Sunteți pe pagina 1din 23

April 28, 2006 13:47 WSPC/124-JEE 00273

Journal of Earthquake Engineering


Vol. 10, Special Issue 1 (2006) 67–89
c Imperial College Press

AN IMPROVED METHOD OF MATCHING RESPONSE


SPECTRA OF RECORDED EARTHQUAKE GROUND
MOTION USING WAVELETS

JONATHAN HANCOCK† , JENNIE WATSON-LAMPREY‡ ,


NORMAN A. ABRAHAMSON§ , JULIAN J. BOMMER∗,† ,
ALEXANDROS MARKATIS† , EMMA McCOY|| and RISHMILA MENDIS†
†Department of Civil and Environmental Engineering, Imperial College London, UK
‡University of California, Berkeley, California, USA
§Pacific Gas and Electricity Company, San Francisco, USA
||Department of Mathematics, Imperial College London, UK

Dynamic nonlinear analysis of structures requires the seismic input to be defined in the
form of acceleration time-series, and these will generally be required to be compatible
with the elastic response spectra representing the design seismic actions at the site. The
advantages of using real accelerograms matched to the target response spectrum using
wavelets for this purpose are discussed. The program RspMatch, which performs spectral
matching using wavelets, is modified using new wavelets that obviate the need to sub-
sequently apply a baseline correction. The new version of the program, RspMatch2005,
enables the accelerograms to be matched to the pseudo-acceleration or displacement
spectral ordinates as well as the spectrum of absolute acceleration, and additionally
allows the matching to be performed simultaneously to a given spectrum at several
damping ratios.

Keywords: Dynamic analysis; accelerograms; wavelets; spectrum-compatible records;


spectral matching; RspMatch.

1. Introduction
Seismic design of structures is invariably based on representation of the earthquake
actions in the form of a response spectrum. In many situations, however, including
the design of critical facilities, highly irregular buildings and base-isolated struc-
tures, the simulation of structural response using a scaled elastic response spectrum
is not considered appropriate to verify the earthquake resistance. In such cases,
dynamic nonlinear analysis of the structure will be required and the seismic input
then needs to be defined in the form of acceleration time-series, which will gener-
ally be required to be compatible with the elastic response spectra representing the
design seismic actions at the site. There are many different options for obtaining
suites of accelerograms for use in engineering design and assessment [e.g. Bommer

∗ Corresponding author: Tel.: +44-20-7594-5984, Fax: +44-20-7594-5934, E-mail: j.bommer@


imperial.ac.uk.

67
April 28, 2006 13:47 WSPC/124-JEE 00273

68 J. Hancock et al.

and Acevedo, 2004], the most widely used approaches being the use of artificial
spectrum-compatible time-series, generated from white noise, and the use of scaled
real accelerograms.
Artificial records constitute a convenient tool but their shortcomings, arising
from their dissimilarity with real earthquake ground motions in terms of num-
ber of cycles, phase content and duration, are widely recognised, and their use in
nonlinear analyses is not recommended. These problems are avoided by using real
strong-motion accelerograms, appropriately scaled to the target spectrum (at least
in the vicinity of the structure’s natural period of vibration), but the inherent vari-
ability of real earthquake motions means that it will often be necessary to run large
numbers of dynamic analyses in order to obtain stable estimates of the inelastic
response of the structure. The required number of inelastic dynamic analyses can
be significantly reduced if the real records are first matched to the target response
spectrum, by eliminating the largest differences between the target spectrum and
the spectral ordinates of individual accelerograms. This is clearly a compromise and
in some sense the records become ‘artificial’ as a result, although the records can
retain most (if not, in fact, all) of the characteristics of real earthquake records.
The choice is essentially one of compromise between engineering pragmatism and
seismological rigour, reducing the number of time-consuming structural analyses
whilst avoiding the use of completely artificial accelerograms generated from mod-
ified white noise.
A commonly used method to reduce the spectral mismatch of the individ-
ual ground motions is to apply spectral matching in the frequency domain by
adjusting the Fourier amplitude spectra [e.g. Rizzo et al., 1975; Silva and Lee,
1987]. This is useful in that it generates accelerograms that are based on real
ground motions and also have a close match to the target spectrum. However,
adjusting the Fourier spectrum corrupts the velocity and displacement time-series
and can result in motions with unrealistically high energy content [Naeim and
Lew, 1995].
An alternative approach for spectral matching adjusts the time history in the
time domain by adding wavelets to the acceleration time-series. Wavelet adjust-
ment of recorded accelerograms has the same advantages as the Fourier adjustment
methods but leads to a more focused correction in the time domain thus intro-
ducing less energy into the ground motion and also preserves the non-stationary
characteristics of the original ground motion. This paper describes the work con-
ducted to create an improved version of the program RspMatch, originally devel-
oped by Abrahamson [1992] using the technique of Lilhanand and Tseng [1987,
1988], which is named RspMatch2005. The wavelet adjustment techniques incor-
porated to this new version of the software have the additional advantage that
they do not cause a drift in the velocity or displacement time-series. RspMatch2005
also allows the records to be matched to a pseudo-acceleration spectrum rather
than only the spectrum of absolute acceleration, and through improved convergence
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 69

