You are on page 1of 114

DRAFT # 1

Prepared by:
Ricardo Taborda

For:
M.I. Todorovska
M.D. Trifunac

UNIVERSITY OF SOUTHERN CALIFORNIA


DEPARTMENT OF CIVIL ENGINEERING

DAMAGE IDENTIFICATION USING DISCRETE WAVELET TRANSFORM

BY

R. Taborda, M.I. Todorovska and M.D. Trifunac

Report CE 04-0?

October, 2004
Los Angeles, California

http://www.usc.edu/dept/civil_eng/Earthquake_eng/

TABLE OF CONTENTS

ABSTRACT .........................................................................................................................................1

INTRODUCTION .................................................................................................................................2

THE WAVELET TRANSFORM ...........................................................................................................5


Discrete wavelet transform..................................................................................................................7
Packet wavelet transform ..................................................................................................................10
Properties and selection criteria ........................................................................................................11
Capability of wavelets on identifying signal singularities...................................................................15

DAMAGE IDENTIFICATION .............................................................................................................19


The occurrence of damage ...............................................................................................................19
Proposed parameters........................................................................................................................20

CASES OF STUDY ...........................................................................................................................25


The Van Nuys Seven-Story Hotel Building .......................................................................................25
The Imperial County Services Building .............................................................................................33

RESULTS ..........................................................................................................................................39
Results from the Imperial County Services Building .........................................................................39
Results from the Van Nuys Seven-Story Hotel Building ...................................................................51
Average normalized square coefficients of drifts ..............................................................................69
Maximum and global maximum level of damage indexes ................................................................83
Sequence of damage occurrence .....................................................................................................85
High Frequency Energy Indexes .......................................................................................................96
Index of damage intensity and its relation with other parameters.................................................. 101

FINAL COMMENTS ....................................................................................................................... 106


Relevance for system identification and soil-structure interaction ................................................. 106
Relevance for structural health monitoring and future application ................................................. 106

CONCLUSIONS ............................................................................................................................. 109

REFERENCES ............................................................................................................................... 110

ABSTRACT
Different alternatives of signal analysis using discrete wavelet transform for structural damage identification
are studied. These alternatives are tested on the signals recorded in two instrumented buildings which
suffered moderate and severe damage. It is shown that, although several of these alternatives seem to be
useful on generally detecting the occurrence of damage, it is the high frequency band component of interstory
drifts the one which leads to the clearest and more consistent results on identifying not only the occurrence of
damage but its time, sequence, location and intensity. Even though two cases of study are not enough for
making general conclusions, preliminary empirical limits that relate wavelet coefficients with the level of
damage in the structures are proposed as a very promising and realistic idea of remote online structural
damage monitoring.

INTRODUCTION
Design of earthquake resistant structures is a final objective of engineering that can only be achieved by the
careful study of realistic data and development of appropriate models for reproducing and understanding the
actual conditions that structures experience during seismic events and their performance. Such rigorous
analysis of data and models will lead us to take the right decisions about planning new and maintaining
existing structures.
As a part of this general objective, several authors have pointed out the importance of health monitoring and
damage identification, but this can not be achieved without the proper use of records from instrumented
structures. Correct and accurate identification of structural properties such as characteristic frequencies,
critical damping ratios or modal shapes, using signal processing techniques, have been widely studied, and,
although some times forgotten, it is becoming a matter of general knowledge that, these structural
characteristics, do suffer changes during the structures lifetime. The fluctuation of these properties in time,
through and between earthquakes, can be due to changes on the structure itself or because of natural behavior
of the whole soil-structure system components and the interaction between them.
Nevertheless, the mentioned changes are not always related with the occurrence of damage, thus the
importance of being able to detect whether damage is occurring or not and what is its relevance on the
structure. But damage detection is only a first step of a more general problem of damage identification, which
includes different stages as classification, severity, location, prediction and estimation of remaining service
life (Kim and Melhem, 2004).

On this matter, nondestructive evaluation techniques as both system

identification of modal parameters and classical spectral analysis present limitations. The former requires a
detail instrumentation of the structure to identify the location of damage, while the later has known
restrictions to detect time-dependent characteristics (Sun and Chang, 2002).
As an alternative approach, wavelet analysis can be seen as an extension of frequency-domain signal
transforms by means of which it is also possible to characterize their time-dependence properties as well as
identifying signal singularities. These two capabilities make of wavelet analysis a useful tool for the problem
of damage identification.
Suggestion of using wavelet analysis for damage detection in instrumented structures, in a global point of
view (i.e. the structure as a whole), was first presented by Rezai et al. (1994) based on the idea that
occurrence of damage is associated with the presence of a sudden higher frequency content in the signal,
which should be detectable in the form of peak values in the wavelet transform coefficients. Rezai et al.
(1994) analyzed the records of one of the instrumented buildings that are studied in this report, which was

damaged during two different earthquakes, using the Haar wavelet transform. Unfortunately, the results of
this work were not clearly conclusive because, as it will be shown later, the selection of the Haar wavelet was
not the most appropriate for this purpose.
An analogous idea was later presented by Sone et al. (1995) who approached the problem with a single degree
of freedom system which suffered fatigue causing degradation on the system stiffness. Vincent et al. (2000)
presented a similar problem for a multi degree of freedom in which the stiffnesses of the system were
suddenly reduced. Hou et al. (2000) did a comparable work for a single degree of freedom system with
multiple paralleled breakable springs and Kitada (1998) did it for non-linear single and multi degree of
freedom systems. All these works shown that wavelet transform excellently detects the moment in which the
change in the stiffness of the system occurs.
At the same time, interest was focused on wavelet analysis for damage identification from a local point of
view (i.e. study of elements instead of an entire system or structure). Wang and Deng (1998) applied the
wavelet transform on a set of spatially distributed structural response measures (spatial wavelets) achieving
detection and location of a crack in a simple supported beam. Ovanesova and Surez (2004) applied this
same concept for a simple frame theoretical example.
Corbin et al. (2000) applied slight modifications of the above mentioned approaches for three different cases:
1) the response of a three degree of freedom system with paralleled breakable springs; 2) the response of a
cantilever beam and; 3) the simulated response of a building. Sun and Chang (2002) presented the case of a
continuous simply supported beam in which, besides detecting the damage, this is quantified in relation with
the energy represented by the wavelet transform coefficients.
A more complete review of the works that have used wavelets on damage detection is out of the objective of
this report and it can be found elsewhere in the work by Kim and Melhem (2004). As a general conclusion, it
can be said that has been well demonstrated the potential capabilities of wavelet analysis on identifying
changes in a system response as those expected to be caused by the occurrence of damage. Nevertheless,
none of these works, except the one of Rezai et al. (1994) has applied this concept on real strong motion data
from instrumented structures that have suffered evident damage.
Based on the study of two buildings that have suffered moderate and severe damage, this report has the
objective of showing that the appropriate use of wavelet transform for signal processing analysis of strong
motion records from instrumented structures is no longer a potential tool but a reality that must start to be
used in damage identification. It is also the purpose of this work, given the obtained results of the two cases

of study, to propose preliminary empirical limits for characterizing the level of damage in a structure known
the results of a wavelet analysis of the recorded response.
In accomplishing these objectives, this report starts briefly reviewing the theoretical concepts of the wavelet
transform and the discrete wavelet transform. Properties of different wavelets and selection criteria for
selecting the most suitable wavelet are discussed and the concept of the relation between damage occurrence
and its identification by means of the wavelet decomposition is explained.
Descriptions of the two buildings studied and their damage are presented. Wavelet analysis of the signals
recorded in these structures is performed with several alternatives based on what conclusions are made about
which of these alternatives is the most appropriate for the purpose of damage identification. Once this
selection is made, all the records from different earthquakes are analyzed under the same parameters.
The analysis of these results leads to propose different parameters that are applied for the earthquakes in
which damage was detected and a more detail analysis of the sequence, location and intensity of damage
occurrence is shown and compared with what happened in the buildings.
Based on a comparison of the level of damage in these two cases throughout the different earthquakes and the
obtained results from the wavelet analysis, empirical relations and limits of damage intensity are proposed
and discussed.

Finally, a preliminary attempt on defining empirical limits that relate parameters estimated

from the high frequency band wavelet coefficients and the intensity of damage in the structures as well as a
futuristic view of what a remote online structural damage monitoring program could be are proposed.

THE WAVELET TRANSFORM


Wavelet analysis has its origin in the early work of mathematicians in the 1930s and even earlier if it is
noticed that the first wavelet basis, Haar wavelet, was introduced in 1910. But as it is known nowadays, the
wavelet transform concept was not formalized until the early 1980s after what its use has been continuously
growing and expanding through a wide variety of fields such as digital signal and image processing,
numerical analysis, physics and geophysics, mechanics and engineering. Since it is not the objective of this
work to review the history and development of the wavelet analysis and its applications the reader is
encouraged to review the works by Kumar and Foufoula-Georgiou (1994), Gurley and Kareem (1999), Misiti
et al. (2000), Kijewski and Kareem (2003) or Todorovska and Hao (2003).
As a simple approach for understanding the meaning of wavelets these may be conceived as localized waves,
i.e. wavelets are constrained signals or wiggles, with zero mean value, which drop to zero after a few number
of oscillations. Wavelets can be real or complex. The mentioned characteristics define wavelets as localized
both in time and frequency. Figure 1 shows an example of Gaussian wavelet in the time domain (top) and
frequency domain (bottom). These properties make of them a particularly useful tool for transient signal
analysis and time varying systems.
The first step in wavelet analysis is selection of a prototype wavelet, (t), which is called the mother wavelet.
Then, a family of wavelets is constructed by scaling (dilating or compressing) and time shifting the mother
wavelets, constituting a set of basis functions which may be expressed as follows.

a ,b ( t ) =

1
a

t b
,
a

a > 0, b \, t \

(1)

Where a and b are the scaling and time shifting factors, respectively. Within the family, the mother wavelet
has time shift b = 0 and scaling factor a = 1. It can be shown that scaling factor is inversely related with the
central frequency of the wavelet, but here it will just said that scaling factor a > 1 will cause dilation of the
mother wavelet, thus reduction of its central frequency and scaling factor a < 1 will be reflected as contraction
of the prototype wavelet thus higher central frequency.
The continuous wavelet transform of a signal f(t) will be defined by

C f ( a, b ) =

1
a

t b
f ( t ) a dt

(2)

1.2
0.8

y( t)

0.4
0
-0.4
-0.8
-1.2
-5

-4

-3

-2

-1

Time (s)

2.5

2.0

y( f )

1.5

1.0

0.5

0
0

0.2

0.4

0.6

0.8

1.0

1.2

Frequency (Hz)

Figure 1 - Mother Gaussian wavelet in time (top) and frequency (bottom) domains

where the bar over the wavelet function in equation 2 means complex conjugate of the shifted wavelet. The
numerical result of the wavelet transform of a signal will be a set of coefficients which represent how well
correlated are the signal and the wavelet. Wavelet transform of a signal, as it can be seen from equation 2
may be also understood as the inner product between the analyzed signal and the wavelet family.
It should be mentioned that, although not presented here, there exist an inverse wavelet transform by means of
which it is possible to recover the signal from its wavelet transform coefficients.

DISCRETE WAVELET TRANSFORM

It can be shown that the continuous wavelet transform of a signal in the full domain of Cf(a,b) is highly
redundant. As a consequence of this, for certain choices of the mother wavelets it is possible, without lose of
generality, to discrete the scaling and shifting factors as
a = 2j

b = k2j

(3)

where j and k are integers. This selection makes possible the definition of the discrete wavelet transform as

C j , k = 2 j 2

f ( t ) ( 2

t k ) dt

(4)

Assuming that there is a limit scaling factor such that the transform Cf(a,b) is only available for low values of
a < a0, then the rest of coefficients containing the information corresponding to Cf(a,b) for a > a0, must be
provided by the introduction of a scaling function, (t). If in addition to this, the limit value a0 is also discreet
as it was done before for a and b, and a reference level j is introduced, then the corresponding coefficients of
the transformation may be divided into

cD j ( k ) =

f ( t ) ( t ) dt
j ,k

(5)

cAj ( k ) =

f ( t ) ( t ) dt
j ,k

(6)

cDj(k) and cAj(k) are known as the detail and approximation coefficients. The described process is known as
subband decomposition because detail coefficients have the signal information about its high frequency
content while the approximation coefficients have the corresponding to the signal low frequency content.
This process can be iterated over the approximation subband, thus the subscript j refers the level of
decomposition. This iterative process is called multiresolution analysis. It was developed by Mallat and
Meyer (Daubechies, 1992), who first saw the connection between wavelet analysis and subband theory
(Mallat, 1989; Meyer, 1990). A more detailed study about the theory behind subband decomposition and
multiresolution analysis and its multiple properties and capabilities for strong motion signal processing may
be found in Todorovska and Hao (2003).
Reconstruction of the original signal from the discrete wavelet transform coefficients is also possible for some
selected wavelet basis and can be expressed as the sum of all detail reconstructed signals up to the Jth level of
decomposition and the Jth level approximation reconstructed signal as expressed in the following equation
f (t ) =
=
=
#
=
=

A1 ( t ) + D1 ( t )
A2 ( t ) + D2 ( t ) + D1 ( t )
A3 ( t ) + D3 ( t ) + D2 ( t ) + D1 ( t )

(7)
AJ ( t ) + DJ ( t ) + DJ 1 ( t ) + " + D2 ( t ) + D1 ( t )
J

AJ ( t ) + D j ( t )
j =1

where Aj(t) and Dj(t) are the corresponding reconstructed signals from the cAj(k) and cDj(k) coefficients.
A very useful and reasonable way of understanding the process of braking down the signal into detail and
approximation coefficients is thinking on it as both low and high pass filtering the signal. One of the most
important properties involved in this process is that discrete wavelet decomposition filters are such that they
subsample the signal in the process.
Figure 2 clearer illustrates this procedure by presenting an ideal view of a N samples signal in a (0,)
frequency domain passed by a three level of decomposition discrete wavelet analysis. Final result of the
transformation is composed by the coefficients corresponding to the shaded areas (i.e. A3, D3, D2 and D1).
Bottom part in figure 2 shows ideal subband frequency limits and the corresponding ideal central frequencies
of each subband.

