You are on page 1of 14

Engineering Structures 78 (2014) 316

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

New insights in the analysis of the structural response


to response-spectrum-compatible accelerograms
Pierfrancesco Cacciola a,, Laura DAmico a, Irmela Zentner b
a
b

School of Environment and Technology, University of Brighton, Lewes Road, BN24GJ Brighton, UK
LaMSID UMR EDF-CNRS-CEA, Clamart, France

a r t i c l e

i n f o

Article history:
Available online 22 August 2014
Keywords:
Ground motion simulation
Quasi-stationary
Non-stationary
Imposed-variability
Response-spectrum-compatible
Code provisions

a b s t r a c t
This paper addresses the study of the seismic response of structures to response-spectrum-compatible
accelerograms. The number of methodologies proposed in the last three decades to simulate articial
earthquake ground motion testies the relevance of this subject in the scientic community. However,
the implications of the selection of models and hypothesis adopted and their impact on the structural
response have not been thoroughly highlighted yet. This contribution shows for the rst time that different ground motion models, having identical response spectrum at 5% damping, peak ground acceleration,
strong motion phase and total duration, can lead to signicant discrepancies in the structural responses
even for proportionally damped linear behaving structures, although all the models are satisfying the
response-spectrum compatibility criteria. The results show clearly the weakness in the current
response-spectrum-compatible criteria provided by seismic codes and the necessity of more robust conditions for the simulation of articial earthquake ground motions.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
The design response-spectrum is up to now the basic representation of the seismic action and the most widespread analysis tool
used by practitioners. Time-history representation of the seismic
action is also allowed by seismic codes and it is employed for a
broad number of engineering applications such as the analysis of
non-linear behaving structures whereas the response-spectrum
technique might not provide accurate results. Furthermore,
time-history analysis provides additional information on damage
mechanism and dissipation of energy due to cyclic loading that
cannot be predicted through the response spectrum analysis. It is
noted that the international seismic codes recommend only the
response-spectrum-compatible criteria that have to be satised
and do not give a method for generating the earthquake timehistories. As a consequence, several methods have been proposed
in literature coping with the generation of response-spectrumcompatible accelerograms. Earlier contributions on this subject
can be found in the review papers by Ahmadi [1] and by Cacciola
[4]. Most common approaches rely on modelling the seismic action
as a realization of a stationary or quasi-stationary stochastic
Gaussian process (see e.g. [24,2]. Even if quasi-stationary Gaussian
Corresponding author. Tel.: +44 (0)1273 642277.
E-mail address: p.cacciola@brighton.ac.uk (P. Cacciola).
http://dx.doi.org/10.1016/j.engstruct.2014.07.015
0141-0296/ 2014 Elsevier Ltd. All rights reserved.

models reliably represent the inherent random nature of seismic


action, they suffer the major drawback of neglecting the nonstationary characteristics of the real records. In this regard, in the
framework of Gaussian stochastic models, very few procedures
have been proposed in literature [23,17,12,3,5] to simulate fully
non-stationary spectrum compatible stochastic processes (i.e. with
amplitude and frequency variation). The above methodologies
possess the common drawback that the simulated time-histories
do not manifest the variability observed for real earthquakes. This
is mainly due to the fact that recorded accelerograms, even if
possessing the same magnitude and epicentral distance, have
strongly different energy distributions leading to a large variability
of ground motion parameters. On the other hand, the simulated
accelerograms are generally determined from a single power spectral density function, as a consequence the ground motion timehistories possess very similar joint timefrequency distribution.
Recently, Cacciola and Zentner [6] proposed a procedure to simulate response-spectrum-compatible accelerograms with ground
motion parameters similar to those observed in nature. To this
aim the evolutionary power spectral density function with random
coefcients and compatible with given target response spectrum
(i.e. spectrum-compatible) has been introduced. This procedure
has been extended by DAmico et al. [7] and Zentner et al. [25] to
include enhanced variability models accounting of the correlation
of spectral accelerations.

P. Cacciola et al. / Engineering Structures 78 (2014) 316

The number of methodologies proposed up to now in the last


three decades to simulate articial response spectrum compatible
accelerograms in addition to the alternative simulation techniques
compatible with database of earthquakes (see e.g. [26] and/or site
characteristics (see e.g. [19] testies the interest of the scientic
community in this topic. Nevertheless, to the best knowledge of
the authors a thorough comparative study of the structural response
to response-spectrum-compatible earthquakes aimed to provide a
depth understanding of the hypothesis adopted and the implications of the selected ground motion model has not been carried
out up to now.
Therefore, this paper aims to bridge this gap of knowledge by
comparing for the rst time through a Monte Carlo study the
response of structures to the widely used Gaussian quasi-stationary and non-stationary models along with the most novel nonstationary model with imposed variability. All the simulated
time-histories fulll the Eurocode 8 provisions and all of them possess also the same peak ground acceleration (PGA), strong-motionphase (SMP) and total duration. Three benchmark structures have
been selected for the current study, namely the linear SMART 2008
nuclear building [14], IASC 1996 5-story building linear steel frame
[10,20], and linear and non-linear SAC 1996 20-story building
structure [15]. The structures have been selected to consider a
spread scenario of high, medium and low fundamental frequencies.
2. Simulation of ground motion gaussian time-histories
In the framework of seismic engineering it is well known that
only a probabilistic approach can afford a rigorous representation
of earthquake ground motion. Accordingly, a ground motion accel g t recorded at a given location can be seen as a sample
eration u
of a stochastic process. A number of stochastic models have been
proposed in literature, for an in depth discussion the readers could
refer to the review paper by Shinozuka and Deodatis [22]. Mainly,
these models can be categorized as Gaussian or non-Gaussian models. Due to their relative simplicity the Gaussian models are the most
used to represent the seismic action. By assuming the ground
motion as a realization of Gaussian stochastic process it is fully
dened by the knowledge of its autocorrelation function Rug t1 ; t2
or by the generally known evolutionary power spectral density function Sug x; t which representation can be alternatively given by the
one-sided power spectral density by the following equation [18]