properties allows the record to be matched to a given target spectrum with several
different levels of damping simultaneously. This feature is particularly useful when
long-period highly damped spectral displacements are relevant, such as in the
direct displacement-based design approaches [e.g. Kowalsky et al., 1995] and in
the analysis of buildings and bridges with base isolation or supplementary damping
devices.

2. Existing Wavelet Methods


There are several different methods of using wavelets to adjust accelerograms so
that they have a closer match to a target response spectrum. Mukherjee and Gupta
[2002a, 2002b] and Suarez and Montejo [2003, 2005] use wavelets and the continuous
wavelet transform (CWT) to de-compose the original acceleration time-series into a
number of time series with energy in non-overlapping frequency bands. An iterative
procedure is used to scale each time history so that when they are added together
they produce a spectrum-compatible ground motion. Although the approximate
duration of the original accelerogram is retained using this type of adjustment pro-
cedure, the adjusted accelerograms have visibly different amplitudes and frequency
contents from the original accelerogram.
The method proposed by Lilhanand and Tseng [1987, 1988] employs wavelets
but uses the response of elastic SDOF systems rather than the CWT. This enables
accelerograms to be made spectrum compatible with smaller adjustments than the
wavelet adjustment methodologies which use the CWT. The Lilhanand and Tseng
[1987, 1988] methodology is adopted as the basis for this work.
A flowchart showing the original procedure as employed in RspMatch is given
in Fig. 1. The essence of the methodology is as follows:

(1) Calculate the response of an elastic SDOF system under the action of the
acceleration time-series for each period and damping level to be matched.
(2) Compare the peak of each SDOF response with the target amplitude and deter-
mine the mismatch.
(3) Add wavelets to the acceleration time-series with the appropriate amplitudes
and phasing so that the peak of each response matches the target amplitude.
One wavelet is used to match one SDOF response.

Each wavelet is applied to the time series so that the time of maximum SDOF
response under the action of the wavelet is the same as the time of the peak response
to be adjusted from the unadjusted acceleration time-series. A fundamental assump-
tion of the method is that the time of the peak response does not change as a result
of adding the wavelet adjustment. This assumption is not always valid and this can
lead to diverging solutions, as is discussed in more detail later.
April 28, 2006 13:47 WSPC/124-JEE 00273

70 J. Hancock et al.

Input in target spectrum


Input in accelerogram

If required subdivide target


Calculate spectral response into frequency bands

Load next frequency


Calculate amplitude and sign of spectral of misfit range to be matched
No

Is misfit less Is last frequency


than tolerance? Yes range?

No Yes

Subdivide target into subgroups, each with Save results to file


periods spread throughout the frequency range

Load next subgroup

Calculate spectral response at periods in subgroup

Calculate spectral misfit at periods in subgroup

Calculate “C ” matrix
Calculat rix for subgroup. This relates the amplitude
of each wavelet to peak response at each period to be adjusted.

Conduct singular value decomposition of C matrix


atrix

Find linear scale factor for each wavelet by


solving C matrix, minimising the spectral misfits

Scale and sum wavelets to create adjustment function

Add adjustment to total adjustment function

Yes Add total adjustment function


No Is this the to accelerogram and clear
last subgroup? total adjustment function

Fig. 1. Basic methodology of RspMatch program.

The amplitude of each wavelet used in the adjustment is determined by the solu-
tion of a set of simultaneous equations that account for the cross correlation of each
wavelet with each response to be matched. This can be expressed in matrix form:

C · b = δR, (1)
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 71

where C is a square matrix with elements that describe the amplitude of each
SDOF response, at the time that the response needs to be adjusted, under the
action of each wavelet, b is a vector of linear scale factors for each wavelet used
in the adjustment, and δR is a vector of the required adjustment, the difference
between the peak SDOF response of the unadjusted time-series and the required
amplitude specified by the target spectra response for each period and damping
level to be matched.
The wavelet scale factors in the b vector are found using amplitude of the
required adjustment and the inverse of the correlation matrix C:

b = C −1 · δR. (2)

The amplitude of the wavelet adjustment function at time t is determined from


the sum of the amplitudes of the wavelets at that time, aj (t), multiplied by their
individual scale factors bj .