N Samples

Original Signal Time-Frequency Domain

S = f(t)

LOWPASS

HIGHPASS

A1

D1

1st Level
of decomposition

Decomposition
Filters

N/2
Samples
LOWPASS

HIGHPASS

A2

D2

2nd Level
of decomposition

N/22
Samples

3rd Level
of decomposition

LOWPASS

HIGHPASS

A3

D3

N/23
Samples

N/23
Samples

A3
Ideal subband
frequencylimits

Ideal subband
central frequency

D3

23

24

D2

D1

22

3
24

3
23

3
22

Figure 2 - Ideal three level of decomposition analysis


of a N samples signal in a (0,) frequency domain

PACKET WAVELET TRANSFORM

Although final analyses in the present report are going to be centered in the use of the discrete wavelet
analysis, the packet wavelet transform would be eventually considered for alternatives comparison, thus the
importance of doing a briefly explanation of it in what follows.
The packet wavelet transform is not other thing but a generalization of the discrete wavelet decomposition. In
it, for each of the levels of decomposition both the approximation and detail subbands are split again in a next
level of decomposition, generating a new pair of detail and approximation subbands for each of the original
ones. Wavelet packet decomposition coefficients may be expressed as

C ij , k =

f ( t ) ( t ) dt
i
j ,k

(8)

where i, j and k are the modulation, scale and time shifting parameters of the wavelet packet defined by

ij , k ( t ) = 2 j k i ( 2 j t k ) ,

i = 1, 2,...

(9)

In equation 8, the wavelets i are obtained from the recursive relationships

2i ( t ) = 2 l ( k ) i ( 2t k )

(10)

k =

2i +1 ( t ) = 2 h ( k ) i ( 2t k )

(11)

k =

Note that both equation 10 and 11 are discrete convolution operations. Therefore, thinking on them in the
frequency domain, the terms l(k) and h(k) correspond to lowpass and highpass filters. These two filters are
quadrature mirror filters respectively associated with the scaling and mother wavelet functions.
It is appropriate to mention that when i = 1, the first of these equations correspond to the before called mother
wavelet.

1 (t ) = (t )

10

(12)

Figure 3 illustrates a three level of decomposition wavelet packet transform of a N samples signal in a (0,)
frequency domain.

As previously presented for the discrete wavelet analysis case, final result of the

transformation is composed by the coefficients corresponding to the shaded areas and in the bottom part of the
figure, the ideal subband frequency limits and their corresponding ideal central frequencies are shown.
As mentioned above, an important characteristics of the wavelet decomposition filters is that they subsample
the signal. When filtering a 100 samples signal, classical filters returns as result a new 100 samples signal
with the desired content of frequency. On the other hand, when using wavelet decomposition filters, the
result is two sets of 50 sample coefficients. If desired, each of these set can be reconstructed in two separate
100 samples signal. Nevertheless, in the coefficients time-frequency wavelet domain, each set is half samples
the original signal or the previous level of decomposition. This characteristic is in agreement with the
Heisenbergs principle of uncertainty which states that a function can not be well localized both in time and
frequency domains. This can be understood if looking at figure 3 is noticed that for narrower the frequency
bands (i.e. higher level of decomposition), lower the time resolution that can be derived from the coefficients
(i.e. less samples in the band).

PROPERTIES AND SELECTION CRITERIA

Some properties of wavelets may have important incidence in the results that are being sought. In addition to
this, even though there is a wide variety of wavelets basis, not all of them allow application of discrete
transform. As a result, just a reduced number of these basis are available for subband decomposition and
suitable for each specific purpose.
Making a classification of wavelets and their properties is out of the scope of this report, but being consequent
with the fact that this is a relevant part of the searched objectives, this section briefly reviews the existing
wavelets and their most relevant properties. What is next presented mainly follows the work by Ovanesova
and Surez (2004). A more rigorous review on the matter can be found elsewhere (Misiti et al., 2000).
Existing wavelets may be listed as:
Gaussian, Mexican Hat, Morlet and Shannon are those wavelets which have an explicit expression, (t),

but do not possess a scaling function, (t). As a result, the analysis using these wavelets is limited to the
continuous transform.

11

N Samples

S = f(t)

Original Signal Time-Frequency Domain

LOWPASS

HIGHPASS

A1

D1

1st Level
of decomposition

LOWPASS

HIGHPASS

LOWPASS

HIGHPASS

A2

DA2

AD2

DD2

2nd Level
of decomposition

3rd Level
of decomposition

LOWPASS

HIGHPASS

LOWPASS

HIGHPASS

LOWPASS

HIGHPASS

LOWPASS

HIGHPASS

AAA3

DAA3

ADA3

DDA3

AAD3

DAD3

ADD3

DDD3

N/23
Samples

N/23
Samples

AAA3
Ideal subband
frequencylimits

Ideal subband
central frequency

Decomposition
Filters

DAA3
2
24

24

N/23
Samples

ADA3
4
24

3
24

N/23
Samples

DDA3
6
24

5
24

N/23
Samples

AAD3
8
24

7
24

N/23
Samples

DAD3
10
24

9
24

N/23
Samples

ADD3
12
24

11
24

DDD3
14
24

13
24

Figure 3 - Ideal three levels of wavelet packet decomposition


analysis of a N samples signal in a (0,) frequency domain

12

N/23
Samples

15
24

Meyer wavelet is an infinitely regular wavelet that does not have an explicit expression for (t). Discrete

wavelet transform is possible because of the existence of the scaling function, (t), but fast wavelet
transform is not available.
Haar, Daubechies (of order N), Symlets (of order N), Coiflet (of order N) are orthogonal wavelets.

Except for Daubechies N = 1, which is the same Haar wavelet, these bases do not have an explicit mother
wavlet function, (t). The existence of (t) allows discrete wavelet transformation using fast wavelet
transform. Nevertheless, these wavelets have poor regularity and symmetry. Important characteristic of
orthogonal wavelets, as it will be seen further on, is that they satisfy the Parsevals equality.
Biorthogonal wavelets (of order Nr.Nd) overcome the limitations of the orthogonal wavelets. These bases

are compactly supported and symmetric. In addition, biorthogonal wavelets provide exact reconstruction
of the signal but two sets of wavelet and scaling functions, instead of one is needed. In their nomination,
the order Nr refers to the order of reconstruction while Nd corresponds to the order of decomposition.
The most relevant properties of wavelets may be summarized as follows:
Regularity: This property is related with the number of possible derivatives of the original function. It is

useful for achieving adequate smoothness in the reconstructed signals.


Support: The support of a function is the smallest time-set outside of which the function is identically

zero. Well supported wavelets are better in time-locating the properties of the original signal.
Number of vanishing moments:

As stated by its name, this number determines the order of the

polynomial that can be approximated to the original signal. It is useful for compression purposes.
Symmetry: this property indicates how much symmetric are the decomposition filters about its central

time. Symmetry is especially useful in avoiding signal dephasing.


Not a general property of wavelets but an important and exclusive characteristic of orthogonal bases is that
they satisfy the Parsevals equality. Explain in more detail by Todorovska and Hao (2003), for the case of
discrete subband decomposition, using the mentioned wavelets, this equality can be expressed as

t =0

f (t ) =

N 2J

k =1

N 2j

cAJ , k + cD j , k
2

(13)

j =1 k =1

where f(t) is the signal in the time domain, cAJ,k are the N/2J approximation coefficients at the Jth level of
decomposition and cDj,k are the N/2j detail coefficients at each one of the jth levels of decomposition (figure
2). This equality is also applicable for the case of orthogonal packet wavelet transform as

13

t =0

J
2J N 2

f (t ) = C j , k
2

(14)

j =1 k =1

Even though units of equations 13 and 14 do not represent energy, it is widely accepted that left hand side of
these expression is an equivalent representation of the energy content of the signal, which is usually
normalized by the time of the so called intense phase of the signal. Therefore, since k is related whit the
shifting in time factor b (equation 3), the importance of this equality because of the interpretation of the
squared coefficients as a time distribution of the energy content in the correspondent subband of
decomposition. Interesting relationships about these concepts, in terms of the signal power spectrum density
and the nodal time-frequency distribution are studied in further detail by Todorovska and Hao (2003). It is
convenient to notice that here when mentioned, the term squared coefficients will make reference to
2

cAJ , k
cD j , k

k = 1, 2,..., N

k = 1, 2,..., N

2J
;

2j

j = 1, 2,..., J

in the case of discrete wavelet transform, and

C j ,k

k = 1, 2,..., N

2J

j = 1, 2,..., 2 J

in the case of packet wavelet transform.


The selection of a particular wavelet mainly depends on the kind of analysis to be done and the sought results.
Nevertheless, this selection is usually done by trial and error or simply by a personal preference. The work by
Ovanesova and Surez (2004) cleverly pointed out that some of the properties of the wavelets should be
carefully seen by the time of making the final selection, particularly thinking in the problem of interest.
Otherwise, even having the right objectives and ideas to procedure, the erroneous selection of the basis for the
wavelet analysis may conducts into misleading results.
When introducing the original idea in which the present study is based on for damage detection, Rezai et al.
(1994) explained the selection of the Haar wavelet for their analysis because it was better for lower
accuracy approximation and for signals with sharp discontinuities. Unfortunately, as highlighted by
Ovanesova and Surez (2004), the irregularity of Haar wavelet is an obstacle for such a purpose.

14

In the cited work of Ovanesova and Surez (2004) a very reasonable and clearly justified criteria for selecting
the appropriate wavelet for damage detection purposes is proposed and here adopted. This criteria states as
follows:

1.

Gaussian, Mexican Hat, Morlet, Shannon and Meyers wavelets must be eliminated because do not allow
carrying out fast wavelet transform.

This, limits the possibilities to orthogonal and biorthogonal

wavelets.
2.

Symmetry and exact reconstruction must be also satisfied. Therefore, the selection is constrained to
Haar and biorthogonal wavelets.

3.

Because of lacking of regularity, Haar wavelet is discarded. Hence biorthogonal wavelets become the
most reasonable selection and, among them, the Bior6.8 wavelet has the highest order of reconstruction
and decomposition.

If not mentioned, results presented here will correspond to transformation using the Bior6.8 wavelet.

CAPABILITY OF WAVELETS ON IDENTIFYING SIGNAL SINGULARITIES

Before going into the damage detection assumptions to be used, it would be better to explain, from a practical
point of view, what is the capability of wavelets on identifying signal singularities. For this purpose, two
simple theoretical examples are presented.
Figure 4 shows the spectral and discrete wavelet analysis of a signal originally composed by two sine
functions with frequencies 1 and 15 Hz, which is later affected by a smoothed step function of amplitude 1 %
the amplitude of the longer period sine. Fourier amplitude spectra of both the original and the affected signal
are presented. On the other hand, in figure 5 the original signal is affected by a Gaussian modulated pulse,
again, of an amplitude 1 % the amplitude of the longer period sine. Both the step function and the pulse have
central frequency of 30 Hz and central time ~2 s.
These two figures clearly reveal advantages of wavelet analysis. Given the location and magnitude of the
introduced changes, signal comparison is worthless.
unknown.

Fourier analysis limitations are evident.

Besides, in real practice, the unaltered signal is


Not even the frequency is detectable, because of

unperiodicity and as a consequence of the small amplitude of the introduced changes if compared to the main
signal. As a matter of fact, the two spectra are by far the same.