Gug x; t 2Sug x; t jax; tj2 Gx; x P 0;


Gug x; t 0; x < 0:

where a(x, t) is the frequency dependent modulating function while


G(x) is the one-sided power spectral density function of the station g t. The processes dened by the evolutionary
ary counterpart of u
power spectral density function in Eq. (1) are able to represent the
common feature of the ground motion acceleration to possess both
amplitude and frequency varying with respect to time (temporal and
spectral non-stationarities), generally known as fully non-stationary
(or non-separable). In the case in which only the amplitude of the
process varies with respect to time, (i.e. a(x, t) = a(t), temporal
non-stationarity), the process is generally known as quasi-stationary
(or separable or also uniformly modulated). Accordingly, Eq. (1) for
quasi-stationary processes modies as follows
2

Gug x; t 2Sug x; t jatj Gx; x P 0;


Gug x; t 0; x < 0:

Finally in the particular case in which the modulating function


a(t) = 1 the process is called stationary. Once the power spectral
density Gug x; t is evaluated, it is possible to simulate the r-th
sample of ground acceleration process via the superposition of Na
harmonics with random phases. That is [21]

gr t
u

Na q
X
r
2Gug iDx; tDx cosiDxt ui

i1
r

where ui are independent random phases uniformly distributed in


the interval [0, 2p).
For simulating articial accelerograms according to the spectral
representation method, the denition of the power spectral density function is the crucial point necessary to assess structural systems through a proper Monte Carlo simulation approach.
3. Quasi-stationary models
When articial ground motions time-histories are used in the
engineering practice, the mean response-spectrum determined
from the simulated time histories has to match the target
response-spectrum provided by the code over a xed frequency
range and within a specied tolerance (i.e. the articial ground
motion time-histories have to be response-spectrum-compatible).
Vanmarcke and Gasparini [24] pointed-out the fundamental relationship between the response-spectrum and the ground motion
power spectral density of the input via the so-called rst passage
problem, specically by assuming the ground-acceleration process as zero-mean Gaussian stationary process. The pseudo-acceleration response-spectrum, RSA(x0, f0), for a given damping ratio
f0 and natural circular frequency x0, can be related to the median
value of largest peak of the response of a single degree of freedom
system by means of the following expression [24,8] as follows

p
RSAx0 ; f0 x20 gU T s ; p 0:5; k0;U ; k1;U ; k2;U k0;U

where gU is the peak factor, Ts is the time observing window, p is the


not-exceeding probability and ki;U i 0; 1; 2 are the response spectral moments dened as

ki;U

xi jHxj2 Gxdx;


1
2
in which jHxj2 x20  x2 4f20 x20 x2
is the energy transfer function of the single degree of freedom system. Based on this
relationship various procedures have been proposed in literature
for determining the response-spectrum-compatible power spectral
density function. A handy recursive expression determining the
power spectral density compatible with a given response-spectrum
has been proposed by Cacciola et al. [2]. Specically,

Gxi 0;

80  x  xa

4f0
Gxi x p4f
0x
i

RSAxi ;f0 2
g 2U xi ;f0

i1

 Dx

!
i1
X
Gxk ;

8x > xa

k1

 U is the peak factor approximately determined according to


where g
the hypothesis of a barrier out-crossing in clumps and spectral
moments determined assuming that the input PSD possesses a
smooth shape and f0  1:




q
1:2
g U xi ; f0 2ln 2NU 1  exp dU pln2NU

With
1

NU 2Tps xi  ln p ;
"

dU 1 

1
1f20

f0

1  p arctan p2
2

2 #1=2

1f0

Moreover, in Eq. (6), xa 1rad sec, is the lowest bound of the


 U . The accuracy of Eq. (6) is generally satisexistence domain of g
factory, however the match can be improved applying iteratively
the following scheme

P. Cacciola et al. / Engineering Structures 78 (2014) 316

G1 x Gx;
j

j1

G x G

RSAx;f0 2
RSAj1 x;f0 2

where RSA(j) is the approximate pseudo-acceleration spectrum


determined at the j-th iteration. Once the spectrum-compatible
power spectral density G(x) is evaluated to simulate samples of
spectrum-compatible ground acceleration via Eq. (3), a proper modulating function has to be dened. To this aim the modulating function proposed by Jenning et al. [13] is generally used

aJ t

8  2
>
t
>
< t1
1
>
>
:
exp bt  t2 

t < t1
t1 6 t 6 t2

10

t > t2

as it possess an adjustable stationary part easy to control for the


generation of ground motion accelerograms. The values of the
parameters t1, t2 and b are evaluated by imposing that the energy
of the stochastic ground motion reaches the values of the 5% and
95% respectively in t1 and t2. Namely by the use of the generalized
Husid function extended for stochastic processes as proposed by
Cacciola and Deodatis [5]

Rt R1

Ht R 0tf R01
0

a2 tGxdt
a2 tGxdt

11

and imposing that H(t1) = 0.05 and H(t2) = 0.95, Eq. (11) leads to the
following analytical values of the Jennings parameters as function
of the duration of the stationary part Ts, that is,

9
2:5
11:5
; t1
; t2
Ts
b
b

12

Therefore, once T s is set (e.g. using the minimum value from


Eurocode 8 equal to 10 s) the other parameters are readily
determined.
4. Non-stationary models
By assuming the ground-acceleration process as zero-mean
Gaussian non-stationary and non-separable process the evaluation
of the evolutionary power spectral density in general cannot be
determined through Eq. (4) as it has been derived under the
hypothesis of stationary process. Hence, in literature alternative
approaches are usually preferred. Based on a statistical approach,
Spanos and Vargas Loli [23] proposed a method for generating
non-stationary spectrum-compatible earthquakes. The authors
adopted the following mathematical model of the evolutionary
power spectrum
N
X
Gug x; t
C k t2 expak tGk x