w
j=N
Adjustment(t) = bj · aj (t), (3)
j=1

where Nw is the total number of wavelets. The adjusted acceleration time-series is


the sum of the original time-series and the adjustment function.
Unfortunately, the correlation matrix C can be singular and of a size that took
considerable time to solve when the method was first proposed. These problems
were overcome by splitting the problem into smaller subgroups and conducting
singular value decomposition on the C matrix.
Users of RspMatch have known for many years that the original time-series
will retain more of its original character if the adjustment is applied in stages over
progressively wider frequency bands. This is also true for RspMatch2005 and in the
example given below the frequency match has been applied in two stages: the first
matches from 0.05 s (20 Hz) to 1.0 s, and the second from 0.05 s to 5.0 s. For brevity
only the results of the final stage are presented.
The Lilhanand and Tseng [1987, 1988] method generally works well, but there
are two main problems with the procedure. Firstly, the wavelet used corrupts the
velocity and displacement time-series of the accelerograms, so a baseline correction
is required after the wavelet adjustment, which can partially undo the spectral
match. Secondly, the method is not always stable and diverges if the user attempts
to match at closely spaced periods and multiple damping levels.

3. Wavelet Functional Form


One of the key features to the adjustment method is the functional form used for
the wavelet adjustment. Wavelets have many different functional forms, but in the
interest of brevity, only those used in the RspMatch and RspMatch2005 programs
are described in this section.
April 28, 2006 13:47 WSPC/124-JEE 00273

72 J. Hancock et al.

3.1. Wavelet used by Lilhanand and Tseng


The original wavelet used by Lilhanand and Tseng [1987, 1988] and RspMatch is a
reverse impulse function:
−ωj    
aj (t) =  exp(−ωj βj (tj − t)) 2βj2 − 1 sin ωj (tj − t)
1 − βj2
  
− 2βj 1 − βj2 cos ωj (tj − t) , (4)
where,
aj (t) is the amplitude of the jth wavelet at time t
tj is the time of the peak response of the jth oscillator under the action of
the jth wavelet
ωj is the circular frequency of the jth wavelet
βj is the damping level (proportion of critical)
 of the jth oscillator
ωj is the damped circular frequency ωj = ωj 1 − βj2 .

Although this wavelet is very efficient in adjusting the response, it has the
disadvantage that it corrupts the velocity and displacement time-history because
the wavelet does not end with zero velocity or displacement (Fig. 2). To over-
come this issue two new displacement compatible wavelets have been created for
RspMatch2005.

3.2. Sinusoidal corrected wavelet


The sinusoidal corrected wavelet is a hybrid wavelet based the wavelet used by
Suarez and Montejo [2003, 2005] that includes a sinusoidal correction to ensure zero
final displacement. The equation describing the Suarez and Montejo wavelet is:

aj (t) = e−βj ωj |t−tj +∆tj | sin(ωj (t − tj + ∆tj )), (5)

where ∆tj is the difference between time of peak response tj and the reference
origin of the wavelet. Unlike the other wavelets described above, this wavelet is only
applied for a fixed number of cycles (Nc ) specified by the user and automatically
reduced by the program as required to ensure the whole wavelet is applied to the
accelerogram.
The error in the final displacement from the Suarez and Montejo wavelet is
obtained by double integration of the wavelet and applying the appropriate initial
conditions:
 
−2βj + e−βj tdj ωj (2βj + tdj ωj + βj2 tdj ) cos(tdj ωj )
 + e−βtdj ω (1 − βj2 − βtdj ωj ) sin(tdj ωj ) 
 
DispErrorj = 2   2 , (6)
 1 + β2 ω2 
j

where tdj is half the duration of the uncorrected jth wavelet equal to π
ωj ∗ Nc .
April 28, 2006 13:47 WSPC/124-JEE
00273

Fig. 2. Acceleration (upper), velocity (middle) and displacement (lower) time-series of reverse impulse wavelet (left), sinusoidal corrected wavelet
(middle) and corrected tapered cosine wavelet (right).
Matching Response Spectra of Recorded Earthquake Ground Motion
73
April 28, 2006 13:47 WSPC/124-JEE 00273

74 J. Hancock et al.

To overcome this problem a sinusoidal half cycle at the start and end of the
wavelet is used to correct the displacement time-series (Fig. 2). The amplitude of
the sinusoidal correction is given by:

−DispError · ωj
SinAmplitudej =  2π  . (7)
ωj + 4tdj

In the unlikely event that there is insufficient space within the record to apply a
sinusoidal correction at the end of the wavelet, a polynomial baseline correction is
applied to the wavelet.