15

TIME DOMAIN:
The signal and its components

FREQUENCY DOMAIN:
Fourier amplitude spectra

12
10
0
0.1
-12
0.001
Inclusion of a smooth step function

0.1

10
0.0
1.95

0.1

2.05

0.001
0.1
TIME-FREQUENCY DOMAIN:
Discrete wavlet transform - 1st level of decomposition

1
10
Frequency (Hz)

Squared approximation coefficients


300
Low
frequency
band

200
100

0 to 25 Hz
0
Squared detail coefficients

2 x 10-7

High
frequency
band

1 x 10-7

25 to 50 Hz
0
0

2
Time (s)

Figure 4 - Spectral and wavelet analysis of a sinusoidal signal


affected with a smoothed step function

16

100

TIME DOMAIN:
The signal and its components

FREQUENCY DOMAIN:
Fourier amplitude spectra

12
10
0
0.1
-12
0.001
Inclusion of a smooth step function

0.1

10
0.0
-0.02
1.95

0.1

2.05

0.001
0.1
TIME-FREQUENCY DOMAIN:
Discrete wavlet transform - 1st level of decomposition

1
10
Frequency (Hz)

Squared approximation coefficients


300
Low
frequency
band

200
100

0 to 25 Hz
0
Squared detail coefficients

2 x 10-7

High
frequency
band

1 x 10-7

25 to 50 Hz
0
0

2
Time (s)

Figure 5 - Spectral and wavelet analysis of a sinusoidal signal


affected with a Gaussian modulated pulse function

17

100

On the other hand, wavelet analysis, just in the first level of decomposition, perfectly locates the moment in
which the signal is affected and also gives information about the frequency band to what it belongs (detail
coefficients).
Although not presented here, it is obvious that a higher level of decomposition or a packet wavelet analysis
would expose further information about the composition of the signal and the characteristics of the introduced
functions.

18

DAMAGE IDENTIFICATION

Numerous studies have been done on the changes of structural properties due to earthquakes. As a result, it
has been widely demonstrated that this changes are reflected as fluctuation on general properties such as
natural frequencies or modal shapes of vibration of the structure. Generally speaking, such changes are
usually due to two principal sources, structural stiffness degradation and soil structure-interaction. It has been
also observed that over the time and even during the earthquakes these properties might be recovered because
of a rearrangement of the participating factors in the soil-structure system. Nevertheless, assignation of the
changes to specific causes being able of determining the location and magnitude of them, is still a challenge in
fields of study as system identification or health structural monitoring.
Classical techniques of system identification using Fourier transform analysis have known limitation as those
mentioned above regarding time domain lost of information and detecting signal singularities. On the other
hand, even though the considerable amount of literature on structural health monitoring, results are still not
convincingly applicable on real data because of typical mathematical models for representing the changes in
the structural systems are either too abstract to be related with structural components or useful only with
simple theoretical models. In addition to this panorama, complexity of the phenomena associated with soil
structure interaction makes even more difficult the pursued objective of identifying and locating damage in
the structure, affecting generally accepted indicators of damage such as interstory drifts (Trifunac and
Ivanovi, 2003).
On the effort of going beyond these current limitations, the explained properties and capabilities of wavelets
for signal processing will be applied in this report for identifying, locating and quantifying the presence and
magnitude of damage using real data from damaged instrumented structures. To do so, it is necessary to first
introduce the postulates in which the present work is based on and propose some of the procedure that will be
followed.

THE OCCURRENCE OF DAMAGE

Understanding of what happens in the structure once damage occurs will lead to a more suitable selection of
the process used to detect it and quantify it. From a seismological point of view, the sudden rupture of a
structural element could be considered as the rupture of a micro fault. Damage, then, will cause analogous
effects in the surrounding medium, i.e. a permanent displacement near the fault and generation of propagating
waves because of the rupture, but with the difference that, due to the oscillatory nature of the motion that
causes the rupture of this micro fault, these effects will rapidly disappear.

19

Figure 6 shows a representation of this phenomenon. A simple concrete frame is considered to be subjected
to an arbitrary signal that leads it into non linear behavior. After a certain period of time, the frame will start
to fail in the most demanded sections. Cracking of these sections will act as subsequent micro faults in the
elements causing a final main failure, involving in the process a release of energy which will be probably
manifested as temporal offsets in the response and high frequency waves due to the micro faults rupture.
Assuming that this representation adequately explains what occurs in real structures, the two described
manifestations should be expected to be detectable, but because of oscillation of the whole system, they
would also be expected to vanish rapidly in the total response.
It would be assumed that the mentioned waves will be clearer noticeable in the system acceleration response,
while the temporal offset should be clearer detectable in the system displacements. Both of them are expected
to be high frequency related. Generated waves should the kind of singularity analyzed in figure 5 while
offsets should be similar to the theoretical example presented in figure 4.
The previous description leads to believe that wavelet analysis is the appropriate tool to catch the occurrence
of damage, its time location and magnitude.

PROPOSED PARAMETERS

For the results that will be presented herein it has been necessary the adoption of empirical parameters to
facilitate its analysis. Such parameters and their intuitive interpretation are explained in what follows.

High Frequency Energy Index

Although the wavelet basis used in this report is not an orthogonal wavelet, it is assumed that the sum of its
squared coefficients is a representative value of the energy contained in the frequency band of analysis. For
the high frequency band of a discrete wavelet analysis, this parameter is formally defined as
N 2

EHf =

cD

1, k

k =1

(15)

where cD1,k are the detail coefficients of the first level of decomposition T is the time duration of the signal.
Given that the sum of the square coefficients is a quantity depending on the band number of coefficients (N/2)

20

Appearence of
cracks and rupture
of elements generating signal singularities

Evidence of damage
in the signal is rapidly
hidden by the oscilatory movement of the
whole structure

Time

Figure 6 - Representation of damage occurrence and its effect in the response of the signal

21

which is also proportional to the record length, the value of T acts as a normalizing factor for comparison of
different earthquake records.

Global Average and Global Maximum High Frequency Energy Index

Since the structures analyzed have multiple instruments, it will be necessary to have a global idea of the level
of the high frequency energy reached in the whole structure. For this purpose the global average and the
global maximum of EHf are respectively defined as

aEHf =

Hf

(16)

mEHf = max { EHf ,i ,..., EHf , I } ,

i = 1, 2,..., I

(17)

where I is the total number of recorded signals (instruments) considered in the analysis.

Average Normalizing Factor

This factor may be also understood as the intrinsic average level of high frequency content in the signal and is
formally defined as follows.
N 2

aCHf =

cD
k =1

1, k

N 2

(17)

It must be noticed that, since it is expected that the peak values of the squared coefficients that will be
associated with the damage occurrence are singular isolated values in the sample, their weight in the average
of all the band coefficients may be expected to be minimum, making of this quantity a good factor for
comparing different signals and normalizing other parameters.

Maximum Squared Coefficient

As it is obviously derived from its name, this parameter is the maximum value in a sample of squared
coefficients. The sample may be composed by all the coefficients in the high frequency band or by a selected
range of coefficients, for a time windowed analysis.

22

2
cDmax
= max cD1, ki ,..., cD1, k f

(18)

cD1,ki and cD1,kf are the initial and final coefficients in the sample. If the entire signal is considered then ki = 1
and kf = N/2.

This parameter provides a first idea of the maximum level of damage reached in the

correspondent signal.

Global Maximum Squared Coefficient

This quantity corresponds to the maximum value of the sample composed by the cD2max from all the signals
under analysis in the structure.
2
2
Cmax = max {cDmax,
i ,..., cDmax, I } ,

i = 1, 2,..., I

(19)

This parameter provides a first idea of the maximum level of damage reached in the structure.

Maximum Level of Damage Index

The maximum level of damage index will be defined as the maximum squared coefficient in a sample of
values divided by the average normalizing factor.

lmax =

max cD1, ki ,..., cD1, k f

aCHf

(20)

Normalization of the peak value of the coefficients seeks to eliminate the intrinsic characteristic frequency
content of each different signal in order to be able of performing conclusive comparisons. Depending on the
definition of ki and kf, lmax may be calculated for the entire signal or for a time window.
Global Maximum Level of Damage Index

This index is the maximum value of the estimated values of lmax for the signals under analysis. It gives a
global idea of the reached level of damage in the whole structure. As cD2max and lmax, it can be also estimated
for signal windows by simultaneous analysis of all considered lmax.
Lmax = max {lmax,i ,..., lmax, I } ,

23

i = 1, 2,..., I

(21)

I is the total number of recorded signals (instruments) considered in the analysis.

Index of Damage Intensity

Knowing the complexity of defining a scale of intensity, it will be roughly assumed that it is possible to group
the damage intensity in the structure according to the index scale shown in table 1.

Table 1 - Index of damage intensity definition


Index

Definition

No damage in the structure.

No visible damage in the structure. Damages could have occurred but is covered or
it was simply not detected or reported.

Moderate damage. Damage is visible. Cracks of considerable width may have


occurred. Previous existent cracks may have been enlarged.

Severe damage. Presence of large cracks. Damage has affected main structural
elements but the structure is stable.

Near collapse. Occurred damage has seriously affected the structural system and it
may be instable.

Collapse.

24

CASES OF STUDY

The previous postulates of how damage occurrence may be identified through wavelet analysis and the
different parameters proposed to characterize it have been tested in two different structures which will be
described in what follows. For each case a description of the building and its instrumentation are covered as
well as a description of the damaged suffered in each of the cases. Main characteristics of the available
records are also included.

THE VAN NUYS SEVEN-STORY HOTEL BUILDING

Description of the Building

The Van Nuys Seven Story Hotel, here referred as VN7SH, is a reinforced concrete building located near the
intersection of Roscoe Blvd. and San Diego Freeway (I-405) in the city of Van Nuys of the metropolitan area
of Los Angeles. Designed in 1965 (Blume et al. 1973) and constructed in 1966, VN7SH building has a total
floor area of about 5,900 m2 and plan dimensions of about 46 19 m. Figure 7 shows the typical floor and
foundation plan of the building. Figure 8 present the typical elevation.
Excepting two small areas at the ground floor, the structure of the building is essentially symmetric. Typical
framing consists of columns spaced 6.3 m on centers in the transverse direction and 5.7 m in the longitudinal
direction. Lateral resistance in each direction is provided by the interior column slab frames and the exterior
column spandrel beam frames. Spandrel beams surrounding the perimeter of the structure provide additional
stiffness conforming exterior frames that are roughly twice as stiff as interior frames. The floor system is a
reinforced concrete slab with different thickness depending on the floor (Figure 8). The north side of the
building, along column line D (Fig. 2a), has four bays of brick masonry walls located between the ground and
the second floor at the east end of the structure. The structure is constructed of regular-weight reinforced
concrete (Blume et al., 1973).
The VN7SH building lies on recent alluvium primarily composed by fine sandy silts and silty fine sands. The
average shear-wave velocity in the top 30 m is 300 m/s. The foundation system shown in figure 7b consists
of 97 cm deep pile caps supported by groups of two to four poured-in-place 61 cm diameter reinforced
concrete friction piles which are centered under the main building columns and are roughly 12 m long. Pile
caps are connected by a beams grid. Each pile has a capacity of over 690 MPa vertical load and up to 138
MPa lateral load.

25

(a)
1

19.10 m

A
8 bays @ 5.715 m = 45.72 m

(b)
1

19.10 m

8 bays @ 5.715 m = 45.72 m

Figure 7 - (a) Typical floor plan and (b) foundation plan

26

A
20.3 cm slab

7th Floor

21.6 cm slab

6th Floor

21.6 cm slab

5th Floor

21.6 cm slab

4th Floor

21.6 cm slab

3rd Floor

21.6 cm slab

2nd Floor

25.4 cm slab

1st Floor

10.2 cm slab

4.11 m

Roof

6.35 m

6.35 m

6.35 m

19.10 m

Figure 8 - Typical elevation

27

20.03 m

2.64 m

5 x 2.65 m

Instrumentation and Strong Motion Records

Instrumentation of the VN7SH building is operated by the Strong Motion Instrumentation Program of the
California Division of Mines and Geology (CDMG).

First digitally available strong motion records

correspond to the 1971 San Fernando earthquake. This event was recorded by three self-contained triaxial
AR-240 accelerographs located in the building as shown in figure 9a. All posterior earthquakes were
recorded by a 13 channel CR-1 central recording system and one tri-component SMA-1 accelerograph with
independent recording system, but common trigger time with the CR-1 recorder. Location of all the 16
channels is shown in figure 9b.
The largest strong motion recorded so far in the building is the 1994 Northridge earthquake. Table 2 lists the
most relevant available records, their magnitudes, M, and epicentral distances to the building, R. Table 2 also
includes an abbreviation that will be eventually used. The accelerograms of the Whittier Narrows, Landers,
Big Bear, and Northridge earthquakes (Shakal et al., 1994) were processed by CDMG. Records of other
earthquakes listed in Table 2 were processed at USC (Trifunac and Lee, 1979; Lee and Trifunac, 1990) from
copies of original recordings supplied by CDMG.