13

k1

where Gk(x) is the unilateral KanaiTajimi spectrum and the


parameters N, Ck and ak as well as the lter parameters are determined through the LevenbergMarquardt procedure in order to
minimize the difference between a target velocity response-spectrum and the approximated one, which is determined through a
simplied stochastic analysis. Due to the number of approximation
involved, the determined accelerations require an iterative procedure to make them spectrum-compatible. The procedure originally
established in Spanos and Vargas Loli [23] has been modied by
Giaralis and Spanos [12] by means of harmonic wavelets transform
to iteratively improve the matching between the target and the
simulated response spectra. Generation of non-separable articial
earthquake accelerograms has been also proposed by Preumont
[17]. The method assumes an empirical model of the evolutionary
power spectral density function possessing the feature that high

frequency component are magnied in the early part of the process


and the iterative correction of the simulated accelerograms. Specifically, the following frequency dependent modulating function, (see
Eq. (1)), is assumed
2
jax; tj t c eaxt

14

with

ax a0 a1 x a2 x2

15

where c, a0, a1 and a2 are adjustable parameters. The matching of


the mean simulated and target response spectra is ensured by the
stationary power spectrum G(x). This is determined by equating
for each frequency the energy of a separable spectrum compatible
model and that of the non-separable process dened in Eq. (1). That
is

Gx

2
jax; tj dt GS x

a2 tdt

16

in which GS(x) is the stationary power spectral density compatible


with a target spectrum and a(t) is the modulating function. The
quasi-stationary model dened in the previous section (e.g. [2]
can be suitably used for this purpose.
Once dened GS(x) and a(t) Eq. (16) provides the direct relationship for determining G(x), that is

R1

a2 tdt
Gx R 1 0
GS x
2
j
a
x
;
t
j
dt
0

17

The procedure proposed by Preumont has been recently


extended by Cacciola and Zentner [6] to deal with alternative frequency dependent modulating functions.
An alternative approach has been proposed by Cacciola [3] that
allows the simulation of fully non-stationary spectrum-compatible
accelerograms as well as the straightforward evaluation of the pertinent non-separable power spectral density function. In the model
proposed by Cacciola [3] it is assumed that the time-histories are
modeled by the superposition of two independent contributions:
the rst one is a fully non-stationary counterpart assumed to be
known and modeled by a record or alternatively by a simulated
earthquake, taking into account the time variability of both intensity and frequency content; the second one is a corrective term
represented by a quasi-stationary zero-mean Gaussian process
that adjusts the response-spectrum of the non-stationary signal
in order to make it spectrum-compatible. That is, the ground
motion can be split in two contributions:

g t u
Rg t u
Sg t;
u

18

Rg t is the non-stationary signal that is assumed known and


where u
Sg t is the random spectrum compatible corrective term whose
u
power spectral density has to be determined. Taking into account
of the statistical independence of the two contributions the evolutionary spectrum-compatible power spectral density function is
given by the superposition of two terms:

Gug x; t GRug x; t GSug x; t

19

GRug

where
x; t is the joint timefrequency distribution of the
recorded accelerogram, while GSug x; t is the separable power spectral density function and represents a corrective term given by

GSug x; t a20 tG0 x

20

According to the quasi-stationary model described in the previous section a0 t is the modulating function and G0 x is determined modifying Eq. (6) as follows

P. Cacciola et al. / Engineering Structures 78 (2014) 316

G0 xi 0; 0 6 x 6 xa

RSAxi ;f0 2 RSAR xi ;f0 2


g 2 xi ;f0

4f0
G0 xi xi p4f
0 xi1

 Dx

i1
X

G 0 xk

x > xa

k1

21
R

 U is the peak factor dened in Eq.(7) and RSA is the pseudowhere g


acceleration response-spectrum for the recorded ground motions
Rg t. It is noted that the iterative scheme is given by
u
1
G0

x G0 x;

j
G0

j1
G0

RSAx;f0 2
RSAj1 x;f0 2

22

where RSA(j) represents the response-spectrum, of the spectrumcompatible stochastic ground motion model determined at the
j-th iteration. It has to be emphasized that Eq. (21) is dened for
RSAxi ; n0 > RSAR xi ; n0 as a consequence a preliminary scaling
procedure is usually required (see [3] for further details on the
procedure).
5. Non-stationary with imposed variability models
The sustained dissemination of database of recorded accelerograms along with the increasing number of strong-motion
networks installed worldwide revealed that the above methodologies for simulating articial earthquakes do not manifest large
variability of the seismological parameters observed for natural
accelerograms. As a consequence, even if the accelerograms are
simulated through a pertinent stochastic approach, the dispersion
of ground motion parameters such as peak ground acceleration
(PGA), Arias Intensity (AI), Cumulative Absolute Velocity (CAV) as
well as the spreading about the mean response-spectrum is
generally much less than those ones manifested by recorded accelerograms. Recently, Cacciola and Zentner [6] proposed a method
for simulating articial earthquakes compatible with a given
response-spectrum taking into account the natural variability
through a pertinent evolutionary power spectrum with random
coefcients. The distribution of the random coefcients has been
determined by the further matching with a target mean standard
deviation response spectra. Specically, in order to match the
target spectra with condence intervals, the following random
evolutionary spectrum is proposed

Gug x; t; a ax2 jax; tj2 Gx

23

in which, for xed xj e [0, X], a(xj) is a Gaussian random variable


possessing mean value la = la(xj) = 1 and standard deviation ra =
ra(xj) given by the following expression
r

ra xj

RSA xj ; f0
 1 8xj 2 0; X
RSAxj ; f0

24

where RSA+r(xj, f0) is the mean + standard deviation responsespectrum. Analogous expression can be found considering the
RSAr response-spectrum. Moreover, the simulation formula given
in Eq. (3) is modied as follows

r
u
g t

N h q
X

ar kDx2 jakDx; tj2 GkDxDx

where ra x is the standard deviation and RSArj x; f0 is the


simulated mean + standard deviation response-spectrum both
determined at the j-th iteration. This procedure has been extended
by DAmico et al. [7] and Zentner et al. [25] to include enhanced variability models (i.e. lognormal distribution) including the correlation of spectral accelerations.
6. Numerical example
In this section the structural responses to three sets of
response-spectrum-compatible ground motion time-histories are
compared, with the purpose of investigating the inuence of the
ground motion model selection on structures with different
dynamic characteristics. The buildings designed for the SMART
2008, IASC 1996 and SAC 1996 benchmarks were selected so to
explore a wide range scenario of fundamental natural frequencies.
6.1. Compatibility with Eurocode 8
A quasi-stationary and two non-stationary spectrum compatible
ground motion models were selected among the above mentioned
procedures. For illustrative purposes, the elastic acceleration
response spectrum prescribed by Eurocode 8 was used, specically
the Type 1 spectrum for ground type D and design ground acceleration of 0.32 g has been chosen assuming the damping ratio 5%
according to the code recommendations. The target response spectrum provided by Eurocode 8 is given below