3.3. Corrected tapered cosine wavelet


The corrected tapered cosine wavelet is an update of the wavelet used by
Abrahamson [1992] that includes an additional correction to ensure zero final dis-
placement (Fig. 2). The equation describing the tapered cosine wave is given by:

aj (t) = cos[ωj (t − tj + ∆tj )] exp[−|t − tj + ∆tj |ψj ], (8)

where ∆t for the tapered cosine wavelet is given by:


 √1−β 
tan−1
j
βj
∆tj = . (9)
ωj

The frequency dependence of ψj should be consistent with the reference time-


history. That is, if the reference time-history has a short duration at a particular
frequency, the ψj should be selected such that the adjustment function at that
frequency will also have a short duration. A tri-linear model for ψj (f ) is used in
this program:


z1 for fj < f1
 (f − f1 )
ψ(f ) = z1 + (z2 − z1 ) for f1 < fj < f2 , (10)

 (f 2 − f1 )

z2 for fj > f2

where f1 , f2 , z1 and z2 are constants and fj is the frequency of the jth wavelet in
Hz; the recommended values of the constants are f1 = 1 Hz, f2 = 4 Hz, z1 = 1.25
and z2 = 0.25. The equation for the corrected tapered cosine wavelet is given by:

aj (t) = cos[ωj (t − tj + ∆tj )] exp[−|t − tj + ∆tj |ψj


+ [c1 (t − tj + ∆tj ) + c2 ] exp[−|t − tj + ∆tj |5ψj ]. (11)

The corrected tapered cosine wavelet is set so that it starts with an initial 1/4
acceleration cycle to avoid long-period drift in the displacement time-series.
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 75

4. Improved Solution Procedure


The fundamental assumption underlying the matrix solution method used in
RspMatch is that the time of the peak response is the same before and after each
adjustment is applied. If this assumption was always valid the problem would be
linear and could be solved exactly with a single iteration. Unfortunately, this is not
the case and the problem is nonlinear, particularly if closely spaced spectral points
and multiple damping levels are to be matched. Movement in the time of a response
peak is caused from either:

• A phase change in the response peak


• A new “secondary” peak becoming critical

Both of these sources of divergence are illustrated in Fig. 3 and are caused by
the cross correlation of different wavelet corrections. Put simply the wavelet added
to correct one period and damping level can also affect the peak response of SDOF
systems at other periods and damping levels. A general framework for describ-
ing non-linear problems is presented by Tarantola [2005]. The specific methods
developed by the authors for the solving this particular problem is described in the
following sections.

Fig. 3. Illustration of sources of diverging response. Note that the response amplitude remains
approximately unchanged at the time of the original peak response, but the response increases
through both a phase shift and the emergence of a secondary peak.
April 28, 2006 13:47 WSPC/124-JEE 00273

76 J. Hancock et al.

4.1. Reduction of off-diagonal C matrix elements


RspMatch subdivides the C matrix into smaller sub-matrices, which increases solu-
tion speed but more importantly reduces the cross correlation of the wavelets and
the likelihood of a phase change in the peak response. The subdivision essentially
sets some of the off-diagonal terms to zero, which is beneficial as it provides numer-
ical stability but is at the expense of the accuracy of the solution because the cross
correlation of some wavelets is not taken into account. RspMatch2005 avoids this
issue by using the full C matrix and obtains numerical stability by reducing the
off-diagonal terms by a constant factor. From conducting trials with different off-
diagonal reduction factors, it is found that a reduction factor of about 0.7 is very
effective in providing numerical stability.

4.2. Preventing secondary peaks


Reducing the off-diagonal terms of the C matrix improves numerical stability, but
it does not reduce the occurrence of secondary peaks. The new solution procedure
checks for divergence using the maximum misfit, which is defined as the greatest
misfit calculated from all of the periods to be matched; misfit at spectral period T
is defined as:
 
 SA(T ) − SAtarget (T ) 

Misfit (T ) =   ∗ 100, (12)
SAtarget (T ) 

where SA(T ) is the spectral acceleration of the adjusted ground motion at this
iteration at period T , and SAtarget (T ) is the target spectral acceleration.
The new solution procedure prevents divergence from secondary peaks by adding
a new wavelet at the period, damping level and time of the new secondary peak.
The adjustment function is recalculated using the secondary wavelet. An example
of the application of this method is shown in Fig. 4.

4.3. Reduction of correction amplitude


The solution can sometimes still diverge, even with the correction of secondary
peaks. This occurs when the cross correlation of the responses causes a phase change
of the response peak. Figure 5 shows one such case; here the original response has
a good match with the target, but wavelets introduced to adjust the response
at other periods cause a phase change and a divergence. This issue is overcome
by increasing the importance of the SDOF with the diverging response. This is
achieved by reducing the amplitude of the adjustment correction, δR, by 30% for all
the points to be matched except the point causing divergence. Although artificially
reducing the amplitude of the adjustment function increases the numerical stability,
it results in a greater number of iterations being required to obtain the required
spectral match.
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 77

Fig. 4. Prevention of diverging response with an additional wavelet adjustment.

Fig. 5. Prevention of diverging response with reduction of correlated wavelet targets.


