.1244 0.0034 3.804 0.361
Taft, California 21-7-1952 S69"E 54.390 4.1025 0.1909 0.6297 0.143
N21"E 54.362 3.7459 0-1047 0.900 1 0.151
Golden Gate Park, 22-3-1951 N1O"E 12500 0.2055 0-3014 1-446 1-282
San Francisco S80"W 12-530 0.3610 0-6890 2.134 1.598
Pacoima Dam 9-2-1 97 1 Radial 25.72 70.383 0.2266 5.69 1.042
Tangential 25-72 42.306 0.0259 7.09 1.176
Vertical 25.72 28-748 0.4032 4.58 0.970
8244 Orion Radial 30.00 7.769 0.01 37 3.65 0-454
Tangential 30.00 4-791 0~0055 3-76 0.436
Vertical 30.00 4.403 0.1659 1-06 0-206
15250 Ventura Radial 27.53 5.274 0.0086 3.44 0.401
Tangential 27.53 3.187 0.0055 3-04 0.326
Hollywood Radial 30.00 1.611 0.1014 1a49 0-369
Storage Building, Tangential 30.00 2-392 0.2210 1.18 0.350
Parking Lot Vertical 30.00 0.429 0.0534 0.49 0-228
1640 Marengo Radial 30.00 1*444 0.0008 4.99 0.634
Tangential 30.00 2.460 0.0027 4.76 0.657
Vertical 30.00 0.399 0.0036 2.78 0.433

The oscillations of accelerograms around their zero acceleration line are statistically measured using a
counting process {No(t)}for t 2 O . This process counts the number of zero crossings occurring up t o time t
and the zero crossings are assumed to be a non-stationary Poisson process having a mean value function,

which is a continuous, differentiable and non-decreasing function of time.

 SIMULATION OF ARTIFICIAL EARTHQUAKES 255

LEAST SOUARE F I T

-
NUMERICAL INTEGRATION

I
0 6 12 18 24 30
TIME ISECI

Figure 3. Expected energy function at 8244 Orion, 9 February 1971 earthquake-radial direction

TIME (SEC)

Figure 4. Expected mean square acceleration at 8244 Orion, 9 February 1971 earthquake-radial direction

The intensity of the number of zero crossings per unit of time is measured by the intensity function of the
process. The intensity function is defined as

It is noted that if the earthquake acceleration process is stationary in frequency content, vo(?) is a constant,
denoted vo, and therefore the mean value function is
Po(?) = vo t (25)
Considering the response record first as a whole, it is possible to postulate a form for the function vo(t) in
equation (24) and to test the hypothesis. Inspection of several plots of the function suggests an intensity
function of zero crossings
vo(t)= voe-yof (26)
where r], is the initial rate of zero crossings and yo is a positive constant which is a measure of the function’s
decay with time. Using equations (24) and (26), the following mean value function, pO(t),is obtained

Estimation of the constants T~ and ’yo are made using the least squares fit technique. Established estimates for
yo and yo are given in Table I1 for the accelerograms noted. The estimation of both parameters makes it
possible to test the hypothesis that {No(?)}is a non-stationary Poisson process with exponential decaying
intensity function. Following P a r ~ e it
n ~can
~ be shown that the original non-stationary Poisson process can
be transformed to a stationary Poisson process by the use of an appropriate change of time scale. Then a
method exists for testing if a process is of the stationary Poisson type at a 95 per cent significance level.
This test was applied to the 26 earthquake records noted in Table I1 and the result of the test was always
256 G . RODOLFO SARAGONI A N D GARY C. HART