T
0 6 T 6 TB
RSAT ag S 1 1:5
TB
RSAT 2:5ag S T B 6 T 6 T C
 
TC
TC 6 T 6 TD
RSAT 2:5ag S
T


TCTD
TD 6 T 6 4 s
RSAT 2:5ag S
T2

27

where ag is the design ground acceleration; S is the soil factor and T


is the natural period; specically the periods TB TC and TD are summarized in Table 1.
The mean response-spectrum of the simulated time-histories
has to match the target response-spectrum within the prescribed
tolerance along the prexed range of periods according to the
following conditions

(
)
RSAT  RSAT
max
 100 6 10%
RSAT

28

where RSA is the mean response-spectrum of at least three simulated accelerograms and RSA(T) is the target one, moreover it has
to be veried that

RSA0 > ag S

29

6.2. Ground motion models specications


The sets of time-histories were simulated according to ground
motion stochastic processes modeled as: quasi-stationary, nonstationary and non-stationary with imposed variability.

k1



r
 cos kDxt uk

25

The value of the stationary component G(x) is updated through


Eq.(9), while the standard deviation ra(x) is updated through the
following iterative scheme
j1
rj
xk
a xk ra

RSAr xk ; f0
RSArj1 xk ; f0

26

Table 1
Target response spectrum parameters of denition: Type 1, soil D.
Parameter

Value

S
ag
TB
TC
TD

1.35
0.32 g
0.20 s
0.80 s
2.00 s

P. Cacciola et al. / Engineering Structures 78 (2014) 316

EPSD [m 2 / s3 ]
EPSD [m 2 / s3 ]

a)

Frequency [rad/ s]

Time [s]

b)

Frequency [rad/ s]
Time [s]

EPSD [m 2 / s3 ]
EPSD [m 2 / s3 ]

c)

Frequency [rad/ s]

Time [s]

d)

Frequency [rad/ s]
Time [s]

Fig. 1. Evolutionary PSD functions: (a) C & P model, (b) quasi stationary; (c) non-stationary; and (d) non-stationary and variability.

35

Quasi-stationary
Non-stationary
30

Mean instantaneous frequency [rad/s]

Among the various procedures available in literature and briey


described in the previous sections in the following numerical analyses only those models than can be compared in terms of non-stationary behavior (i.e. amplitude non-stationarity and frequency
non-stationarity) have been considered. Specically, the procedures developed by Vanmarcke and Gasparini [24] in the recursive
form proposed by Cacciola et al. [2] was used to simulate quasi-stationary processes. The method proposed by Preumont [17] and
extended by Cacciola and Zentner [6] was employed for modeling
the non-stationary processes. Lastly, the model proposed by
DAmico et al. [7] was adopted for the simulation of non-stationary
signals with imposed variability. The other models described in
this paper (e.g. [23,12,3] have not been included in the comparisons as their non-stationary behavior cannot be controlled directly
in the simulation procedure.
All the quasi-stationary and non-stationary models can be seen
as a particular case of Eq. (23) by setting different values of a(x)
and a(x, t). Specically, the non-stationary model with imposed
variability was dened by assuming the random function a(x) as
log-normally distributed, with la = 1 and standard deviation
ra(x), determined according to Eq. (24) and assuming the target
variability of 1.5 RSA(T). This value of target variability has been
set after performing numerical studies on a sample of records from
the European Earthquake Database (see [6]. Note that, mean/median response spectrum along its variability should be determined

Non-stationary
and variability

25

20

15

10

10

15

20

Time [s]
Fig. 2. Mean instantaneous frequency for the response-spectrum compatible
models adopted.

a)

4
0
-4

Acceleration [m/s2 ]

P. Cacciola et al. / Engineering Structures 78 (2014) 316

4
0
-4

b)

T ime [s]

4
0
-4
0

T ime [s]

4
0
-4
10

15

4
0
-4

c)

4
0
-4

4
0
-4

4
0
-4

Time [s]

10

Time [s]

T ime [s]

4
0
-4

15

10

15

Time [s]

Fig. 3. Simulated time histories (a) quasi-stationary; (b) Non-stationary; (c) non-stationary and variability.

Fig. 4. Response spectral accelerations with damping ratio variation plotted versus: (a) period and (b) frequency.

Fig. 5. Percentage differences of the response spectra for 1% and 2% damping


evaluated with respect to the simulated 5% damped response spectra.

from the same earthquake database for performance based design


studies. However, for the purpose of demonstration and in absence
of variability data for the Eurocode response spectrum a feasible
variability has been used in the following numerical application.

Fig. 6a. SMART 2008 structure: Sap 2000 model.

For the other two models (non-stationary and quasi-stationary)


a(x) = 1. The non-separable frequency dependent modulating
function a(x, t) is selected according to the evolutionary Clough
Penzien model, that is

P. Cacciola et al. / Engineering Structures 78 (2014) 316

0.10
1.00

0.20

0.20

0.50

0.50

1.05

0.68

0.80

0.15

1.25

1.05

1.05

0.10
0.225
0.85

0.15

0.10

1.20

1.20
0.50

0.70
3.65

0.85

1.20

1.20

3.65
0.50
0.10

0.90

1.25

1.25
0.50

0.10

0.70

1.25

0.95

0.10

1.05

0.50

1.00

Fig. 6b. Geometrical data of the SMART 2008 structure.

a)

600

Quasi-stationary
Non-stationary

500

Non-stationary
and variability

b)

0.75

CDF

PDF

400
300

0.5

200
0.25

Quasi-stationary
Non-stationary

100

Non-stationary
and variabiity

0.003

0.006

0.009

0.012

Displacements [m]

0.003

0.006

0.009

0.012

Displacements [m]

Fig. 7. Smart output displacements: (a) probability density function and (b) cumulative density function.