April 28, 2006 13:47 WSPC/124-JEE 00273

78 J. Hancock et al.

4.4. Pseudo spectral acceleration


Spectral displacements are required for very long-period structures, tunnels, base
isolated structures and for direct displacement-based design. Nonlinear static
(pushover) methods of analysis may require elastic spectral displacements for peri-
ods up to about 5 seconds and damping levels from 5% of critical up to about
30% in most cases. As spectral displacements are directly related to the pseudo
spectral acceleration (PSA), not the absolute spectral acceleration, PSA should be
used when matching spectral displacements. PSA can be calculated directly from
the spectral displacement, SD for any period T :
 2

P SA = SD . (13)
T
RspMatch2005 uses pseudo-spectral accelerations for this reason. The difference
between the pseudo and absolute spectral accelerations only becomes significant
for damping levels above about 20%.

4.5. New solution algorithm


A flowchart detailing the new solution algorithm is shown in Fig. 6. Although the
new algorithm prevents the solution from diverging it does not guarantee that the
solution will converge to within the requested tolerance. For cases with multiple
damping levels and closely spaced spectral points it may be necessary to accept a
more relaxed tolerance criteria than for cases where only a single damping level is
to be matched. A balance needs to be maintained between the goodness-of-fit to
the response spectra and the degree of adjustment made to the accelerogram.
The ability of using different sub-groups has been left in the program to ensure
compatibility with earlier versions of the code; however, this is not recommended
with use of the new algorithms that employ off-diagonal reduction.

5. Examples of New Matching Procedure


To illustrate the new matching procedure a spectrum-matched accelerogram is pro-
duced for the scenario of a stiff soil site at 10 km from an Ms 7 earthquake. The
first half of this section shows the results of matching to the target acceleration and
displacement spectra at a 5% damping level. The second half of this section shows
that RspMatch2005 is capable of producing a ground motion that simultaneously
matches the 5, 10, 20 and 30% damping levels, provided a relaxation of the solution
tolerance is accepted.

5.1. Selection of seed accelerogram


The first step in the process is to select a suite of real accelerograms that may
be linearly scaled to obtain an approximate match with the spectral ordinates
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 79

Input in target spectrum


Input in accelerogram

If required subdivide target


Calculate spectral response into frequency bands

Load next frequency


Calculate amplitude and sign of spectral of misfit range to be matched
No

Is misfit less Is this the last


than tolerance? Yes frequency range?

No Yes

Subdivide target into subgroups, each with Save results to file


periods spread throughout the frequency range

Load next subgroup

Add wavelet to adjust


Calculate spectral response at periods in subgroup
secondary peak

Calculate spectral misfit at periods in subgroup

Calculate ““C” matrix


atrix for subgroup. Applying off-diagonal
reduction.

Conduct singular value decomposition of C matrix

Find linear scale factor for each wavelet by


solving C matrix, minimising the spectral misfits
Reduce amplitude of
Scale and sum wavelets to create adjustment function elements in δR except
that of diverging
Temporally add adjustment to total adjustment function spectral point
and check response

Is this the Yes Is the solution


No No
last subgroup? converging?

Is the
No mismatch peak
Yes
Add total adjustment within a half cycle of an
Yes
function to existing matched
accelerogram point?

Fig. 6. RspMatch2005 methodology including new solution algorithms (bold text).


April 28, 2006 13:47 WSPC/124-JEE 00273

80 J. Hancock et al.

before applying the wavelets adjustments. The issues related to selecting and scaling
accelerograms are beyond the scope of this paper but the reader is referred to
the following papers for guidance — and different perspectives — on this issue:
Watson-Lamprey and Abrahamson [2006a], Bommer and Acevedo [2004], Naeim
et al. [2004], Malhotra [2003]. The accelerograms in this paper have been selected
in accordance with the recommendations of Bommer and Acevedo [2004] who show
that distance has little influence on spectral shape and so they recommend a narrow
search window in terms of magnitude but allow broad limits in terms of distance.
Some engineers might consider the scale factors used in this paper are quite large;
however, a recent study by Watson-Lamprey and Abrahamson [2006b] found that
spectral matched accelerograms with scale factors of over a factor of 10 could be
used without causing a bias in nonlinear response.
For the illustrative example shown herein, a seed accelerogram has been selected
from the 3551 records of the PEER NGA dataset [PEER 2005]. Initial selection is
conducted based on an approximate match to the earthquake magnitude and the
spectral shape using the RMS of the difference in normalised spectral accelera-
tion (∆SAn RMS ), Equation (14). Other methods of matching spectral shapes are
possible: for example, the shape could be normalised to a high-frequency spectral
acceleration rather than PGA, or to the log of the normalised spectral acceleration.