Table 11. Zero crossing and maxima results

Station Date
El Centro, 30-12-1934 N-S 169 9.282 0.0259 132 7.870 0.0366
California E-W 175 9.577 0.0278 133 7.309 0,0276
Helena, Montana 3 1-10-1935 N-S 131 15.06 0.0202 79 9.393 0.0239
E-W 108 11.82 0*0000 78 8.751 04035
El Centro, 18-5-1 940 N-S 346 9.224 0.0139 257 6.892 0.01 54
California E-W 331 8.498 0.0112 278 6.888 0.0124
Olympia, 13-4-1 949 S1O"E 299 16.830 0.041 1 194 9.5 1 0.0275
Washington S8O"W 283 17.050 0.0466 190 10-67 0.0397
Taft, California 21-7-1 952 N21"E 380 1 1.627 0.0179 321 7.771 0.0105
S69"E 352 10.110 0.0179 307 7.235 0.0108
Golden Gate Park, 22-3-1 957 N1O"E 181 18.13 0.0359 116 10.927 0.0242
San Francisco S80"W 199 16.20 o*oooo 120 9.943 0.0075
Pacoima Dam 9-2-1 971 Radial 299 14.990 0.0242 292 13.28 1 0.0125
Tangential 349 13.355 0~0000 325 12.462 0~0000
Vertical 405 16.798 0.0077 305 11.555 0~0000
8244 Orion Radial 299 12.967 0.0737 230 9-113 0.0177
Tangential 154 12.840 0.0774 229 9.361 0.0193
Vertical 175 11.710 0.0477 223 9.157 0.0154
15250 Ventura Radial 180 14.035 0.0679 186 9.761 0.0303
Tangential 179 9562 0.0226 161 7.393 0.0168
Hollywood Radial 204 11.135 0.0406 196 8.095 0.0158
Storage Building, Tangential 137 7.602 0.0452 183 8.413 0.0250
Parking Lot Vertical 245 16.143 0.0580 298 124452 0.0167
1640 Marengo Radial 227 12.610 0.0397 223 10556 0.0255
Tangential 226 9.700 0.0204 275 10.641 0.0274
Vertical 301 20.2 15 0.0572 276 15.614 0.0413
~

satisfactory. This implies that the original zero crossing processes can be assumed to be a non-stationary
Poisson process with exponential decaying intensity function.
From the results given in Table I1 some additional observations and interpretations can be drawn. Some
of the records from the 1971 San Fernando earthquake, e.g. 8244 Orion and 15250 Ventura radial, yield very
large values of yo. The Olympia accelerograms yielded the largest values of yo for any pre-1971 San Fernando
event. Other records such as Helena E-W, Golden Gate S8O"W and Pacoima tangential yield practically zero
values for yo. Also, it is noted that some of the records obtained at the same station exhibited for different
directions quite different values of the parameters. In the case of the 1971 earthquake the values of r), for
vertical components were the largest and the radial components yielded larger values of yo than the tangential
components.
One way to compare the variation of zero crossing intensity with time for different processes is to plot the
decay factor (i.e. e-yol). Such plots are shown in Figure 5. From this figure it can be generally concluded that
the records of the 1971 earthquake and Olympia 1949 exhibit the largest variation of zero crossing intensity.
The counting process of zero crossings is one parameter used to characterize the time dependent spectral
content of the earthquake acceleration process. When we consider the non-stationary earthquake ground
acceleration process to be sectionally stationary (i.e. having contiguous time segments which have a unique
but stationary power spectral density) then the expected number of zero crossing per unit of time in the ith
time segment defined by ti-l < t < ti is a constant denoted ~ ~ ( ti)
t ~and
- ~is, given by expression (24)

where S(w) is the power spectral density function of the ground acceleration in this region. The constant
<
ti) is the temporal average of vo(t)in the time segment defined by the interval (ti 5 t ti+l), i.e.
vO(tiF1,

J ti-1 1
 SIMULATION OF ARTIFICIAL EARTHQUAKES 257

Figure 5(a). Normalized intensity functions of zero crossings pre-1971 earthquakes

r
PACOIMA TANGENTIAL
PACOlMA RADIAL I

I I I I
0 10 20 30
TIME . SECS
Figure 5(b). Normalized intensity functions of zero crossings 1971 San Fernando earthquake-horizontal components

Since v0(ri+ ti) in general changes from segment to segment it is evident that S(w) will be a function of
time. Note that equation (26) is by itself insufficient to characterize S(w) of a given time segment; therefore,
in order to complete the characterization of the time variation of the frequency content, we now consider the
statistics of the maxima.
The statistics of maxima are characterized by considering the stochastic process {N,(t)} which counts the
number of maxima occurring in an interval of time. The counting process of maxima is also assumed to be a
non-stationary Poisson process and to possess the same basic considerations as the counting process of
zero crossings. Therefore, the counting process of maxima has a mean value function, pm(t)and an intensity
function, v,(t).
258 G . RODOLFO SARAGONI AND GARY C. HART