Table 2
Smart 2008 structural responses: displacements statistics (m).
SMART

Mean

Fractile 5%

Fractile 50%

Fractile 95%

Quasi-stationary
Non-stationary
Non-stationary and variability
DNS%
DNSV%

0.0051
0.0050
0.0046
0.8676
8.1265

0.0040
0.0039
0.0025
3.2997
37.1894

0.0050
0.0049
0.0039
1.8624
20.5808

0.0064
0.0066
0.0088
4.1451
38.5640

10

P. Cacciola et al. / Engineering Structures 78 (2014) 316

0.0075

0.008

0.0065

Fractile 5% [m]

Mean Response [m]

7samples

Quasi-stationary
Non-stationary

7samples

Non-stationary
and variability

0.0055

0.006

0.004

0.0045

(a)
0.0035

50

0.002

100 150 200 250 300 350 400 450 500


Number of samples

0.0075

(b)
0

50

0.012

7samples

7samples
0.0065

Fractile 95% [m]

Median Response [m]

100 150 200 250 300 350 400 450 500


Number of samples

0.0055

0.009

0.006

0.0045

0.0035

(c)
0

50

100 150 200 250 300 350 400 450 500


Number of samples

0.003

(d)
0

50

100 150 200 250 300 350 400 450 500


Number of samples

Fig. 8. SMART building convergence of the output displacements statistics (a) mean; (b) fractile 5%; (c) median; and (d) fractile 95%.

with

15
t
tf

31

xf t 0:1xg t

32

xg t 20 

and for soft soil according after Der Kiureghian and Neuenhofer [9]
the following damping values have been selected

fg t 0:2; ff t 0:6

33
J

The modulating function at a t adopted is given in Eq. (10)


with the stationary part T s equal to 10 s leading to b = 0.9,
t1 = 2.78 s t2 = 12.78 s. Finally, the stationary counterpart of the
power spectral density function G(x) is determined according to
Eq. (6) for the quasi-stationary, non-stationary and non-stationary
with imposed variability models. The relevant evolutionary power
spectral density functions used for the simulations of the articial
accelerograms are depicted in Fig. 1. Fig. 2 shows the variation of
the mean instantaneous frequency versus the time in order to
highlight the differences in terms of non-stationary frequency content of the models adopted. Five hundred response-spectrum-compatible ground motion time-histories were generated according to
Eq. (3) by setting Na = 1000, Dx = 0.1 rad/s and total duration of
20 s. Samples from the quasi-stationary, the non-stationary and
non-stationary with imposed variability sets are displayed in
Fig. 3. The accelerograms are baseline corrected adding a polynomial function to the time histories as follows [16]
Fig. 9. 5-Story building: SAP 2000 model.

cg u
g t a1 a2 t a3 t 2 a4 t3
u


2
1 4f2g t xx
g t
J
ax; t a t 

2 2

2
1  xx
4f2g t xx
g t
g t


1

x
xf t

x
xf t

2 2

4

4f2f t xxt
f

2

30

34

cg is the corrected accelerogram, and a1, a2, a3, a4 the polynowhere u


mial coefcients determined to impose zero end value to the
ground motion velocity time histories.
The average of the ensemble of the articial accelerograms satises Eq. (28) as required by Eurocode 8 in the range between of
periods between 0 and 4 s as shown in Fig. 4. Furthermore the average of the PGA calculated from the set of articial accelerograms
are: 4.77, 4.62 and 4.97 m/s2 respectively for the quasi-stationary,

11

P. Cacciola et al. / Engineering Structures 78 (2014) 316

40

(a)

Non-stationary
and variability

0.75

CDF

30

PDF

(b)

Quasi-stationary
Non-stationary

20

10

0.5

0.25

Quasi-stationary
Non-stationary
Non-stationary
and variability

0.05

0.1

0.15

0.2

0.05

Displacements [m]

0.1

0.15

0.2

Displacements [m]

Fig. 10. 5-Story building output displacements: (a) probability density function and (b) cumulative density function.

Table 3
5-Story building structural responses: displacements statistics (m).
5 Story building

Mean

Fractile 5%

Fractile 50%

Fractile 95%

Quasi-stationary
Non-stationary
Non-stationary and variability
DNS%
DNSV%

0.0423
0.0387
0.0356
8.4710
15.9936

0.0260
0.0234
0.0139
9.7425
46.5767

0.0415
0.0377
0.0268
9.1024
35.2950

0.0633
0.0579
0.0762
8.5655
20.3271

7samples

0.07

0.06

Quasi-stationary
Non-stationary

0.05

Non-stationary
and variability

Fractile 5% [m]

Mean Response [m]

0.07

0.04

7samples

0.05

0.03

0.03

(b)

(a)
0.02

50

0.01

100 150 200 250 300 350 400 450 500

50

Number of samples

Number of samples
0.3

7samples

7samples

0.06

Fractile 95% [m]

Median Response [m]

0.07

100 150 200 250 300 350 400 450 500

0.05
0.04

0.2

0.1

0.03
0.02

(c)
0

50

100 150 200 250 300 350 400 450 500

Number of samples

(d)
0

50

100 150 200 250 300 350 400 450 500

Number of samples

Fig. 11. 5-Story building convergence of the output displacements statistics (a) mean; (b) fractile 5%; (c) median; and (d) fractile 95%.

non-stationary and non-stationary with variability models, all satisfying the Eurocode 8 condition to be greater than the value
ag S = 4.23 m/s2 (see Eq.(27)). Remarkably, the three sets of accelerograms full all the requirements of the response spectrum compatibility criteria provided by the Eurocode 8 and possess, within

negligible differences (i.e. below 5%), the same 5% damping mean


response spectrum, PGA and duration of the stationary part of
10 s. In Fig. 4 the mean response spectra from the same set of accelerograms for various damping ratios are also depicted. Note that
according to Eurocode provisions the accelerograms have to be