 Np  2
 1  PSA0 (Ti ) PSAs (Ti )
∆SAn RMS =  − , (14)
Np i=1 PGA0 PGAs

where Np is the number of periods, P SA0 (T i) is the pseudo spectral acceleration


from the record at period Ti , P SAs (Ti ) is the target pseudo spectral acceleration
at the same period; P GA0 and P GAs are the peak ground acceleration of the
accelerogram and the zero-period anchor point of the target spectrum.
The record selected is the 1989 (Mw 6.9) Loma Prieta earthquake recorded
71 km from the fault rupture at Diamond Heights (record 00794T in the NGA
database). The code allows for scaling of the accelerogram to either PGA or a
selected scale factor. Here we have chosen to linearly scale by a factor of 3.2 so
that the difference between the record and target PSA and SD is minimised. For
engineering projects where high-frequency ground motion is of importance it is
appropriate to scale to PGA or a high-frequency spectral acceleration in order to
preserve the high-frequency characteristics of the ground motion.

5.2. Target spectra


The median 5% damped target spectra is generated according to the method pro-
posed by Bommer et al. [2000] using the peak ground motions proposed by Tromans
and Bommer [2002] notwithstanding the limitations associated with these formula-
tions arising from their derivation using analogue recordings [Boore and Bommer,
2005]. The target spectra for damping levels other than 5% are obtained using
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 81

the formulae derived by Bommer et al. [2000], which was subsequently adopted by
Eurocode 8 [CEN 2002]:

10
η= , (15)
5+β
where, η is the linear scale factor between the 5% damped response spectrum and
the required spectrum at β damping at intermediate periods. Equation (15) is a
simplification and is used herein only for illustrative purposes; the scaling of the
5%-damped spectral ordinates for higher damping levels has recently been shown
to be a function of the strong-motion duration [Bommer and Mendis, 2005; Mendis
and Bommer, 2006].

5.3. Matching 5% damped spectrum


The ground motion acceleration is adjusted so that it matches the target spec-
trum between 0.05 s and 5 s period (Fig. 7). Matching to longer periods is possible
but not conducted because the filter frequency of the seed accelerogram is 5 s;
indeed the useable frequency range will be less than the filter frequency (e.g. Akkar
and Bommer, 2006). The average spectral misfit between 0.05 s and 5 s period for
the 5% damping level has improved from 15% in the linearly scaled record to 1%
after wavelet adjustment with RspMatch2005. The average of the spectral misfit is
defined as:
Np  
1   PSAo (T i ) − PSAs (T i ) 
AverageMisfit =  ∗ 100. (16)
Np i=1  PSAs (T i )
Note that precisely the same average misfit is calculated if the PSA terms in
Eq. (16) are replaced with SD. When comparing results the misfit must be cal-
culated at closely-spaced periods, not just those used to conduct the spectral
matching.
Examination of the acceleration, velocity and displacement time-series before
and after the wavelet adjustment shows that the characteristics of the original
records have been retained (Fig. 8). Checks of the build up of Arias intensity also
demonstrate that the energy distribution within the record is similar to the original
ground motion and that the total energy content has been changed by less than
about 5% by the wavelet adjustment (Fig. 9).

5.4. Matching multiple damping levels


The ability of RspMatch2005 to adjust accelerograms to fit multiple damping levels
is investigated by running the program four times, fitting the accelerogram used in
the previous section to increasing numbers of damping levels (Fig. 10). Figure 10
shows that matching to the 5% damping spectra alone does not ensure a good
match at other damping levels. Although the match is not exact with four damp-
ing levels, the program has reduced the average spectral misfit at all damping
April 28, 2006 13:47 WSPC/124-JEE 00273

82 J. Hancock et al.

Fig. 7. 5% spectral acceleration (upper) and displacement (lower) of the target response (dashed
black line), original linearly scaled ground motion (solid grey line) and adjusted ground motion
(solid black line).

levels by a factor of about 3 (Table 1). This demonstrates RspMatch2005’s abil-


ity to match multiple damping levels, provided a reduced convergence tolerance
is accepted when matching increased numbers of damping levels. Examination of
the acceleration, velocity and displacement time-series before and after the wavelet
adjustment shows that the characteristics of the original record have been retained
(Fig. 11). Checks of the build up of Arias intensity also demonstrate that the energy
distribution within the record is similar to the original ground motion and that the
total energy content has been changed by less than about 10% by the wavelet
adjustment (Fig. 12).

6. Discussion
An improved method is presented for the wavelet adjustment of recorded ground
motions to achieve a match between the target design spectrum and the response
April 28, 2006 13:47 WSPC/124-JEE
00273

Fig. 8. Comparison of the acceleration, velocity and displacement time-series of the original linearly-scaled ground motion (grey line) and adjusted
Matching Response Spectra of Recorded Earthquake Ground Motion

ground motion (black line).