0 10 20 30
-
TIME SECS.
Figure 5(c). Normalized intensity functions of zero 1971 San Fernando earthquake-Vertical components

Inspection of 44 mean value function plots again suggests a decaying character of p,(t) and an intensity
function V m ( t ) of the exponential form
Y m ( t ) = 7, e-yJ (30)
where q,, is the initial rate of occurrence of maxima, and y m is a positive constant which is a measure of the
function decay with time. The mean value function of the process is

pm(r = E(N,(t)) = %(1 - e-72) (31)

Ym
Estimates of rlrn and ym were made again using a least square fit technique and the values obtained appear
in Table 11. The same test as was used for the zero crossing processes to verify that the process was non-
stationary was applied and the results were found to be equally satisfactory. From the results given in
Table I1 the following observations can be made:
1. Marengo vertical, Olympia and El Centro 1934 yielded the largest values of y,.
2. Some samples such as Pacoima tangential and vertical have practically a zero value for ym.
3. 15250 Ventura shows large differences for 7, and ym between the different components.
A comparison between various normalized intensity functions of maxima (i.e. e-y.J> for the earthquakes
analysed appears in Figure 6. This comparison reveals that normalized intensity functions of maxima for
records of pre-1971 earthquakes and the 1971 earthquake are not significantly different.
Now consider, as was done for zero crossings, that the expected number of maxima per unit time in the
ith stationary time segment defined by t i - , < t < t i is a constant, V m ( r i - , , f i ) and it is given by expression
(24), i.e.

It is noted that equation (32) only has statistical significance when Y , ( t ) is a slowly varying function of time
in the time segment (ti-1 < t < ti) and therefore practical estimation limitations are placed upon the time
duration and numbers of time regions for which the frequency content may vary.
The characterization of S(w) for each time region and variation with time is now possible using equations
(26) and (28) and by idealizing each segment's power spectral density function as a uni-modal function of the
mathematical form
S(O) = So I w 1' e-lw"J (33)
 SIMULATION OF ARTIFICIAL EARTHQUAKES 259

2
1
HELENA E.W
HELENA N-S
GOLDEN GATE S 80"E
GOLDEN GATE N 10'E
TAFT S 69"E
TAFT N 21"E

.5 -
EL CENTRO 1934 N S EL CENTRO 1940 N S
EL CENTRO 1934 E-W EL CENTRO 1940 E.W

OLYMPIA S 10"E
OLYMPIA S 80"W

0
0 10 20 30 40 50 60
TIME -SECS

Figure 6(a). Normalized intensity functions of maxima pre-1971 earthquakes

VENTURA TANGENTIAL

HOLLYWOOD RADIAL
'
17
HOLLYWOOD TANGENTIAL

MARENGO RADIAL
MARENGO TANGENTIAL

0 '
I I

where So, P and Q are positive real constants which in general are different for each time segment. The
functional form of S(w) idealizes the power spectral density functions of earthquake acceleration processes
and has some physical justifications.26The previously discussed condition
Jo*S(w)dw = 1 (34)

places the following constraint on So

Qp+1
so=qFxj (35)
260 G. RODOLFO SARAGONI AND GARY C. HART

Figure 6(c). Normalized intensity functions of maxima 1971 San Fernando earthquake-vertical components

The scaling of the power spectral density in each region to have unit area enables us to separate the time
variation in acceleration amplitude and frequency content.
Inherent in the use of equations (28) and (32) is the necessity to define the beginning and ending times for
each time region. This may be accomplished by:
1. Inspection of an existing record if the wish is to generate artificial earthquakes corresponding to a
historical record.
2. Estimation of the arrival times of seismic waves at the site under study.
3. Using the zero crossing or maxima intensity functions.
Herein we used the latter method to divide the time axis into three regions. Figure 7 shows a normalized
intensity function of zero crossings for the 8244 Orion Radial accelerogram. The time regions were defined

4.72 12.22 30.00

\ " h
Y
,TIME-(SEC)
V
I
REGION I REGION II REGION Ill

Figure 7. Division of time axis into stationary regions

 SIMULATION OF ARTIFICIAL EARTHQUAKES 26 I

by subdividing the intensity axis into three parts as shown in the figure. Table I11 summarizes the time regions
so established and the corresponding power spectral density shape parameters.