12

P. Cacciola et al. / Engineering Structures 78 (2014) 316

difference among them fairly represents the differences between


the models as all the sets of accelerograms possess the same identical mean response spectrum 5% damping. Interestingly, note that
for different damping ratios various differences arise especially in
the case of light damping (e.g. <2%). This is likely due to the fact that
light damped system possesses a narrow band transfer function and
therefore they are more sensitive to local non-stationary contribution of the input. It has to be emphasized that for the simplest linear
decaying non-stationary models adopted in this application, can be
observed differences respect the non-stationary case of about 20%
especially for higher frequencies (see Fig. 5). Clearly, alternative
non stationary models will provide different results. It is therefore
necessary to clarify that on purpose the differences between the
results from various ground motion models are labelled as D%
and not as a percentage error. Nowadays, it cannot be stated which
is the most accurate model among them. The (amplitude and frequency) non-stationary model selected in this paper is only one
of the possible feasible models available in literature, and up to
now does not exist any non-stationary model universally recognized as the most appropriate for engineering applications.
6.3. Structural responses and analysis tool
In this section the responses of the structures under responsespectrum-compatible accelerograms are presented. Three structures have been selected so to cover a wide scenario of fundamental
frequencies. Namely, the SMART 2008 reinforced concrete building,
the IASC 1996 and the SAC 1996 steel structures were selected as
representative of rigid, medium and slender structures respectively. All the structures were studied assuming a linear behavior.
Furthermore, the non-linear response of the moment resisting
frame SAC building was also studied to investigate the inuence
of the non-stationary input also in the non-linear range.
6.4. SMART 2008: 2-story reinforced concrete building

Fig. 12. SAC building: Sap 2000 model.

compatible with the response spectrum dened for 5% damping


ratio, even if the structure that need to be analyzed possesses a
damping ratio different from 5% [11]. The response spectra, for values of damping ratio different from the 5%, are purged by the difference between the non-stationary models and the quasi-stationary
one calculated for the 5% damping ratio. As a consequence the

(a)

(b)

Quasi-stationary
Non-stationary
Non-stationary
and variability

0.75

CDF

1.5

PDF

This structure designed for the Seismic Design and best


estimate methods assessment for Reinforced Concrete Building
subjected to Torsion is known as SMART 2008 project [14]. The
reinforced concrete building has been built at CEA Seismic Laboratory of Saclay in France, meeting the seismic French nuclear regulations as well as EC2 and EC8 prescriptions. The model has been
implemented in SAP2000 v15.0.1 assuming linear behavior with
a damping ratio equal to 2% for all the modes. The building is half
of a three story structure with trapezoidal plan, pertinent geometrical data are given in Fig. 6. The mass of the structure includes the

0.5

0.5

0.25

Quasi-stationary
Non-stationary
Non-stationary
and variability

0.5

1.5

Displacements [m]

2.5

0.5

1.5

2.5

Displacements [m]

Fig. 13. SAC building linear model output displacements: (a) probability density function and (b) cumulative density function.

13

P. Cacciola et al. / Engineering Structures 78 (2014) 316


Table 4
SAC building linear model structural responses: displacements statistics (m).
SAC building linear model

Mean

Fractile 5%

Fractile 50%

Fractile 95%

Quasi-stationary
Non-stationary
Non-stationary and variability
DNS%
DNSV%

0.7951
0.7977
0.7990
0.3362
0.4916

0.4747
0.4746
0.3743
0.0212
21.1422

0.7593
0.7590
0.7020
0.0368
7.5480

1.1804
1.2488
1.6365
5.7908
38.6425

0.9

7samples

Fractile 5% [m]

Mean Response [m]

0.8

Quasi-stationary
Non-stationary

0.6

50

0.7

0.5

Non-stationary
and variability

(a)
0.4

7samples

0.3

100 150 200 250 300 350 400 450 500

(b)
0

50

100 150 200 250 300 350 400 450 500

Number of samples

Number of samples
2.5

7samples

Fractile 95% [m]

Median Response [m]

0.8

0.6

0.4

(c)
0

100

200

300

400

500

7samples

1.5

0.5

(d)
0

50

100 150 200 250 300 350 400 450 500

Number of samples

Number of samples

Fig. 14. SAC building linear model convergence of the output displacements statistics: (a) mean; (b) fractile 5; (c) median; and (d) fractile 95%.

3000
2000

Moment [KNm]

self-weight of the building and the additional loads that are of


12 tons for the rst and the second slab and 11 tons for the roof.
The modulus of elasticity and the unit weight used in the numerical analysis are 3.2  1010 N/m2 and 2460 kg/m3, respectively.
The rst two natural frequencies of the structure are 8.9 Hz and
15.9 Hz. Fig. 6a shows the Finite Element model of the SMART
2008 structure. The structure has been forced by the three set of
accelerograms acting along the x-direction according to the reference system shown in Fig. 6b. In Fig. 7 the distributions of the maxima of the displacement in the x-direction of the selected node (see
circled joint in Fig. 6a) are represented. Relevant statistical
response quantities are displayed in Table 2. Differences in absolute value of the non-stationary model (NS) and non-stationary
with variability model (NSV) respect to the quasi-stationary model
(QS) are also reported in Table 2. It can be observed that the nonstationary model do not provide signicant differences with
respect to the quasi-stationary case, whereas the non-stationary
with variability model presents signicant differences in all the
relevant statistics. Remarkably, as might be expected, the highest
difference is detected for the fractiles and in particular the 95%
fractile of the non-stationary with variability model with a value
of the 38% higher than the correspondent one for quasi-stationary
process. The convergence of the fractiles of the response is also
reported in Fig. 8. It is noted that Eurocode 8 requires a minimum
of 3 accelerograms for the time-history analysis and suggest to use
the average of the response quantities if a minimum number of 7