83
April 28, 2006 13:47 WSPC/124-JEE 00273

84 J. Hancock et al.

Fig. 9. Build up of Arias intensity from the original linearly-scaled accelerogram (grey line) and
adjusted ground motion (black line).

spectra of the accelerograms. New wavelets have been developed that have zero
final velocity and displacement, ensuring that records do not require a baseline
correction after wavelet adjustment. The procedure is applied using pseudo-spectral
acceleration so that spectral displacements can be matched. This method enables
records to be adjusted so that they match the target response spectrum at more
damping levels than previously possible, although the goodness-of-fit to the target
spectrum reduces as the number of target damping levels increases.
The option of adjusting real strong-motion recordings to achieve a match to the
target spectrum renders the use of artificial spectrum-compatible signals generated
from white noise redundant. The choices that remains then are to use natural
accelerograms scaled to achieve an approximate match to the target spectrum
over a specified period range or to adjust the records using the wavelets tech-
nique to achieve a close match with the target spectral ordinates. The latter option
reduces the variability of the inelastic response, which is particularly beneficial as
the number of accelerograms required to predict the response to a given confidence
level depends on the standard deviation of the response. This means that inelastic
response can be predicted with greater confidence and fewer analyses using accelero-
grams matched to the elastic response spectrum. Studies by Carballo [2000] and
Watson-Lamprey and Abrahamson [2006b] suggest that matched accelerograms can
reduce the standard deviation of the inelastic response by a factor of 2 compared to
linearly scaled accelerograms. This reduces the number of accelerograms to estimate
the inelastic response to a given confidence level by a factor of about 4.
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 85

Fig. 10. Spectral matching for different damping levels. Matched to 5% damped spectrum (top
row); 5 and 10% damped spectra (second row); 5, 10 and 20% damped spectra (third row); 5, 10,
20 and 30% damped spectra (bottom row). Pseudo spectral acceleration (left column) and spectral
displacement (right column).
86
J. Hancock et al.
April 28, 2006 13:47 WSPC/124-JEE
00273

Fig. 11. Acceleration, velocity and displacement time-series from original libearly scaled ground motion and that adjusted to match the 5, 10, 20
and 30% damping levels from 0.05 to 5 seconds period.
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 87

Table 1. Average spectral misfit between 0.05 and 5 seconds period for adjustment
conducted at different damping levels.

Damping Level
Damping level matched 5% 10% 20% 30% All
Original 14.8 12.7 8.7 6.6 10.7
Matched 5% 1.0 6.1 10.6 12.3 7.5
Matched 5 and 10% 2.8 2.3 6.3 9.2 5.2
Matched 5, 10 and 20% 4.7 2.8 2.1 3.6 3.3
Matched 5, 10, 20 and 30% 5.0 3.2 2.5 2.5 3.3

Fig. 12. Arias Intensity from original linearly scaled ground motion (grey line) and that adjusted
to match the 5, 10, 20 and 30% damping levels from 0.05 to 5 seconds period (black line).

Although using scaled natural accelerograms may be preferable in terms of


conserving the characteristics of real ground motion, using spectrally-matched
accelerograms has the advantage that the variability in spectral amplitude is greatly
reduced. If the target response spectrum has been obtained from probabilistic seis-
mic hazard analysis (PSHA), then the ground-motion variability will already be
incorporated into the ordinates of the target spectrum; using scaled natural records
can thus mean double counting of this aleatory variability.
The program is available on request from the corresponding author, provided
together with a user manual.

Acknowledgements
We would like to express our thanks to Luis Montejo for sending us his thesis and
journal publications and Luis Suarez for interesting discussions on this subject. We
April 28, 2006 13:47 WSPC/124-JEE 00273

88 J. Hancock et al.

would also like to acknowledge fruitful discussions with John Douglas and David
Boore on the possible methods of measuring the difference between recorded and
target spectral shape. The paper has benefited from the thorough reviews of Miguel
Castro and two anonymous reviewers, for which we are most grateful. The work
of the first and seventh authors is supported by doctoral training grants from the
EPSRC and Marie Curie Fellowships.