Table 111. Summary of frequency description parameters

~~~

Station Component tl f2 f3 Qi p2 Q'2 p3 Q,

El Centro, N-S 6.73 14.90 25.26 1-699 0.1180 2.449 0.1771 3.916 0-3101
California E-W 6.56 14.60 24.96 2.347 0.1387 2.331 0.1691 2.311 0.2169
Helena. Montana N-S 2.97 6.12 9.49 6-838 0.1812 7.348 0.2046 0.2337 0.2337
E-W 3.02 6-07 9.15 2.778 0.1144 2.921 0.1182 0.1224 0.1 224
El Centro, N-S 13.87 31-08 53.73 2.72 1 0.1590 3.035 0.2121 3-498 0.3082
California E-W 14.53 31.89 53.47 1.853 0.1345 2.024 0.1692 2.252 0-2241
Olympia, S1O"E 6.59 15-66 30.27 9.636 0.2402 4.795 0.1918 2.316 0.1811
Washington S8O"W 6.23 15-05 30.27 5-736 0.1553 4.345 0.1 772 2.942 0-2320
Taft. California N21"E 12.99 29-94 54.39 2.745 0.1364 1.651 0.1314 0.7692 0.1348
S69"E 12.98 29.93 54.36 2.578 0.1428 1.583 0.1401 0.7571 0.1462
Golden Gate Park, N1O"E 3.58 7-68 12.50 7.205 0.1626 5.504 0.1500 4-1635 0.1421
San Francisco S80"W 4-05 8.22 12.53 8.440 0.1939 10.626 0.2366 14.054 0.3036
Pacoima Dam Radial 7.80 17.43 30.00 0-769 0.0533 0.3849 0.0540 0.0353 0.0564
Tangential 30 0.890 0.0557
Vertical 9.251 19.21 30.00 3-247 0.0927 2.295 0.0795 1.5851 0.0697
8244 Orion Radial 4.78 12-22 30.00 1.899 0.0979 0.2252 0-0749 0-1870 0.3045
Tangential 4.62 11-88 30.00 1.599 0.0902 0.1 208 0-0714 0.41 19 0.2737
Vertical 6.13 14.84 30.00 1.343 0,0877 02974 00768 1.7628 0.91 15
15250 Ventura Radial 4.88 12.22 27.53 2.407 0.1032 0.7998 0*0900 1.5264 0.4062
Tangential 6.94 15.69 27-53 2.804 0.1434 1-764 0.1364 0.9294 0.1361
Hollywood Radial 6.59 15.62 30.00 2.044 0.1143 0.7921 0.0998 1.6548 0.2098
Storage Building, Tangential 6.29 15.11 30.00 0-093 0.0728 0.0642 0.1957 0.9430 0.2914
Parking Lot Vertical 5.54 13.75 30.00 1.337 0.0644 0.1743 0.0546 4.2301 0.7950
1640 Morengo Radial 6.65 15-71 30.00 1.231 0.0771 0.6794 0.0830 0.1515 0.0971
Tangential 8-12 17.85 30.00 0.363 0.0639 0.5790 0.0862 0.9073 0.1156
Vertical 5.58 13-84 30.00 1.831 0.0606 1.0922 0.0692 0.3369 0.0941

A plot of the normalized power spectral density function for each region of the 8244 Orion Radial motion
is shown in Figure 8. It is apparent from this figure that the dominant frequency content of the record is
centred about lower frequencies for later time regions. Figure 9 shows the time variation in power spectral
density function for four historically significant and commonly used earthquake ground motions. Figures
showing the power spectral density functions for all the earthquakes noted in this paper are given in
Reference 21.

4.78-12.22 SECS

P
10 20 30 40 50 60
FREQUENCY (RADlANS/SECl

Figure 8. Time-frequency content: 8244 Orion Radial 9 February 1971

 NORMALIZED
POWER SPECTRAL DENSITY
(X102J
- . - a
0 h ) P m ) O ) o N
NORMALIZED NORMALIZED
POWER SPECTRAL DENSITY POWER SPECTRAL DENSITY
(Xl0.2) (X152)
m
iD
w
I
e
w
Q)
VI
rn
t:
Z9Z
 NORMALIZED NORMALIZED
POWER SPECTRAL DENSITY POWER SPECTRAL DENSITY
3! (Xlo-2) (Xlo-2)
N 0 R MA L I ZED NOR M A L I 2 E0
POWER SPECTRAL DENSITY POWER SPECTRAL DENSITY
I I I
N N
E9Z
264 G . RODOLFO SARAGONI A N D GARY C. HART

EXAMPLE SIMULATION
This section demonstrates the mechanics of the non-stationary earthquake simulation procedure described in