1000
0
-1000
-2000
-3000
-0.04

-0.02

0.02

0.04

Rotation
Fig. 15. Moment-rotation relationship for moment resisting column beam connection of the analyzed node at the roof.

accelerograms are used. From Fig. 8 it is evident that even for the
simplest quasi-stationary case the suggested numbers of accelerograms do not lead to stable statistics.
6.5. IASC 1996: 5-story steel frame
The second benchmark structure analyzed in this paper is a
5-story steel frame IASC 1996 shown in Fig. 8. As for this structure
the stiffness matrix was already available [20,10], namely

14

P. Cacciola et al. / Engineering Structures 78 (2014) 316

3.5

(a)

Non-stationary

Non-stationary
and variability

2.5

0.75

CDF

PDF

(b)

Quasi-stationary

1.5
1

0.5

0.25

Quasi-stationary
Non-stationary

0.5
0

Non-stationary
and variability

0.4

0.8

1.2

1.6

0.4

0.8

1.2

1.6

Displacements [m]

Displacements [m]

Fig. 16. SAC building non linear model output displacements: (a) probability density function and (b) cumulative density function.

Table 5
SAC building non-linear model structural responses: displacements statistics (m).
SAC building non-linear model

Mean

Fractile 5%

Fractile 50%

Fractile 95%

Quasi-stationary
Non-stationary
Non-stationary and variability
DNS%
DNSV%

0.6275
0.6283
0.6041
0.1319
3.7272

0.4277
0.4208
0.3354
1.6049
21.5823

0.6183
0.6064
0.5861
1.9261
5.2093

0.8359
0.8841
0.9624
5.7702
15.1320

0.75

0.7

0.65

Fractile 5% [m]

Mean Response [m]

0.65

0.55
Quasi-stationary
Non-stationary

0.45

0.35

Non-stationary
and variability

7samples

50

0.6
0.55
0.5
0.45
0.4
0.35

(a)

100 150 200 250 300 350 400 450 500

0.3
0.25

7samples

50

(b)

100 150 200 250 300 350 400 450 500

Number of samples [m]

Number of samples [m]


1.3

0.75

0.65

Fractile 95% [m]

Median Response [m]

1.2

0.55

0.45

0.35

1.1
1
0.9
0.8
0.7
0.6

(c)

7samples

100

200

300

400

500

Number of samples [m]

0.5
0.4

7samples

50

(d)

100 150 200 250 300 350 400 450 500

Number of samples [m]

Fig. 17. SAC building non linear model convergence of the output displacements statistics: (a) mean; (b) fractile 5; (c) median and (d) fractile 95.

4270
6
6 3124
6
K6
6 64
6
4 1
15

3124 64

1

15

7
7
7
7 kN=m
3156 6174
3156 74
7
7
65
3156 6172
3079 5
9
74
3079 2978

6174

3156 65

35

the structure has been modelled in MATLAB assuming equally concentrated masses at each oor level of 1042 kg and with damping
ratio of 0.2% for all the modes. The rst natural frequency is
4.27 Hz. The structure has been forced by the three sets of accelerograms in the x direction of the referring system shown in the
equivalent SAP2000 model depicted in Fig. 9 and the distributions
of the maxima displacement of the top oor along the x direction

P. Cacciola et al. / Engineering Structures 78 (2014) 316

are shown in Fig. 10. Relevant response data are summarized in


Table 3. In this case signicant differences can be observed with
respect to all the fractiles and mean values. Proof of convergence
is also reported in Fig. 11.
6.6. SAC building: 20-story steel frame
Finally, the third benchmark structure analyzed in this paper is
the moment resisting frame for the SAC Phase 2 Steel project
designed by Brandow & Johnston Associates (1996). The SAC joint
venture aimed to study structural control strategies for non-linear
3, 9 and 20-story buildings. The 20-story steel frame shown in
Fig. 12 has been selected and modelled in SAP2000 pertinently
with the regulation for the Los Angeles, California region. The study
of the structural response is performed through the 2-dimensional
analysis of the moment resisting frame. Geometrical details are
reported in [15]. Pertinently with the benchmark presentation
the rst ve natural frequencies of the reproduced model are
0.261 Hz, 0.75 Hz, 1.30 Hz, 1.83 Hz, 2.4 Hz; 2% damping is assigned
to the rst and fth frequency according to Rayleigh model.
The linear case is analyzed rst. Fig. 13 shows the distributions
of the displacement of the top oor for the three different set of
accelerograms. Relevant results are reported in Table 4. Negligible
discrepancies can be observed between the quasi-stationary and
the non-stationary model, while the non-stationary model with
imposed variability provided, as expected, signicant differences
in the tails leading to a 95% response fractile 38% also in this case
higher than the correspondent one for quasi-stationary process.
This is in contrast with the common understanding that quasi-stationary models lead to over-conservative response with respect to
the non-stationary ones. This general idea should be revised in the
light of these results. Clearly, it depends how the accelerograms are
generated. Proof of convergence is also reported in Fig. 14.
The non-linear time history analysis has been conducted on the
same structure according to FEMA-356 recommendations adopted
to dene the plastic hinges. The non-linear model associated to the
moment-rotation is shown in Fig. 15. Distributions of the maximum displacements of top oor are reported in Fig. 16 and summarized in Table 5. Also in this case signicant differences can be
observed with respect to the tails especially for the 5% and 95%
fractiles. Interestingly for this case, the discrepancies between
the 95% fractile in the case of linear and non-linear model are more
evident in the linear case, again in opposite of the general view that
the non-stationarity of the ground motion models is more relevant
for non-linear behaving structures. Proof of convergence is also
reported in Fig. 17. It is observed that for non-linear time-history
analysis the statistics of the response converge more slowly in
comparison to the linear case. Remarkably the minimum number
of 7 accelerograms for the non-linear time-history analysis to use
the average of the response quantities should be signicantly
increased (e.g. 100 samples), especially if higher order statistics,
such as 95% fractile, will be used for design purpose.
7. Concluding remarks
In this paper the seismic analysis of structural systems forced
by response-spectrum-compatible accelerograms generated
according to the Eurocode 8 provisions has been performed. Three
different stochastic ground motion models (i.e. quasi-stationary,
non-stationary and non-stationary with imposed variability) have
been adopted possessing the common feature to having the same
mean 5% damping response spectrum, PGA, strong motion phase
and the same total duration. From the comparisons of the structural responses to the three sets of accelerograms it has been highlighted that even if the Eurocode 8 provisions have been fullled