Table 2 - Digital records available in the VN7SH building


No.

Earthquake

Abb.

Date

R (km)

San Fernando

SFR

02/09/1971

6.6

22

Whittier Narrows

WTN

10/01/1987

5.9

41

Whittier Narrows aftershock

WTA

10/04/1987

5.3

38

Pasadena

PSD

10/03/1988

4.9

32

Montebello

MTB

06/12/1989

4.1

34

Malibu

MAL

01/19/1989

5.0

36

Sierra Madre

SMD

06/28/1991

5.8

44

Landers

LAN

06/28/1992

7.5

186

Big Bear

BBR

06/28/1992

6.5

149

10

Northridge

NOR

01/17/1994

6.5

1.5

11

Northridge aftershock (392)

N03

03/20/1994

5.2

1.2

12

Northridge aftershock (436)

N12

12/06/1994

4.5

10.8

28

(a)

8
9

5
4
6

(b)

3
9
4

13.24 m

5.29 m

10
1.80 m
2.65 m

4.11 m

6
11

15
16

12
14

7.95 m

8
2.65 m
4.11 m

13
40.0 m
8.10 m

Figure 9 - (a) Instrumentation at the moment of 1971 San Fernando earthquake and
(b) the sensors that have recorded posterior earthquakes

29

SAN FERNANDO VALLEY


35 25' N
San Fernando, 1971

Epicenter

2 km

Northridge, 1994:
6 Dec. Aftershock
Northridge, 1994:
Main Event

2 mi

2 sec

Newhall

3
4

5 sec

6
4
7
3

8 sec
9
2

118

22

-42

SITE

Big Bear, 1992

Canoga Park

35 k
m

Reseda

Mal
ibu,

Encino

Sherman
Oaks

198
9

101

118 39' W

Pa sa
de na
,

CH 13

CH 1
Northridge, 1994:
20 Mar. Aftershock

210

km
, 19 91 44
Sierra Madre
Landers, 1992 186 km

-90

34 09' N

v = 2 km/s

62

on

te

170 Whi
ttie
r,
be

llo

,1

98

32 km

Burbank
198
7

41
k

34

405

19 88

149 km

km

118 18' W

Figure 10 - Location of the VN7SH building relative to the 1971 San Fernando
and 1994 Northridge earthquake faults and other earthquakes considered

30

Figure 10 (simplified from Trifunac, 1974) shows the location of the building relative to the faults of the 1971
San Fernando and 1994 Northridge earthquakes. It also shows the epicenters of two Northridge aftershocks,
and the direction of wave arrival for other earthquakes here considered.

Damage from the 1971 San Fernando Earthquake

Damage in the structure of the building caused by this earthquake was minor. It mainly consisted of cracks in
the spandrel beam-to-column connection at the north-east corner of the building and damage at the partitions
and exterior plaster. Spalled concrete of the second floor beam-column joints on the north side and east end
of the building was repaired using epoxy. Main presence of damage was concentrated on the second and third
floors while minimum damages were observed at the sixth and seventh floors. Nonstructural damage was
extensive and it was necessary to repair drywall partitions, bathroom tiles and plumbing fixtures (Blume and
Assoc., 1973).
Peak horizontal acceleration at the ground level during the 1971 San Fernando Earthquake was 0.25 g, and
peak horizontal velocity was 27 cm/s.

Damage from the 1994 Northridge Earthquake and Early Aftershocks

Between 1971 and 1994, the building was shaken several times, but no visible damage was detected. The
largest peak acceleration in this period was 0.16 g, recorded during the 1987, Whittier Narrows earthquake.
And the largest velocity was 11 cm/s, recorded during the 1992, Landers earthquake (Trifunac et al., 1999).
During the January 17, 1994 Northridge earthquake, the peak ground acceleration was 0.44 g, and the peak
velocity was 51 cm/s (EW component).
Main event of Northridge earthquake caused severe structural damage in the building. Figure 11 shows a
schematic representation of this damage. South and north frames A and D, designed to take most of the
lateral loads in the EW direction, were seriously affected. Severe shear cracks occurred at the middle
columns of frame A near the contact with the spandrel beam of the fifth floor, significantly decreasing the
axial, moment and shear capacity of these elements. The shear cracks that appeared in the north frame (D),
on the third and fourth floors, and the damage of columns D2, D3, and D4 on the first floor, caused minor to
moderate changes in the structural capacity of the elements.
There was no visible damage to the interior longitudinal frames (B and C), although some cracks could have
been hidden by large furniture. Excepting small cracks in the slab around the central columns on 5th and 6th

31

(a)

FRAME D (North view)


9

1
2

3
Roof

7th Floor
4
6th

"x" shear cracks

Floor

<0.5 cm

4th Floor

cracks through
the beam
0.5 - 1.0 cm

"x" shear cracks


<0.5 cm

5th Floor
"x" shear cracks
<0.5 cm

cracks
along the column,
0.5 cm

"x" shear cracks


<0.5 cm

6
3rd Floor

2nd Floor

13

diagonal cracks
in beams
<0.5 cm

diagonal cracks
along the column,
0.5 cm

cracks through
the beam
0.5 - 1.0 cm

7
'short column' cracks,
0.5 - 1.0 cm

cracks between bricks

cracks between bricks

1st Floor

(b)

FRAME A (South view)


1

9
3

Roof

7th Floor
" x " shear craks
~5.0 cm

"x" - shear cracks ~ 5.0 cm ,


bending of long
reinforcement

6th Floor

"x" - shear cracks ~ 5.0 cm ,


bending of long. and trans.
reinforcement

5th Floor
"x" - shear cracks > 5.0 cm ,
bending of long.
reinforcement

4th Floor
5

"x" - shear cracks ~ 5.0 cm ,


bending of long.
reinforcement

cracks in the beam


< 1.0 cm

3rd Floor
7
2nd

cracks in the beam


< 1.0 c m

Floor

complete first floor on the south side was covered - there may have been some damage there

1st Floor

13

Figure 11 - Schematic representation of damage following the 1994 Northridge earthquake

32

floors, no damage was observed in the reinforced concrete slabs. There were no signs of large deformations
in the foundation, neither at the 1st floor slab nor in the pavement around the building. In the east side of the
building, minor nonstructural cracks were seen, mainly along the first floor and in general, nonstructural
damage was significant. Almost every room suffered considerable damage. There were numerous cracks in
the bathtubs, and many ceramic tiles fell off. Severe cracks were noticed in the masonry brick walls and in
the exterior cement plaster.
After the main event of Northridge earthquake, the building was restrained to support the weakened structural
elements. Wooden braces were placed at selected bays of the exterior longitudinal frames. After the March
20 aftershock large shear cracks, with deformations of the longitudinal reinforcement larger than those
observed after the main event were noticed in the frame A. No additional damage was noticed in frame D nor
in the interior frames and no significant change was noticed regard to nonstructural damage.
A more detailed report of the damage is out of the objective of the present work and is better presented
elsewhere (Trifunac et al., 1999; Trifunac and Hao, 2001).

THE IMPERIAL COUNTY SERVICES BUILDING

Description of the Building and its Instrumentation

The Imperial County Services building, here referred as ICS, was a six-story reinforced concrete structure in
the El Centro area. At the time of its operational life the surrounding area was primary a farming community.
As a result, the building was one of the major structures in the zone.
The ICS building was designed according to the considerations of the 1967 Uniform Building Code, and was
finished in 1969. Its plan dimensions were 41.70 26.02 m (figure 12). Height of the building was 25.48 m.
Foundation system of the building was composed of piles groups and pile caps directly located under the
columns and walls ends. Pile caps were also connected each other by ground-level beams. Underlying soil
consisted of soft to medium-stiff damp sandy clay with organic materials and with inter-layers of medium
dense moist sand from 0 to 9 m, and stiff, moist sandy clay and silty clay beneath 9 m.
Structural elements were made of reinforced concrete with minimum ultimate compressive strength of 27.6
MPa for walls, beams and slabs, 34.5 MPa for columns and 20.7 MPa for the foundation elements.
Reinforcement steel was 276 MPa. Structural configuration in the NS (transverse) direction was composed

33

38.1 m
7.62 m

7.62 m

7.62 m

7.62 m

7.62 m

7.62 m

7.62 m

22.86 m

7.62 m

1
Foundation and Ground Level

41.7 m
1.80 m

1.80 m

1.58 m

26.0 m

1.58 m

Typical Floor Layout

Figure 12 - Foundation and ground level plan and typical floor layout of the ICS building

34

4.01 m

Roof

5th Floor

4 @ 4.11 m

25.5 m

6th Floor

4th Floor
3rd Floor
"Free Field"
Accelerometers

2nd Floor

5.03 m

Ground Floor

41.7 m
104 m

E-W Section

41.7 m
INSTALATION NOTES:

26.0 m

- Accelerometers 1 through 4 attached


to the underside of roof slab.

- Accelerometers 5 through 13
attached to topside of slabs.

3
4

- Horizontal starter adjacent


and parallel to accelerometer 4.
- Vertical starter adjacent to triaxial
package on ground floor.

Roof Plan

5
"Free Field"
Accelerometers

4th Floor Plan

F3
F2
10

11

12

2nd Floor Plan

3m

13

Ground Floor Plan

104 m

Figure 13 - ICS building instrumentation

35

F1

by two concrete panels at the east and west ends of the building (figure 12). These panels extended only from
the second floor level to the roof, and were supported by cantilever parts of the frame beams which extended
in the EW direction. At the ground level, four panels were located between axis 2 and 3 along lines A and C
through D. In the EW (longitudinal) direction, the structural system was composed by four beam-column
frames. Faade columns were cross-sectional variable, changing from rectangular to trapezoidal shape at the
second floor level.
Seismic instrumentation of the building consisted on thirteen force balance accelerometers (FBA-1) and a
free field triaxial accelerometer (SMA-1). Figure 13 shows orientation and location of these instruments.
Accelerometers 1 through 4 were attached under the roof slab. Accelerometers 5 through 13 were at the
topside of floor slabs. The recording system was ruled by a horizontal starter adjacent and parallel to
accelerometer 4, and by a vertical starter adjacent to triaxial arrange on the ground floor. The free field
triaxial accelerometer was approximately located 104 m east from the north-east corner of the building.
Further information on the structural system and the building instrumentation may be found in the work by
Koji et al. (1984).

Damage from the 1979 Imperial Valley Earthquake

On October 19, 1979 Imperial Valley, California suffered the occurrence of a ML = 6.6 earthquake originated
in the Imperial Fault near to El Centro area. The earthquake had epicenter at a depth of 8 km and about
26 km southeast from the building, and it was excellently recorded by the 16 channels network of the
building. Peak accelerations in the NS directions were 571 cm/s2 and 339 cm/s2 at the roof and the ground
floor levels, respectively. In the EW direction these values were 461 cm/s2 and 331 cm/s2. Records were
digitized at USC (Trifunac and Lee, 1979).
During this earthquake, the building resulted seriously damaged and as a result it had to be later demolished
(Koji et al., 1984). Figure 14 shows a schematic representation of the main damage in the building.
Principal failures took placed at the ground level of the east frame (F) columns. Vertical reinforcement was
exposed and buckled and the core concrete could not be retained, resulting in a shortening of the columns and
causing cracking of the floor beams and slabs near column line F on the second, third and higher floors.
Columns in lines A, B, D and E also suffered damage. Columns in frames A and E did not suffer such
extensive damage as shortening and buckling of the bars of the east side columns, but large concrete cracks
and exposed reinforcing bars could be seen near the base. Columns in interior frames B through E had visible

36

Roof

6th Floor

5th Floor

4th

Shortening of line F columns


caused cracking of floor beams
and slabs between column
lines E and F on the second,
third and higher floors

Floor

3rd Floor

2nd Floor
Ground Floor
Large concrete cracks,
spalling of concrete
cover and exposed
reinforcement

Large concrete cracks


and spalling of the
concrete cover

Vertical reinforcement
exposed and buckled.
Concrete core was
not retained

Figure 14 - Schematic representation of damage in the ICS building


following the 1979 Imperial Valley earthquake

37

cracking and spalling of the concrete cover. Further description of the damage and the inelastic response of
the building may be sought in Koji et al. (1984).