the previous section. In particular, it demonstrates how one may generate an ensemble of artificial earthquake
accelerograms each possessing the important characteristics present in a single historical (or target) earth-
quake. The historical target accelerogram considered in this section is the radial accelerogram recorded at
8244 Orion during the 1971 San Fernando earthquake.
The previously described non-stationary characterization procedure was used to define the 8244 Orion
Radial process and it follows by substituting the values of /I,y and 01 for this process that
$ ( t ) = 0.117t1.825 e-0.2771
(36)
The variation in frequency content is introduced in the artificial earthquake process (a(t)>by the considera-
tion of three time regions which each possess a unique power spectral density function. The three regions

0.4

--
a
0.2

2
0
l-
d o
W
-I
W
0
V
Q
-.2

-A
0 5 10 15 20 25
TIME (SEC)
ARTIFICIAL EARTHQUAKE #l

0.4

z
0
t-
d o
W
-J
W
0
0
a
-.2

-.4
5 10 15 20 25 30
TIME (SECI
ARTIFICIAL EARTHQUAKE #2
Figure 10. Representative artificial accelerograms
 SIMULATION OF ARTIFICIAL EARTHQUAKES 265

used in this simulation are noted in Table I11 and it therefore follows that
A,(t) = U ( t ) - U(t-4.78) (37a)
A,(t) = U(t- 4.78) - U(t - 12.22) (37b)
A,(?) = U(t- 12.22)- U(t-30.00) (37c)
Sixteen samples functions of the 8244 Orion Radial process were generated. The generation of each sample
takes approximately 3 sec of time on an IBM 360/91 computer. Figure 2 shows the actual Orion record and
Figure 10 shows two typical artificial earthquake samples. A visual comparison of the artificial accelero-
grams shows a satisfactory resemblance to the original record.
A more complete comparison between the simulated accelerograms and the original record was performed
by comparing the average values of the previously discussed non-stationary characterization functions
obtained from the artificial accelerograms with the values of the functions for the 8244 Orion Radial target
process. Figure 11 shows a comparison between the expected energy function of the target process and the
mean and one sigma levels of the artificial accelerograms. The comparison is acceptable. Figure 12 presents
10
1

0 10 20 30
TIME- (SEC)
Figure 11. Comparison of expected energy functions for artificial and target accelerograms
200
I

TlhlE-(SEC)

Figure 12. Comparison of zero crossings for simulated and target accelerograms
266 G. RODOLFO SARAGONI AND GARY C. HART

L
TARGET ACCELEROGRAM:
8244 ORION RADIAL

150 7

1 I

TIME -(SECl
Figure 13. Comparison of maxima for simulated and target accelerograms
Table IV. Non-stationary characteristics of artificial earthquakes

1 7.842 0.01 30 3-560 0,4280 136 12.18 00846 122 8.6100 0.0544
2 8.789 0.0541 3.004 0.4400 140 13.41 0-0907 126 8.4880 0.0451
3 6.090 0.0023 4.712 0.5360 130 12.36 0-0918 122 9.8710 0.0652
4 9.652 0.0014 4.996 0.5040 130 14.08 0.1056 122 9.4700 0.0601
5 7.764 0.0147 3.582 0.4460 135 14.45 0.1037 121 8.6444 0.0516
6 7.084 0.01 84 3.426 0.4440 130 11.85 0.0857 124 9.7920 0.0591
7 5-511 0.0173 3.268 0.4500 141 13.33 0.0880 126 8.6610 0.0407
8 8-222 0.0132 3.682 0.4520 121 13.52 0-1099 137 9.5140 0.0481
9 6.664 0.0127 3.452 0.41 80 129 13.50 0.1015 118 7.9910 0.0400