References
Abrahamson, N. A. [1992] “Non-stationary spectral matching,” Seismological Research
Letters 63(1), 30.
Akkar, S. and Bommer, J. J. [2006] “Influence of long-period filter cut-off on elastic spec-
tral displacements,” Earthquake Engineering and Structural Dynamics, accepted for
publication.
Boore, D. M. and Bommer, J. J. [2005] “Processing of strong-motion accelerograms:
needs, options and consequences,” Soil Dynamics and Earthquake Engineering 25(2),
93–115.
Bommer, J. J., Elnashai, A. S. and Weir, A. G. [2000] “Compatible acceleration and
displacement spectra for seismic design codes,” 12th World Conference on Earthquake
Engineering, Auckland, paper 207.
Bommer, J. J. and Acevedo, A. B. [2004] “The use of real earthquake accelerograms as
input to dynamic analysis,” Journal of Earthquake Engineering 8 (Special Issue 1),
43–91.
Bommer, J. J. and Mendis, R. [2005] “Scaling of displacement spectral ordinates
with damping ratios,” Earthquake Engineering and Structural Dynamics 34(2),
145–165.
Carballo-Arévalo, J. E. [2000] Probabilistic Seismic Demand Analysis Spectrum Matching
and Design, PhD Thesis, Stanford University, Stanford USA.
CEN [2002] Eurocode 8: Design of Structures for Earthquake Resistance. Part 1: General
Rules, Seismic Actions and Rule for Buildings, Draft No. 5, May 2002, Document
CEN/TC250/SC8/N317, Comité Européen de Normalisation, Brussels.
Kowalsky, M. J., Priestley, M. J. N. and McRae, G. A. [1995] “Displacement-based
design of RC bridge columns in seismic regions,” Earthquake Engineering & Structural
Dynamics 24(12), 1623–1643.
Lilhanand, K. and Tseng, W. S. [1987] “Generation of synthetic time histories compatible
with multiple-damping design response spectra,” Transactions of the 9th International
Conference on Structural Mechanics in Reactor Technology, Lausanne, K1, 105–110.
Lilhanand, K. and Tseng, W. S. [1988] “Development and application of realistic earth-
quake time histories compatible with multiple-damping design spectra,” Proceedings
of the 9th World Conference on Earthquake Engineering, Tokyo Japan, II, 819–824.
Malhotra, P. K. [2003] “Strong-motion records for site-specific analysis,” Earthquake
Spectra 19(3), 557–578.
Mendis, R. and Bommer, J. J. [2006] “Constructing over-damped displacement spectra
for seismic design codes,” 100th Anniversary Earthquake Conference Commemorating
the 1906 San Francisco Earthquake, EERI, San Francisco, paper 1743.
Mukherjee, S. and Gupta, V. K. [2002a] “Wavelet-based characterisation of design ground
motions,” Earthquake Engineering and Structural Dynamics 31, 1173–1190.
Mukherjee, S. and Gupta, V. K. [2002b] “Wavelet-based generation of spectrum-
compatible time-histories,” Soil Dynamics and Earthquake Engineering 22, 799–804.
April 28, 2006 13:47 WSPC/124-JEE 00273

Matching Response Spectra of Recorded Earthquake Ground Motion 89

Naeim, F., Alimoradi, A. and Pezeshk, S. [2004] “Selection and scaling of ground motion
time histories for structural design using genetic algorithms,” Earthquake Spectra
20(2), 413–426.
Naeim, F. and Lew, M. [1995] “On the use of design spectrum compatible time histories,”
Earthquake Spectra 11(1), 111–127.
PEER [2005] PEER NGA Database. http://peer.berkeley.edu/nga/index.html, Pacific
Earthquake Engineering Research, University of California, Berkeley.
Rizzo, P. C., Shaw, D. E. and Jarecki, S. J. [1975] “Development of real/synthetic time
histories to match smooth design spectra,” Nuclear Engineering and Design 32,
148–155.
Silva, W. J. and Lee, K. [1987] “WES RASCAL code for synthesizing earthquake ground
motions,” State-of-the-Art for Assessing Earthquake Hazards in the United Stated,
Report 24, Miscellaneous Paper S-73-1. US Army Corps of Engineers, Vicksburg,
Mississippi.
Suarez, L. E. and Montejo, L. A. [2003] “Generacion de registros artificiales compatibles
con un espectro de respuesta mediante la transformada wavelet,” Proceedings of II
Congreso Nacional de Ingenieria Sismica, Medellin.
Suarez, L. E. and Montejo, L. A. [2005] “Generation of artificial earthquakes via
the wavelet transform,” International Journal of Solids and Structures, 42(21–22),
5905–5919.
Tarantola, A. [2005] “Inverse problem theory and methods for model parameter estima-
tion,” Society for Industrial and Applies Mathematics, Philadelphia, ISBN 0-89871-
572-5.
Tromans, I. and Bommer, J. J. [2002] “The attenuation of strong-motion peaks in Europe,”
Proceedings of the 12th European Conference on Earthquake Engineering, Paper 394.
Watson-Lamprey, J. and Abrahamson, N. [2006a] “Selection of ground motion time series
and limits on scaling,” Soil Dynamics and Earthquake Engineering 26(5), 477–482.
Watson-Lamprey, J. A. and Abrahamson, N. A. [2006b] “Bias caused by use of spectrum
compatible motions,” 100th Anniversary Earthquake Conference Commemorating the
1906 San Francisco Earthquake, San Francisco, paper 0909.

S-ar putea să vă placă și