38

RESULTS

Different alternatives were considered for analyzing the records from the VN7SH and the ICS buildings using
the discrete and packet wavelet transforms. In this section the most significant results are presented and
relevant conclusions are taken in order to go into the analysis of the most meaningful alternative. It must be
mention that, although some of these alternatives are not generally or consistently conclusive, they are still
presented because they serve on defining the method limitations and scopes.
First part of this section is centered in the analysis of records from the ICS building and the identification of
the occurrence of damage during the 1979, Imperial Valley earthquake. As it will be seen, because of its
severity, damage in the ICS building results so obviously detectable that it makes of the analysis a redundant
exercise. Nevertheless, it serves to confirm the assumptions made about the capabilities of wavelet analysis
on detecting signal singularities on real records. Second part deals with the analysis of records from the
VN7SH building. Although it is principally centered on the study of the main event of the 1994, Northridge
earthquake, other records from events where no damage was detected are used for testing these results.
Conclusions made from the results of both of the buildings are used on defining a final criterion for analyzing
remaining records and finally going into detail on detecting, locating and quantifying the damage. Final
sections study the appearance time of the damage, its sequence of occurrence, distribution in the structure and
magnitude, and analyzes the different results obtained from the proposed parameters, the relations between
them and their relation with the actual damage observed in the structures suggesting the possibility of defining
empirical relations between the results form wavelet analysis and index of intensity defined.

RESULTS FROM THE IMPERIAL COUNTY SERVICES BUILDING

Absolute Accelerations

Figure 15 shows the horizontal accelerations recorded in the building during the 1979, Imperial Valley
earthquake. The available signals were digitized at 200 samples per second and band pass filtered between
0.125 and 25.0 Hz. Peak values in the roof and the ground floor were 571 cm/s2 and 339 cm/s2 in the NS
direction, and 461 cm/s2 and 331 cm/s2 in the EW direction. Figure 16 shows the square coefficients
corresponding to the detail band of the first level of decomposition (12.5 to 25.0 Hz) for records in (a) the NS
direction and (b) the EW direction.

39

ACCELERATION (cm/s2 )
0.125 to 25.0 Hz

NS DIRECTION
600
300
0
-300
-600

CH 1

CH 2

CH 3

CH 7

CH 8

CH 9

CH 10

CH 11

EW DIRECTION
600
300
0
-300
-600

CH 4

CH 5

CH 6

CH 13

10

15

20

Time (s)

Figure 15 - Horizontal accelerations from the 1979, Imperial Valley earthquake

40

SQUARE COEFFICIENTS (cm 2 /s4 )


12.5 to 25.0 Hz

NS DIRECTION
5000
2500

CH 1

0
4000
2000

CH 2

0
5000
2500

CH 3

0
4000
2000

CH 7

0
2000
1000

CH 8

0
60000
30000

CH 9

0
1200
600

CH 10

0
1600
800

CH 11

0
4

10

12

Time (s)

Figure 16 - (a) Acceleration detail square coefficients


from the first level of decomposition in the NS direction

41

14

SQUARE COEFFICIENTS (cm 2 /s4 )


12.5 to 25.0 Hz

EW DIRECTION
700
350

CH 4

0
400
200

CH 5

0
500
250

CH 6

0
700
350

CH 13

0
4

10

12

Time (s)

Figure 16 - (b) Acceleration detail square coefficients


from the first level of decomposition in the EW direction

42

14

Shaded regions in figure 16 correspond to a first selection of regions where peak values are consistently
present in the signals. These regions are better defined in the NS than in the EW direction. The highest value
is in channel 9 about 11 s. It corresponds to the accelerograph just above the main damage of the ICS
building (figures 13 and 14), and it is also directly detectable in the records (figure 15).
In the NS direction, the first selection goes from 5.8 to 7 s and reaches its highest values in channels 7 and 2,
which are in the west side of the second floor level and in the middle of the roof, respectively. Second
interval is located between 7.8 and 9.2 s and its highest value comes from cannel 1 at about 8.9 s. The third
region in the NS component goes from 10.5 to 11.5 s. It is maybe the clearest defined because of the
appearance of the former mention peak value at 11 s. Except for the range between 10.8 and 11.7 s, which
clearly corresponds to the same range in the NS component, shaded regions in the EW direction are not so
clearly defined.
A rough analysis of this results suggest that damage of the building started in the west and interior frames at
the ground level (channels 7 and 8) at about 6 s, and were later on migrating to the east side of the building
until causing the major failure of ground level columns in the east (D) frame. Nevertheless, it must be noticed
that these results have some inconsistencies with the observed damage in the building. Square coefficients
from channel 1 and 2 suggest the occurrence of something at the west side and center of the roof and none of
the observations agree with this result neither channel 3 corroborate this. Besides, results from channel 13
seem to be affected by the high frequency content of the signal making difficult the identification. It must be
also pointed out that amplitude scale varies considerably from one channel to another. In a same scale figure,
magnitude reached by channel 9 would have clouded other channel results.
Figure 17 shows the square coefficients of the highest frequency band at the second level of decomposition
using packet wavelet transform (i.e. DD2 as showed in figure 3), which theoretically correspond to the
frequency band from 18.75 to 25 Hz. Although going into a second level of decomposition means loss of
time resolution some positive effects may be noticed. Shaded regions are narrower defined and a smoothness
effect is seen in channel 13.
Nevertheless, while in channel 13 this smoothness is a positive effect on hiding high frequency content of the
ground level signal, in channels 4 an 5, amplitudes of peaks about 11 s are significantly reduced, what in other
cases, where damage is not so evident as here, this could lead to loss of detectability. Besides, generally
speaking, no new conclusions can be made different of what mentioned before. Channel 1 and 2 continue
showing inconsistent results relative to the magnitude of the values reached in other channels.

43

SQUARE COEFFICIENTS (cm 2 /s4 )


18.75 to 25.0 Hz

NS DIRECTION
5000
2500

CH 1

0
5000
2500

CH 2

0
4000
2000

CH 3

0
4000
2000

CH 7

0
2000
1000

CH 8

0
60000
30000

CH 9

0
2400
1200

CH 10

0
2000
1000

CH 11

0
4

10

12

14

Time (s)

Figure 17 - (a) Acceleration square coefficients of the highest frequency band at the second
level of decomposition using packet wavelet transform in the NS direction

44

SQUARE COEFFICIENTS (cm 2 /s4 )


12.5 to 25.0 Hz

EW DIRECTION
300
150

CH 4

0
200
100

CH 5

0
600
300

CH 6

0
1600
800

CH 13

0
4

10

12

14

Time (s)

Figure 17 - (b) Acceleration square coefficients of the highest frequency band at the second
level of decomposition using packet wavelet transform in the EW direction

45

Absolute Velocities and Displacements

Figures 18 and 19 show horizontal absolute velocity and absolute displacements records, respectively.
Figures 20 and 21 show the detail square coefficients from the first level of the correspondent discrete wavelet
transform of the signals of the NS direction. EW records were also analyzed but not relevant information was
seen different from the NS component. Excepting the case of channel 10 in velocities, same scale was use for
the figures.
Shaded circular areas highlight the most noticeable peaks in the square coefficients. It can be seen that
absolute velocity and absolute displacement coefficients leads to better located results than those seen from
acceleration records. Main times of damage occurrence are within those intervals observed in the acceleration
records, thus preliminary conclusions do not vary significantly.
Displacement results seem affected by some level of high frequency content in the signals, but inconsistent
peak value from channel 1 is no more present in the results. For the case of velocities an incomprehensible
high value is reached in channel 10 and, although damage occurred near this channel, it should not be
expected from other different to channel 9 to produce such a value. Nevertheless, velocity and displacement
results are, in general, seem to be alternatives that lead to clearer detectable results than acceleration.

Interstory Drifts

Results from all accelerations, velocities and displacements have common difficulties. The magnitude of the
coefficients obviously appears to be highly dependent on the vicinity of the instrument relative to the
damaged point, thus high local values no comparable to the general damage in the building. Besides, it also
seems to be dependent on the general amplitude of the signal, thus high values in the roof, where there was no
damage. Therefore, analysis of relative interstory drifts rises as a good alternative for taking all records into a
common scale which, by the way, is generally accepted to be related with the occurrence of damage in
structures.
Before going into the results it must be mentioned that, although it has been demonstrated that omission of the
influence of soil-structure interaction in the relative inter-story drifts due to rocking of the structure may leads
to non-conservative errors (Trifunac and Ivanovi, 2003), it is here assumed that these effects are mainly
affecting the low frequency band of the signal nor the detail coefficients of it. Therefore, analysis of the high
frequency band is out of this influence, giving to the wavelet analysis an additional positive characteristic.

46

VELOCITY (cm/s2 )
0.125 to 25.0 Hz

NS DIRECTION
80
40
0
-40
-80

CH 1

CH 2

CH 3

CH 7

CH 8

CH 9

CH 10

CH 11

EW DIRECTION
120
60
0
-60
-120

CH 4

CH 5

CH 6

CH 13

10

15

20

Time (s)

Figure 18 - Horizontal velocities from the 1979, Imperial Valley earthquake

47

DISPLACEMENTS (cm/s2 )
0.125 to 25.0 Hz

NS DIRECTION
20
10
0
-10
-20

CH 1

CH 2

CH 3

CH 7

CH 8

CH 9

CH 10

CH 11

EW DIRECTION
40
20
0
-20
-40

CH 4

CH 5

CH 6

CH 13

10

15

20

Time (s)

Figure 19 - Horizontal displacements from the 1979, Imperial Valley earthquake

48

SQUARE COEFFICIENTS (cm 2 /s2 )


12.5 to 25.0 Hz

NS DIRECTION
1.2
0.8
0.4

CH 1

0
1.2
0.8
0.4

CH 2

0
1.2
0.8
0.4

CH 3

0
1.2
0.8
0.4

CH 7

0
1.2
0.8
0.4

CH 8

0
1.2
0.8
0.4

CH 9

0
160
120
80
40
0

CH 10

1.2
0.8
0.4
CH 11

0
4

10

12

Time (s)

Figure 20 - Velocity detail square coefficients from


the first level of decomposition in the NS direction

49

14

SQUARE COEFFICIENTS (cm 2 )


12.5 to 25.0 Hz

NS DIRECTION
0.0004
0.0002

CH 1

0
0.0004
0.0002

CH 2

0
0.0004
0.0002

CH 3

0
0.0004
0.0002

CH 7

0
0.0004
0.0002

CH 8

0
0.0004
0.0002

CH 9

0
0.0004
0.0002

CH 10

0
0.0004
0.0002

CH 11

0
5

11

Time (s)

Figure 21 - Displacements detail square coefficients from


the first level of decomposition in the NS direction

50

13

Figure 22 shows the relative interstory drifts assessed from the records between vertically adjacent channels
and figure 23 shows the correspondent detail square coefficients from the first level of decomposition for (a)
NS direction and (b) EW direction. Opposite to the previous analysis, this time was possible to define a
common scale for all pair of channels in each direction.
Results from figure 23 are totally consistent with previous results from absolute acceleration, velocity and
displacements, and besides of that, eliminates inconsistencies found in channels 1 and 10, and are almost free
of noise. Shaded circular sections highlight instances of main damage occurrence, which agree with observed
damage in the west (A) and east (F) frames.
Unfortunately, direct processing of drifts signals makes difficult the identification of minor and intermediate
damage which, according to the results from the other signals, should be more significantly present in the
areas marked with dash line circles.

Additional comments

Other alternatives were tested trying to improve the presented results. Relative displacements between nonadjacent records did not conduce to any better results. Discontinuities from the signals were lost because of
the longer distances. Other major effort was done using quotients of acceleration signals in an attempt for
eliminating the effect of high frequency content on the ground records. No positive result was achieved.
Indeed, these pseudo-transfer functions led to full of noisy spurious peaks without any sense or relation with
the damage.
Further analysis in the results presented so far is redundant.

For improvement on identification of

intermediate and minor damage it will be better to study the strong motion records from the VN7SH, which
case could be additionally tested in records from which it is known that no damage was detected.

RESULTS FROM THE VAN NUYS SEVEN-STORY HOTEL BUILDING

Given the results from the ICS building, analysis from the VN7SH building will be presented in three parts.
First part deals with the absolute motion of the building, for which the results from acceleration and
displacement records are shown. Velocity has not been considered because, according to what was seen in
the ICS building, acceleration and displacement cover the two extreme cases of full of richness and poorness
of high frequency content. Second part center the attention on the relative interstory drifts and third part

51

INTERSTORY DRIFTS
0.125 to 25.0 Hz

NS DIRECTION
0.008
0.004
0
-0.004
-0.008

CH 1 to 7

CH 2 to 8

CH 3 to 9

CH 7 to 10

CH 8 to 11

CH 9 to 11

EW DIRECTION
0.016
0.008
0
-0.008
-0.016

CH 4 to 5

CH 5 to 6

CH 6 to 13

10

15

20

Time (s)

Figure 22 - Relative interstory drifts assessed from the records


between vertically adjacent channels

52

NS DIRECTION

SQUARE COEFFICIENTS
12.5 to 25.0 Hz

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 1 to 7

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 2 to 8

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 3 to 9

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 7 to 10

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 8 to 11

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 9 to 11

EW DIRECTION
1.8E-010
1.2E-010
6E-011
CH 4 to 5

0
1.8E-010
1.2E-010
6E-011

CH 5 to 6

0
1.8E-010
1.2E-010
6E-011

CH 6 to 13

0
5

11

13

Time (s)

Figure 23 - Drifts detail square coefficients from the first level of decomposition

53

presents an alternative to improve limitation observed regard to detection of minor and intermediate damage
in the analysis of interstory drifts.