10 7.399 0.0115 3-894 0.4940 144 15.67 0.1045 148 10.6250 0.0524
11 7.789 0.0140 3.924 0.5160 130 11.58 0.0812 117 10.0300 0.0656
12 9.665 0.0115 3.934 0.4760 120 12.31 0.0997 117 11.0800 0.0844
13 8.133 04086 3.884 0.4540 127 13-83 0.1052 128 9.3890 0.0507
14 8.673 0.0065 4.560 0.5680 137 15.22 0.1084 127 9.1800 0.0578
15 7.633 0.0056 4-302 0.5100 140 12.82 0.0852 134 11.1280 0.0682
16 9.999 0.0063 4.350 0.5060 127 13.64 0.1029 127 8.9770 0.0442
Mean 7.931 0.0132 3.908 0.4776 132.31 13-36 0.0968 126 9.4656 0.0555
U 1.216 0.01 16 0.533 0.0414 6.81 1.12 0.0094 7.8 0.8550 0.01 13
6.7 16 0.0248 4.441 0.5 190 139.12 14.48 0.1062 133.8 10.3206 0.0668
(a)
(b) 9.147 04016 3.375 0.4362 125.50 12-24 0.0874 118.2 8.6106 0.0442
Target 7.769 0.01 37 3.650 0,4540 135 14.822 0.1068 129 12.7800 0.0956
(a), (b) population la levels.
a comparison of the occurrence of zero crossings for the target accelerogram and the mean value of the
artificial accelerograms. A similar comparison between the maxima occurrences is shown in Figure 13.
The method of non-stationary characterization discussed in the previous section was applied to each
artificially generated earthquake record. Table IV summarizes the results.
It is felt that the similarity between the target process and the simulated processes is acceptable.
CONCLUSIONS
A procedure has been presented and mathematical equations derived for the simulation of artificial earth-
quake motions which possess a time variation in mean square amplitude and frequency content. The pro-
cedure has been successfully demonstrated using a target earthquake and then generating 16 artificial earth-
quake accelerograms which resemble that target earthquake.
 SIMULATION OF ARTIFICIAL EARTHQUAKES 267

The time variation in frequency content and mean square amplitude has been characterized for 26 historical
earthquake accelerograms. This study indicated an exponential decay with time in zero crossings and maxima
and a frequency content which tended toward lower frequencies as time evolved.

ACKNOWLEDGEMENTS

The authors wish to acknowledge the financial support provided to Mr. Saragoni by the University of
California-University of Chile Co-operative Program. Also, financial assistance for this research was pro-
vided by the National Science Foundation under contract NSF-GI-29932.
Professors C. M. Duke and R. B. Matthiesen, the latter now with the Seismological Field Survey, NOAA,
made many helpful suggestions and served on the dissertation committee of Mr. Saragoni. The authors also
wish to acknowledge the sometimes casual, but helpful and honest, conversations over the last few years with
J. R. Benjamin, C. A. Cornell, G. Housner, P. Jennings, M. Nelson, J. Penzien, M. Shinozuka and J. H.
Wiggins, Jr. These discussions were invaluable in combining to form the basic concepts expressed in this paper.

REFERENCES
1. G.W. Housner and P. C. Jennings, ‘Generation of artificial earthquakes’, J. Engng Mech. Div., ASCE, 90,113-150 (1964).
2. M. Amin and A. H.-S. Ang, ‘Nonstationary stochastic model of earthquake motions’, J. Engng Mech. Div., ASCE,
94, 559-583 (1968).
3. G.W.Housner, ‘Behavior of structures during earthquakes’, J. Engng Mech. Div., ASCE, 85, 109-129 (1959).