15

the distributions of relevant response parameters such as the peak


displacement are sensibly different for the three models adopted
with differences of about 3040% of the response fractiles. This
result manifests unequivocally the weakness of the responsespectrum-compatible criteria currently available in the seismic
codes. Moreover, it has been shown that the response of structures
having damping ratio smaller than the nominal 5% (used as a
target) is sensitive to the non-stationary model adopted in the simulation. Furthermore the general view that the non-stationarity of
the ground motion models is more relevant for non-linear behaving structures in comparison with the linear case should be
revised, as the results shown in this paper manifest that it is possible the opposite. Finally, the general accepted understanding that
quasi-stationary models lead to over-conservative response compared to the non-stationary model should be also revised, as it
depends on the strategy used to generate the accelerograms. The
ndings provided in this paper enforce the importance to dene
reliable non-stationary models for ground motion simulations,
which up to now has been highlighted only for non-linear behaving structures and the necessity to redene the seismic code provisions for the time-history representations of the ground motion.
Acknowledgments
The rst author wishes to acknowledge the support by the
EPSRC First Grant EP/K004867/1 Vibrating Barriers for the control
of seismic waves (ViBa).
References
[1] Ahmadi G. Generation of articial time-histories compatible with given
response spectra a review. SM Arch 1979;4(3):20739.
[2] Cacciola P, Colajanni P, Muscolino G. Combination of modal responses
consistent with seismic input representation. J Struct Eng 2004;130(1):4755.
[3] Cacciola P. A stochastic approach for generating spectrum compatible fully
non-stationary earthquakes. Comput Struct 2010;88(1516):889901.
[4] Cacciola P. Stochastic ground motion modelling for the seismic analysis of
structures: a review. Comput Technol Rev 2011;4:6591.
[5] Cacciola P, Deodatis G. A method for generating fully non-stationary and
spectrum-compatible ground motion vector processes. J Soil Dyn Earthquake
Eng 2011;31(3):35160.
[6] Cacciola P, Zentner I. Generation of response spectrum compatible articial
earthquake accelerograms with random joint time frequency distributions.
Probabilist Eng Mech 2012;28:528 .
[7] DAmico L, Zentner I, Cacciola P. Simulation of spectral acceleration correlated
and response spectrum compatible ground motion accelerograms. In: 15th
World conference on earthquake engineering, Lisbon, Portugal; 2012.
[8] Der Kiureghian A. Structural response to stationary excitation. J Eng Mech Div
1980;106:1195211.
[9] Der Kiureghian A, Neuenhofer A. Response spectrum method for multiplesupport seismic excitation. Earthquake Eng Struct Dyn 1992;21(8):71340.
[10] Djajakesukma SL, Samali B, Nguyen H. Study of a semi-active stiffness damper
under various earthquake inputs. Earthquake Eng Struct Dyn
2002;31:175776.
[11] Eurocode 8. Design of structures for earthquake resistance, Part 1: General
rules, seismic actions and rules for buildings; 2010.
[12] Giaralis A, Spanos PD. Wavelets based response spectrum compatible
synthesis of accelerograms Eurocode application (EC8). Soil Dyn
Earthquake Eng 2009;29:21935.
[13] Jennings PC, Housner GW, Tsai C. Simulated earthquake motions for design
purpose. In: Proc 4th World conference earth. Engineering Santiago, A(1);
1969. p. 14560.
[14] Lermitte S, Chaudat T, Payen T, Vandeputte D, Viallet E. SMART 2008: seismic
design and best estimate methods assessment for RC buildings subjected to
torsion. In: 14th World conf on earthquake eng, Beijing, China; 2008.
[15] Ohtori Y, Christenson RE, Spenser BF, Dyke SJ. Benchmark control problems for
seismically
excited
nonlinear
buildings.
J
Eng
Mech
(ASCE)
2004;130(4):36685.
[16] Preumont A. The generation of spectrum compatible accelerograms for the
design of nuclear power plants. Earthquake Eng Struct Dyn 1984;12:48197.
[17] Preumont A. The generation of non-separable articial earthquake
accelerograms for the design of nuclear power plants. Nucl Eng Des
1985;88:5967.
[18] Priestley MB. Spectral analysis and time series. Academic Press; 1981.
[19] Rezaeian S, Der Kiureghian A. Simulation of synthetic ground motions for
specied earthquake and site characteristics. Earthquake Eng Struct Dyn
2010;39:115580.

16

P. Cacciola et al. / Engineering Structures 78 (2014) 316

[20] Samali B. System identication of a ve storey benchmark model using modal


analysis. In: Proceedings of the Asia Pacic vibration conference; 2007. p. 7016.
[21] Shinozuka M. Stochastic elds and their digital simulation. In: Shinozuka M,
editor. Stochastic methods in structural dynamics. Boston: Martinus Nijhoff
Publishers; 1987. p. 93133.
[22] Shinozuka M, Deodatis G. Stochastic process models for earthquake ground
motion. Probabilist Eng Mech 1988;3:11423.
[23] Spanos PD, Vargas Loli LM. A statistical approach to generation of design
spectrum compatible earthquake time histories. Soil Dyn Earthquake Eng
1985;4(1):28.

[24] Vanmarcke EH, Gasparini DA. Simulated earthquake ground motions. In: Proc
4th int conf on smirt, K1/9, San Francisco; 1977.
[25] Zentner, DAmico, Cacciola. Simulation of non stationary ground motion
compatible with NGA-spectra. New York: ICOSSAR; 2013.
[26] Zentner I, Poirion F. Enrichment of seismic ground motion databases using
Karhunen Love expansion. Earthquake Eng Struct Dyn 2012;41(14):194557.

You might also like