Absolute accelerations and displacements from the 1994, Northridge earthquake

Figures 23 and 24 show absolute accelerations and displacements recorded in the building during the 1994,
Northridge earthquake. Figures 25 and 27 show their correspondent detail square coefficients, respectively.
Shaded areas highlight the most relevant peak values and consistent along the channels, which are those
believed to be associated with damage occurred in the building.
As well as in the ICS building, square coefficients from acceleration records are richer in high frequency
content, especially in the ground level, i.e. channels 1, 13 an 16, making the definition of a common scale
useless for comparisons between channels. Nevertheless, from figure 26 it is possible to roughly identify a
first group of time intervals during which damage should have taken place. Damage seems to had started at
about 3.5 s near channels 4 and 5 and a general damage in the building should had occurred at about 4 s, when
almost all the channels show a significant peak value in the NS direction and about 4.5 to 5 s in the EW
direction. This first interval goes up to 6.2 s. A few seconds later a second range of time may be detected in
the NS component. Starting at channel 3 at about 6.7 s, this interval goes until 9.5 s, approximately. In the
EW direction, analog interval starts after 8 s and finishes just after the one of the NS component. Discarding
the ground level channels, the most representative peak values of these intervals are in channels 3 an 8 in the
NS direction, and in channels 10 and 11 in the EW direction. A final interval is present in the NS direction in
channels 4 and 6 at about 11 s.
Analysis of the displacements helps in diminishing the high frequency and allows the definition of a common
scale for all the channels without significant loss of information. Four intervals within the ones defined from
acceleration records are here better time and channel localized. Once again generally spread damage in the
building in the NS direction seems to had occurred between 4 and 5 s. Channels at 1st and 2nd floor (1, 7, 8,
12, 16) in both directions reveal the occurrence of damage from 5.5 to 6.5 s and from 7.2 to 8 s in the NS
component and from 7.1 to 7.4 in the EW component. Within these last intervals, damage is also detectable
near channel 4. Some isolated peak values are present in the roof channels in the NS direction (2 and 3) at
about 6.8 s and in east side channels (4, 6, 8 and 13) at about 11 s. Other relevant peaks suggesting the
occurrence of damage in the EW direction are in the higher floors of the east side between 9 and 9.5 s. It
should be also noticed that the most significant peak values in the EW direction are concentrated in the
ground level (channel 16), and they are excellently time located at 5.2, 5.9, 6.2, 7.3, 8.2, 8.6 and 9 s.

54

ACCELERATION (cm/s2 )
0.125 to 25.0 Hz

NS DIRECTION
600
300
0
-300
-600

CH 2

CH 3

CH 4

CH 5

CH 6

CH 7

CH 8

CH 1

CH 13

EW DIRECTION
600
300
0
-300
-600

CH 9

CH 10

CH 11

CH 12

CH 16

10

15

20

Time (s)

Figure 24 - Horizontal accelerations from the 1994, Northridge earthquake

55

DISPLACEMENT (cm)
0.125 to 25.0 Hz

NS DIRECTION
30
15
0
-15
-30

CH 2

CH 3

CH 4

CH 5

CH 6

CH 7

CH 8

CH 1

CH 13

EW DIRECTION
30
15
0
-15
-30

CH 9

CH 10

CH 11

CH 12

CH 16

10

15

20

Time (s)

Figure 25 - Horizontal displacements from the 1994, Northridge earthquake

56

SQUARE COEFFICIENTS (cm 4 /s2 )


12.5 to 25.0 Hz

NS DIRECTION
1600
1200
800
400
0

CH 2

2000
1000
CH 3

0
1200
800
400

CH 4

0
800
400

CH 5

0
1600
800

CH 6

0
1600
800

CH 7

0
3000
2000
1000

CH 8

0
3000
2000
1000

CH 1

0
4000
3000
2000
1000
0

CH 13

12

16

Time (s)

Figure 26 - (a) Acceleration detail square coefficients


from the first level of decomposition in the NS direction

57

20

SQUARE COEFFICIENTS (cm 4 /s2 )


12.5 to 25.0 Hz

EW DIRECTION
800
600
400
200
0

CH 9

1200
800
400
CH 10

0
1600
800

CH 11

0
1600
1200
800
400
0

CH 12

8000
6000
4000
2000
0

CH 16

12

16

Time (s)

Figure 26 - (b) Acceleration detail square coefficients


from the first level of decomposition in the EW direction

58

20

SQUARE COEFFICIENTS (cm 2 )


12.5 to 25.0 Hz

NS DIRECTION
0.00012
8E-005
4E-005

CH 2

0
0.00012
8E-005
4E-005

CH 3

0
0.00012
8E-005
4E-005

CH 4

0
0.00012
8E-005
4E-005

CH 5

0
0.00012
8E-005
4E-005

CH 6

0
0.00012
8E-005
4E-005

CH 7

0
0.00012
8E-005
4E-005

CH 8

0
0.00012
8E-005
4E-005

CH 1

0
0.00012
8E-005
4E-005

CH 13

0
3

10

11

12

13

Time (s)

Figure 27 - (a) Displacements detail square coefficients


from the first level of decomposition in the NS direction

59

14

15

SQUARE COEFFICIENTS (cm 2 )


12.5 to 25.0 Hz

EW DIRECTION
0.00016
0.00012
8E-005
4E-005
0

CH 9

0.00016
0.00012
8E-005
4E-005
0

CH 10

0.00016
0.00012
8E-005
4E-005
0

CH 11

0.00016
0.00012
8E-005
4E-005
0

CH 12

0.00016
0.00012
8E-005
4E-005
0

CH 16

10

11

12

13

Time (s)

Figure 27 - (b) Displacements detail square coefficients


from the first level of decomposition in the EW direction

60

14

15

The previous description of detected damage from absolute acceleration and displacement records has very
good agreement with the observed damage in the VN7SH building during the 1994, Northridge earthquake.
Preliminary conclusions could roughly described the damage starting at 4 s wide spread in the building. After
this, the damage concentrates in the 1st, 2nd and 3rd floors in both directions, subsequently occurring between 5
and 8 s, and probably related with the observed damage at the east and west ends (figure 11). Finally, last
main damage seems to had occurred at intermediate floors, between channels 4, 6 and 8, and 9, 10 and 11
(transverse and longitudinal components, respectively), in agreement with damage at 5th floor in the east side
of the building.
However, some of these results present inconsistencies with the observations. Channel 2 shows in both
accelerations and displacements, prominent peaks not easily associated with damage. In order to clean these
spurious values, as previously done for the ICS building, it will be next evaluated the interstory drifts of the
recorded signals.

Interstory Drifts from the 1994, Northridge earthquake

Figure 28 shows interstory drifts assessed from the records of the VN7SH building during the 1994, for
vertically adjacent pairs of channels. Figure 29 show the correspondent detail square coefficients from the
first level of their discrete wavelet transform in (a) the NS direction and (b) the EW direction. Amplitude of
square coefficients in figure 29 are on a common scale for all channels, and it should be noticed that this was
the same scale use in the results from drifts analysis in the ICS building, thus the advantage of drifts on
making possible the comparison between different events and structures.
Analysis of the results shown in these figures leads to the following conclusions. Drifts are certainly useful
on eliminating without agreement with actual damage results, like those from channel 2. But, as mentioned
before, when analyzing the case of the ICS building, in the process, although highlighting and better time
locating most of the results, they also vanish other relevant ones.
It can be seen in figure 29a that damage associated shaded regions can be narrower defined. Damage in the
transverse direction is very well space and time located. Regard to the 1st to 3rd floors of the building (drifts
form channels 5 to 7, 7 to 1, 6 to 8 and 8 to 16) damaged occurred at 4 to 5, 6, 7.2, 7.9, 8,7, 10.2 and 11 s,
while it took place near the east side of the roof (channels 3 to 4) about 4.5, 6.7, 8,7 and 10.6 s. In the
longitudinal direction (figure 19b) results just confirm the times observed in channel 16 in absolute
displacements.

61

INTERSTORY DRIFTS
0.125 to 25.0 Hz

NS DIRECTION
0.016
0.008
0
-0.008
-0.016

CH 2 to 5

CH 3 to 4

CH 4 to 6

CH 5 to 7

CH 6 to 8

CH 7 to 1

CH 8 to 13

EW DIRECTION
0.016
0.008
0
-0.008
-0.016

CH 9 to 10

CH 10 to 11

CH 11 to 12

CH 12 to 16

10

15

20

Time (s)

Figure 28 - Relative interstory drifts assessed from the records


between vertically adjacent channels

62

NS DIRECTION

SQUARE COEFFICIENTS
12.5 to 25.0 Hz

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 2 to 5

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 3 to 4

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 4 to 6

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 5 to 7

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 6 to 8

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 7 to 1

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 8 to 13

10

11

12

13

Time (s)

Figure 29 - (a) Interstory drifts detail square coefficients


from the first level of decomposition in the NS direction

63

14

15

EW DIRECTION

SQUARE COEFFICIENTS
12.5 to 25.0 Hz

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 9 to 10

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 10 to 11

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 11 to 12

1.6E-009
1.2E-009
8E-010
4E-010
0

CH 12 to 16

10

11

12

13

14

Time (s)

Figure 29 - (b) Interstory drifts detail square coefficients


from the first level of decomposition in the EW direction

64

15

Unfortunately, as also mentioned in the case of the ICS building, somehow, drifts analysis makes some of the
results to vanish. Drifts square coefficients from channels 4 to 6 (NS), and 9 to 10 and 10 to 11 (EW),
practically disappeared (dash line circles in figures 29a and 29b). Several attempts not presented here were
tried in order to correct this problem and next section present the so far most appropriate solution.

Average Normalized Square Coefficients of Drifts from the 1994, Northridge earthquake

For solving the mentioned problem, the square coefficients of drifts were normalized using the Average
Normalization Factor proposed in equation 17.

This factor was separately calculated for the square

coefficients of each drift signal. Results for the 1994, Northridge earthquake, same just analyzed in the
previous section, are shown in figure 30 for (a) the NS and (b) the EW direction.
Kindness of this alternative is immediately visible. Narrowed time location of the damage continue being as
good as in the unnormalized drift coefficients but results from relevant pairs of channels that were previously
lost are here recovered. Average normalization factor can be interpreted as the overall level of high frequency
content in the signal (drifts in this case), thus, when normalizing, singular peak values that do not globally
affect its assessment, rises over the general trend of the signal. As a result, for the specific case showed
herein, damage associated with drift from channels 4 to 6 (NS), and 9 to 10 and 10 to 11 (EW) are recovered
without affecting other results.
Great advantage of this normalization was noticed when trying to relate the scale of magnitude with the
observed damage. Table 3 presents the maximum peak values for each pair of channels showed in figure 30.
This value corresponds to what was defined as the Maximum Level of Damage Index (equation 20) for the
entire sample of coefficients. Table 3 also shows the direction and location of each pair of channels, and the
time at what the maximum level of damage index is reached.