4. G.W.Housner, ‘Properties of strong ground motion’, Bull. Seis. SOC.Am. 45, 197-218 (1955).
5. W.T.Thomson, ‘Spectral aspect of earthquakes’, Bull. Seis. SOC.Am. 49, 91-98 (1959).
6. G. N. Bycroft, ‘White noise representation of earthquakes’, J. Engng Mech. Div., ASCE, 86, 1-16 (1960).
7. H. Tajimi, ‘A statistical method of determining the maximum response of a building structure during an earthquake’,
Proc. 2nd Wld Conf. Earthquake Engng, 2, 781-797, Science Council of Japan, Japan, 1960.
8. M. F. Barstein, ‘The application of probability methods for design-the effect of seismic forces on engineering struc-
tures’, Proc. 2nd Wld Conf. Earthquake Engng, 2, 1467-1482, Science Council of Japan, Japan, 1960.
9. E. Rosenblueth and J. Bustamante, ‘Distribution of structural response to earthquakes’, J. Engng Mech. Div., ASCE,
88, 75-106 (1962).
10. C. A. Cornell, ‘Stochasticprocess models in structural engineering’, Department of Civil Engineering, Stanford University,
Stanford, California, Technical Report No. 34, 1960.
11. A. Arias and L. Petit-Laurent, ‘Un modelo teorico para 10s accelerogramasde Temblores Puertes’, Revista del Z.D.Z.E.M.,
Universidad de Chile, 4, No. 1, Santiago, Chile, 1965.
12. A. G. Brady, ‘Studies of responses to earthquake ground motions’, Earthquake Engineering Research Laboratory,
California Institute of Technology, Pasadena, California, 1966.
13. P. Ruiz and J. Penzien, ‘Stochastic seismic response of structures’, J. Engng Mech. Div., ASCE, 97,441-456 (1971).
14. V. V. Bolotin, ‘Statistical theory of the aseismic design of structures’, Proc. 2nd Wld Conf. Earthquake Engng, 2,
1365-1374, Science Council of Japan, Japan, 1960.
15. M. Shinozuka and Y. Sato, ‘Simulation of non-stationary random process’, J. Engng Mech. Div., ASCE, 93,1140(1967).
16. P. C. Jennings, G. W. Housner and C. Tsai, ‘Simulated earthquake motions for design purposes’, Proc. 4th Wld Conf.
Earthquake Engng, 1. A-1, 145-160,Chilean Association on Seismology and Earthquake Engineering, Chile, 1969.
17. R. N.Iyengar and K. T. Iyengar, ‘A nonstationary random process model for earthquake accelerations’, Bull. Seis. SOC.
Am. 59, 1163-1188 (1969).
18. P. R. Beaudet, ‘Synthesis of nonstationary seismic Signals’, Bull. Seis. SOC.Am. 60, 1615-1624 (1970).
19. M. D. Trifunac, ‘A method for synthesizing realistic strong ground motion’, Bull. Seis. SOC.Am. 61, 1739-1753 (1971).
20. R. E. Holman and G. C. Hart, ‘Structural response to segmented nonstationary random excitation’, J. AZAA 10, 1473-
1478 (1972).
21. G.R. Saragoni and G. C. Hart, ‘Nonstationary analysis and simulation of earthquake ground motions’, UCLA Technical
Report, No. UCLA-Eng-7238, 1972.
22. A. Papoulis, The Fourier Integral and Its Applications, New York, McGraw-Hill, 1962.
23. E. Parzen, Stochastic Processes, San Francisco, Holden Day, 1962.
24. S. 0.Rice, ‘Mathematical analysis of random noise’, in Selected Papers on Noise and Stochastic Processes (Ed. N. Wax),
Dover Publications, New York, 1954, 133-204.
25. A. K.Mal, ‘Rayleigh waves from a moving fault’, 67th Annual Meeting of the Seismological Society of America, 1972,
Honolulu, Hawaii.