Table 3 - Maximum level of damage index assessed from drifts signals,


for the VN7SH building during 1994, Northridge earthquake

Channel
Pair

Direction

Location

Maximum Level
of Damage Index
(lmax)

Time

2 to 5

NS

rd
Roof to 3 Floor
West Side

43

8.3 s

5 to 7

NS

rd
nd
3 to 2 Floor
West Side

51

7.3 s

7 to 1

NS

94

7.3 s

nd

to 1st Floor
West side

65

NS DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 2 to 5

0
120
80
40

CH 3 to 4

0
120
80
40

CH 4 to 6

0
120
80
40

CH 5 to 7

0
120
80
40

CH 6 to 8

0
120
80
40

CH 7 to 1

0
120
80
40

CH 8 to 13

0
3

10

11

12

13

14

15

Time (s)

Figure 30 - (a) Average normalized interstory drifts detail square coefficients


from the first level of decomposition in the NS direction

66

EW DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 9 to 10

0
120
80
40

CH 10 to 11

0
120
80
40

CH 11 to 12

0
120
80
40

CH 12 to 16

0
3

10

11

12

13

14

15

Time (s)

Figure 30 - (b) Average normalized interstory drifts detail square coefficients


from the first level of decomposition in the EW direction

67

Table 3 (continues) - Maximum level of damage index assessed from drifts signals,
for the VN7SH building during 1994, Northridge earthquake

Channel
Pair

Direction

Location

Maximum Level
of Damage Index
(lmax)

Time

3 to 4

NS

th
Roof to 6 Floor
East Side

86

6.7 s

4 to 6

NS

th
rd
6 to 3 Floor
East Side

63

6.1 s

6 to 8

NS

rd
nd
3 to 2 Floor
East Side

67

6.1 s

8 to 13

NS

52

6.0 s

9 to 10

EW

th
Roof to 6 Floor
East Side

60

9.2 s

10 to 11

EW

th
rd
6 to 3 Floor
East Side

68

4.9 s

11 to 12

EW

rd
nd
3 to 2 Floor
East Side

89

4.7 s

12 to 16

EW

79

6.2 s

nd

to 1st Floor
East side

nd

to 1st Floor
East side

It can be seen that these results are in excellent agreement with the previous analysis of acceleration,
displacements and unnormalized drifts. Times perfectly coincides with the main intervals and their location
match the observed damage in the building (figure 11). It would be useful, for future comparisons with other
cases to introduce what as defined as the Global Maximum Level of Damage Index (equation 21). This value
corresponds to the maximum value of the lmax values of all the channels pairs. For the case the VN7SH
building, during the 1994, Northridge earthquake, the global maximum level of damage index is
Lmax = 94
and it seems to be associated with the failure of columns in the east side of the building near the 2nd floor.
At this level of the study, it is appropriate to test the so far most relevant conclusions in records from other
earthquakes where no damaged was detected or reported in the VN7SH building. It is also important to go
back over the previous results from the ICS building to check if the proposed analysis using average
normalized square coefficients of drifts fit with the results presented above. Next section will pursue this and
other matter of interest.

68

AVERAGE NORMALIZED SQUARE COEFFICIENTS OF DRIFTS

In this section, the assessment of normalized square coefficients of drifts is used in the VN7SH building for
other events different from the main event of the 1994, Northridge earthquake, whose results were just
presented in the previous section. This procedure is also tested in the ICS building.

Results from earthquakes in the VN7SH building with minor or no damage

Four additional earthquakes are herein analyzed.

The first ones are the two aftershocks of the 1994,

Northridge earthquake. As was mention in the damage description of the building, it is known that after the
March aftershock some of the cracks in the structure were lager and it is assumed that additional damaged
occurred. On the other hand, for the December aftershock no damage was reported. Third and fourth events
here analyzed are the main event of the October 10, 1987 Whittier Narrows earthquake and the January 19,
1989 Malibu earthquake. For none of these two earthquakes damage was reported. Main characteristics of
these four events were presented in table 2. All other earthquakes included in table 2 were also analyzed, but
no significant results, different from those that will be here discussed, were found.
Part (a) of figures 31 to 34 shows the assessed relative interstory drifts signals for the Northridge March
aftershock, December aftershock and Whittier Narrows and Malibu earthquakes. Part (b) of the mentioned
figures shows their correspondent average normalized square coefficients. Analysis of these results led to the
following comments.
Reported damage during the March 1994, Northridge aftershock was detected in several of the records in both
directions. Selected shaded regions in figure 31b suggest that damage occurred at two different instants. First
evidence takes place between 3 and 4 s and is mainly associated with the lower levels (channel pairs 7 to 1, 6
to 8 and 8 to 13). Second damage should have taken place near the lower level of the west side of the
building (channel pair 7 to 1).
Except for the appearance of two low magnitude values (<45) between the 1st and 2nd floor (channel pairs 7 to
1 and 12 to 16), results shown in figure 32b confirms the fact that no damage occurred in the building during.
Figure 33b presents the obtained average normalized square coefficients for the 1987, Whittier Narrows
earthquake. Although no damage was reported after the this event, results from several pair of channels in
both directions suggest the occurrence of some possible damage between 5.9 and 6.5 s. Nevertheless,
magnitudes of these normalized coefficients (~60) are not significant compared to the ones obtained from

69

NS DIRECTION

INTERSTORY DRIFTS
0.125 to 25.0 Hz

0.003
0.0015
0
-0.0015
-0.003

CH 2 to 5

CH 3 to 4

CH 4 to 6

CH 5 to 7

CH 6 to 8

CH 7 to 1

CH 8 to 13

EW DIRECTION
0.0012
0.0006
0
-0.0006
-0.0012

CH 9 to 10

CH 10 to 11

CH 11 to 12

CH 12 to 16

10

15

20

Time (s)

Figure 31 - (a) Relative interstory drifts assessed from the records between vertically
adjacent channels in the VN7SH building during the March 1994, Northridge aftershock

70

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

NS DIRECTION
120
80
40

CH 2 to 5

0
120
80
40

CH 3 to 4

0
120
80
40

CH 4 to 6

0
120
80
40

CH 5 to 7

0
120
80
40

CH 6 to 8

0
120
80
40

CH 7 to 1

0
120
80
40

CH 8 to 13

0
0

10

11

12

13

14

15

Time (s)

Figure 31 - (b) Average normalized square coefficients of drifts


in the VN7SH building during the March 1994, Northridge aftershock

71

EW DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 9 to 10

0
120
80
40

CH 10 to 11

0
120
80
40

CH 11 to 12

0
120
80
40

CH 12 to 16

0
0

10

11

12

13

14

15

Time (s)

Figure 31 - (b) (continues) Average normalized square coefficients of drifts


in the VN7SH building during the March 1994, Northridge aftershock

72

NS DIRECTION

INTERSTORY DRIFTS
0.125 to 25.0 Hz

0.0012
0.0006
0
-0.0006
-0.0012

CH 2 to 5

CH 3 to 4

CH 4 to 6

CH 5 to 7

CH 6 to 8

CH 7 to 1

CH 8 to 13

EW DIRECTION
0.0012
0.0006
0
-0.0006
-0.0012

CH 9 to 10

CH 10 to 11

CH 11 to 12

CH 12 to 16

10

15

20

Time (s)

Figure 32 - (a) Relative interstory drifts assessed from the records between vertically
adjacent channels in the VN7SH building during the December 1994, Northridge aftershock

73

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

NS DIRECTION
120
80
40

CH 2 to 5

0
120
80
40

CH 3 to 4

0
120
80
40

CH 4 to 6

0
120
80
40

CH 5 to 7

0
120
80
40

CH 6 to 8

0
120
80
40

CH 7 to 1

0
120
80
40

CH 8 to 13

0
0

10

11

12

13

14

15

Time (s)

Figure 32 - (b) Average normalized square coefficients of drifts


in the VN7SH building during the December 1994, Northridge aftershock

74

EW DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 9 to 10

0
120
80
40

CH 10 to 11

0
120
80
40

CH 11 to 12

0
120
80
40

CH 12 to 16

0
0

10

11

12

13

14

15

Time (s)

Figure 32 - (b) (continues) Average normalized square coefficients of drifts


in the VN7SH building during the December 1994, Northridge aftershock

75

NS DIRECTION

INTERSTORY DRIFTS
0.125 to 25.0 Hz

0.006
0.003
0
-0.003
-0.006

CH 2 to 5

CH 3 to 4

CH 4 to 6

CH 5 to 7

CH 6 to 8

CH 7 to 1

CH 8 to 13

EW DIRECTION
0.003
0.0015
0
-0.0015
-0.003

CH 9 to 10

CH 10 to 11

CH 11 to 12

CH 12 to 16

Channel 16 did not record this event

10

15

20

Time (s)

Figure 33 - (a) Relative interstory drifts assessed from the records between vertically
adjacent channels in the VN7SH building during the 1987, Whittier Narrows earthquake

76

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

NS DIRECTION
120
80
40

CH 2 to 5

0
120
80
40

CH 3 to 4

0
120
80
40

CH 4 to 6

0
120
80
40

CH 5 to 7

0
120
80
40

CH 6 to 8

0
120
80
40

CH 7 to 1

0
120
80
40

CH 8 to 16

0
0

10

11

12

13

14

15

Time (s)

Figure 33 - (b) Average normalized square coefficients of drifts


in the VN7SH building during the 1987, Whittier Narrows earthquake

77

EW DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 9 to 10

0
120
80
40

CH 10 to 11

0
120
80
40

CH 11 to 12

0
0

10

11

12

13

14

15

Time (s)

Figure 33 - (b) (continues) Average normalized square coefficients of drifts


in the VN7SH building during the 1987, Whittier Narrows earthquake

78

NS DIRECTION

INTERSTORY DRIFTS
0.125 to 25.0 Hz

0.0006
0.0003
0
-0.0003
-0.0006

CH 2 to 5

CH 3 to 4

CH 4 to 6

CH 5 to 7

CH 6 to 8

CH 7 to 1

CH 8 to 13

EW DIRECTION
0.0006
0.0003
0
-0.0003
-0.0006

CH 9 to 10

CH 10 to 11

CH 11 to 12

CH 12 to 16

10

15

20

Time (s)

Figure 34 - (a) Relative interstory drifts assessed from the records between vertically
adjacent channels in the VN7SH building during the 1989, Malibu earthquake

79

NS DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 2 to 5

0
120
80
40

CH 3 to 4

0
120
80
40

CH 4 to 6

0
120
80
40

CH 5 to 7

0
120
80
40

CH 6 to 8

0
120
80
40

CH 7 to 1

0
120
80
40

CH 8 to 13

0
0

10

11

12

13

14

Time (s)

Figure 34 - (b) Average normalized square coefficients of drifts


in the VN7SH building during the 1989, Malibu earthquake

80

15

EW DIRECTION

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

120
80
40
CH 9 to 10

0
120
80
40

CH 10 to 11

0
120
80
40

CH 11 to 12

0
120
80
40

CH 12 to 16

0
0

10

11

12

13

14

15

Time (s)

Figure 34 - (b) (continues) Average normalized square coefficients of drifts


in the VN7SH building during the 1989, Malibu earthquake

81

AVERAGE NORMALIZED SQUARE COEFFICIENTS


12.5 to 25.0 Hz

NS DIRECTION
120
80
40

CH 1 to 7

0
120
80
40

CH 2 to 8

0
120
80
40

CH 3 to 9

0
120
80
40

CH 7 to 10

0
120
80
40

CH 8 to 11

0
120
80
40

CH 9 to 11

EW DIRECTION
120
80
40
CH 4 to 5

0
120
80
40

CH 5 to 6

0
120
80
40

CH 6 to 13

0
5

11

Time (s)

Figure 35 - Average normalized square coefficients of drifts


in the ICS building during the 1979, Imperial Valley earthquake

82

13

records of earthquakes where it is known that damage did occurred. As a result, Whittier Narrows will be
here classified as a not visible damage or not damage reported earthquake.
Finally, figure 34b show the average normalized coefficients from the 1989, Malibu earthquake. During this
earthquake no damage was reported and as a result of that there is not even one peak value that rises over the
general trend of the entire sample of coefficients.
Except for the case of the 1971, San Fernando earthquake, for which records from channels 4 to 6 are not
available, thus drifts were not possible to be assessed, because of what it is here omitted; results for the rest of
earthquakes mentioned in table 2 were by far the same case of Malibu earthquake. No damage reported and
no damage detected. This confirms the suitability of the proposed approach to determine the occurrence of
damage.
It is good to mention that all (b) parts of figures 31 to 34 were plotted in a common scale. The same that was
used in the previous section for the 1994, Northridge earthquake and the same that will be used in what
follows for the ICS building during the 1979, Imperial Valley earthquake.

Results from the 1979, Imperial Valley earthquake in the ICS Building

Figure 35 shows the average normalized square coefficients of drifts from the records of the ICS building
during the 1979, Imperial valley earthquake. Correspondent signals are those previously shown in figure 22.
Although, because of the magnitude of the main damage occurred in this building, some of the peak values
get lost in the normalizing process, it can be seen that the severe damage of columns at the east side of the
building (frame F) are very well detectable, and so it is the damage at the west side (frame A), at 6.4 and 7.1
s, respectively.

MAXIMUM AND GLOBAL MAXIMUM LEVEL OF DAMAGE INDEXES

When analyzing the results from the averaged normalized square coefficients of drifts from the 1994,
Northridge earthquake in the VN7SH building, the proposed maximum level of damage index (lmax) was
introduced as a practical way of analyzing the occurrence of damage and so was the global maximum level of
damage index (Lmax). Table 4 shows the results for these two parameters in all the available records from the
VN7SH building, except for the 1971, San Fernando earthquake, for which the lacking of records from
channels 4 to 6 (figure 9) makes unviable the assessment of drifts.

83

Table 4 - Maximum level of damage index (lmax) assessed from drifts signals for the VN7SH building
Earthquake
Channel
Pair

Dir.

Location

WTN

WTA

PSD

MTB

MAL

SMD

LAN

BBR

NOR

N03

N12

(10/01/87)

(10/04/87)

(10/03/88)

(06/12/89)

(01/19/89)

(06/28/91)

(06/28/92)

(06/28/92)

(01/17/94)

(03/20/94)

(12/06/94)

2 to 5

NS

Roof to 3rd Floor


West Side

43

11

23

19

14

43

29

10

5 to 7

NS

3rd to 2nd Floor


West Side

44

12

12

26

51

33

10

7 to 1

NS

2nd to 1st Floor


West side

61

13

31

11

14

10

94

74

40

3 to 4

NS

Roof to 6th Floor


East Side

45

18

13

10

29

15

86

22

14

4 to 6

NS

6th to 3rd Floor


East Side

23

11

10

10

24

10

63

43

10

6 to 8

NS

3rd to 2nd Floor


East Side

32

10

17

10

16

67

71

13

8 to 13

NS

2nd to 1st Floor


East side

33

10

15

10

19

15

11

52

55

14

9 to 10

EW

Roof to 6th Floor


East Side

27

11

14

16

15

11

60

19

th

rd

10 to 11

EW

6 to 3 Floor
East Side

28

21

13

14

20

13

11

68

34

15

11 to 12

EW

3rd to 2nd Floor


East Side

58

12

29

24

11

89

62

12 to 16

EW

2nd to 1st Floor


East side

---

32

24

35

10

11

79

78

43

Global Maximum Level of Damage


(Lmax)

61

32

31

16

19

35

19

14

94

78

43

>0

> 30

> 50

> 70

84

In table 4, some of the cells have been highlighted by an arbitrary defined scale that is shown at the bottom of
the table.

It can be seen that only the 1987, Whittier Narrows earthquake and the 1994, Northridge

earthquake and its aftershocks have valuees of lmax > 40.


Given the base of defining the 1987, Whittier Narrows earthquake as a not reported damage earthquake, it
would be reasonable that the proposed indexes fairly classify the recorded earthquakes in the building, being
the 1994, Northridge earthquake main event the one with the highest value, followed by its March aftershock,
where damage was less sever, and with the Whittier Narrows earthquake and the rest of the events where
there was no evidence of damage behind.
For an upcoming discussion it must be here noticed that the two maximum values from the 1987, Whittier
Narrows earthquake comes from the same two pair of signals in which the highest values where reached
during the 1994, Northridge earthquake.
Figure 36 compares the global maximum level of damage indexes (Lmax) from the recorded earthquakes in the
VN7SH building and also includes the value of Lmax for the ICS building during the 1979, Imperial Valley
earthquake. Also included on the right hand side of the chart in figure 36, a classification of the damage
observed in the buildings according to the values obtained for Lmax. Results from the earthquakes in which no
damaged was observed suggest that damage occur for values of Lmax > 40.

SEQUENCE OF DAMAGE OCCURRENCE

When analyzing the two cases of study it was seen in deep detail, i.e. signal by signal, the time occurrence of
damage but its magnitude was just a matter of being able or not to detect it. Previous section analyzed better
the magnitude and relevance of it in the structure, but left behind the objective when it was taking place. In
what follows, both approaches will be put together for understanding how damage of the structure progress in
time.
Figures 37 to 40 show a time window analysis of lmax (thin lines) and Lmax (thick line) for the cases in which
damage was detected. According with their definitions (equations 20 and 21), Lmax is the envelope of lmax for
all the channel pairs, which, for clarity, are not discretely indicated in the figures. Length of the window was
arbitrary fixed at ~0.5 s.

85

Near
Colapse
100
Severe
Damage
80
Intermediate
Damage
60
Not Reported
Damage
40
No
Damage

20

VN7SH
Building

IMV

N12

N03

NOR

BBR

LAN

SMD

MTB

MAL

PSD

WTA

0
WTN

Global Maximum Level of Damage (Lmax)

120

ICS
Building

Figure 36 - Comparison of values for the global maximum level of damage


from the records of the VN7SH and the ICS building

86

120

l max and Lmax

100
80
60
40
20
0
0

10

12

14

16

18

20

Time (s)

Figure 37 - Sequence of damage occurrence in the VN7SH building


during the 1987, Whittier Narrows earthquake

120

l max and Lmax

100
80
60
40
20
0
0

10

12

14

16

18

Time (s)

Figure 38 - Sequence of damage occurrence in the VN7SH building


during the 1994, Northridge earthquake

87

20

120

l max and Lmax

100
80
60
40
20
0
0

10

12

14

16

18

20

Time (s)

Figure 39 - Sequence of damage occurrence in the VN7SH building


during the March 1994, Northridge aftershock

120

l max and Lmax

100
80
60
40
20
0
0

10

12

14

16

18

Time (s)

Figure 40 - Sequence of damage occurrence in the ICS building


during the 1979, Imperial Valley earthquake

88

20

According to the levels of damage defined in figure 36, damage is present when the values of either lmax
(locally) or Lmax (for the whole building) exceed a limit of about 40. Therefore, it can be seen from these
figures how damage was evolving in time in the whole structure.
In the VN7SH building, for the case of the 1987, Whittier Narrows earthquake, although no damage was
reported, results suggest the occurrence of some low level damage between 6 and 8 s, approximately. During
the 1994, Northridge earthquake, main damages seem to had taken place at 4.5 and near 7 s, but generally in
the building, damage started at 4 s and finished at 11 s. Finally, analysis of the March 1994, Northridge
aftershock suggest the occurrence of damage between 3 and 4 s, and an isolated low presence at 9 s.
In the ICS building, during the 1979, Imperial Valley earthquake, damage occurred in two separate but well
concentrated instances at 6.5 and 11 s. In this case Lmax index reached the highest value (about 120).
Comparison between these different cases suggests other interesting conclusions. Distribution of the peak
values in time shows that, although damage was more severe in the ICS building, concentrated in two well
defined peak values, thus the highest value of Lmax, damage in the VN7SH building during the Northridge
main event was more spread in the structure, i.e. presence of multiple peak values of Lmax. Consequently, for
the latest, it would be a good idea to see how damage travels along the structure in time.
Even though the instrumentation distribution is not optimum to se damage in the interior frames, figure 41
shows a schematic animation of the evolution of damage presence in the VN7SH building at selected
instances between 3 and 12 s, approximately. In figure 41, first (top-left) frame recall the location of
accelerometers in the building and just aside (top-right), concentric circles describe the used scale for
representing the reached value of lmax. Since damage is being evaluated by the values of lmax from the average
normalized square coefficients of drifts between vertically adjacent instruments, concentric circles meaning
damage in the following frames are located between the instruments showed in the first frame. For the east
side of the building, where two pairs of channels is available (for both NS and EW oriented instruments), the
plotted value corresponds to the maximum of them. At the bottom of each frame, selected time and its
correspondent value of Lmax is included.
This representation excellently illustrates the way damage is detected by each pair of channels (drifts) and
how it is felt by other pairs at the same instant. It also gives a very good idea of how the occurrence of
damage in some place a kind of a chain effect that continues causing damage in other points, and suggest the
possibility of a future study on how damage migrates through the building. In concordance with previous
results, figure 41 shows that main damage in the VN7SH building during 1994, Northridge earthquake
occurred at about 4.5, 6.5 and 8.7 s in the east side of the building and about 7 s in the west end.

89

INSTRUMENTS CONFIGURATION
2

SCALES
3,9
lmax > 5

lmax > 15

lmax > 25

lmax > 35

lmax > 45

lmax > 55

lmax > 65

lmax > 75

lmax > 85

4,10

6,11

8,12

13,16
t=0s

Lmax = 0

t = 3.30 s

Lmax = 16

t = 4.06 s

Lmax = 34

t = 4.10 s

Lmax = 53

t = 4.30 s

Lmax = 53

t = 4.38 s

Lmax = 53

t = 4.42 s

Lmax = 72

Figure 41 - Schematic animation of the sequence of damage occurrence in the VN7SH


building during the 1994, Northridge earthquake according to the values of lmax

90

t = 4.46 s

Lmax = 89

t = 4.62 s

Lmax = 89

t = 4.94 s

Lmax = 89

t = 4.98 s

Lmax = 89

t = 5.02 s

Lmax = 68

t = 5.22 s

Lmax = 39

t = 5.49 s

Lmax = 34

t = 5.61 s

Lmax = 41

Figure 41 - (continues) Schematic animation of the sequence of damage occurrence in the


VN7SH building during the 1994, Northridge earthquake according to the values of lmax

91

t = 5.77 s

Lmax = 52

t = 5.81 s

Lmax = 67

t = 6.33 s

Lmax = 79

t = 6.37 s

Lmax = 79

t = 6.45 s

Lmax = 86

t = 6.53 s

Lmax = 86

t = 6.73 s

Lmax = 86

t = 7.01 s

Lmax = 94

Figure 41 - (continues) Schematic animation of the sequence of damage occurrence in the


VN7SH building during the 1994, Northridge earthquake according to the values of lmax

92

t = 7.05 s

Lmax = 94

t = 7.56 s

Lmax = 51

t = 7.64 s

Lmax = 51

t = 7.72 s

Lmax = 51

t = 7.88 s

Lmax = 51

t = 8.04 s

Lmax = 51

t = 8.20 s

Lmax = 52

t = 8.32 s

Lmax = 66

Figure 41 - (continues) Schematic animation of the sequence of damage occurrence in the


VN7SH building during the 1994, Northridge earthquake according to the values of lmax

93

t = 8.64 s

Lmax = 66

t = 8.76 s

Lmax = 66

t = 8.96 s

Lmax = 60

t = 9.16 s

Lmax = 60

t = 9.52 s

Lmax = 55

t = 9.67 s

Lmax = 26

t = 9.83 s

Lmax = 14

t = 10.11 s

Lmax = 41

Figure 41 - (continues) Schematic animation of the sequence of damage occurrence in the


VN7SH building during the 1994, Northridge earthquake according to the values of lmax

94

t = 10.59 s

Lmax = 45

t = 10.67 s

Lmax = 45

t = 11.15 s

Lmax = 16

t = 11.23 s

Lmax = 14

Figure 41 - (continues) Schematic animation of the sequence of damage occurrence in the


VN7SH building during the 1994, Northridge earthquake according to the values of lmax

95

HIGH FREQUENCY ENERGY INDEXES

When comparing the sequence of damage in the VN7SH building during the main event of Northridge
earthquake with the results of the ICS building it was noticed that for the former damage was wide spread in
time (figure 37) while for the latest damage occurred in two well defined instants (figure 40). It would be a
good logical understanding of this to think that in the case of the VN7SH building, the energy release by
damage was the resultant sum of several faults releasing certain amount of energy, while on the ICS building
it was just a reduced number of faults but releasing a big amount of energy.
From this point of view, the proposed High Frequency Energy Index (EHf, equation 15) becomes a good base
to proportionally quantify the amount of energy in one specific signal or, using the proposed Global Average
(aEHf, equation 16) and Global Maximum High Frequency Energy Indexes (mEHf, equation 17), in an entire
structure.
Table 5 shows the results of these quantities for the high frequency square coefficients of the first level of
decomposition of drifts assessed from vertically adjacent instruments in the VN7SH building. Table 5
includes all the earthquakes in which assessment of interstory drifts was possible (i.e. 1971, San Fernando
earthquake was omitted).
Some cells in table 5 are filled according with the gray scale defined at the bottom of the corresponding part
of the table. The result of doing this shows a very similar pattern to the one observed before in the case the
maximum level of damage index (table 4). Maximum values in table 5 are associated with the main event and
the March aftershock of Northridge earthquake which perfectly agree with damage observed during these
events. Results in table 5 also suggest, as was noticed before, the occurrence of some minor damage during
the 1987, Whittier Narrows earthquake.
A very interesting result to highlight from table 5, and that was no present in previous results, is that, all along
the different earthquakes, from Whittier Narrows to the main event of Northridge and its aftershocks, the
maximum values were always present in the same pairs of signals, i.e. east and west side of the building
between the 3rd and 2nd floor. This behavior might be with the fact that some of the major damage occurred
during the 1994, Northridge earthquake was at these points.
Figure 42 and 43 respectively show the same results of mEHf and aEHf for the events recorded in the VN7SH
building, but also including the results from the ICS building during the 1979, Imperial Valley earthquake.

96

Table 5 - High frequency energy indexes (EHf x 10-12) assessed from drifts signals for the VN7SH building
Earthquake
Channel
Pair
2 to 5
5 to 7
7 to 1

Dir.

NS
NS
NS

Location
Roof to 3rd Floor
West Side
rd

nd

nd

st

3 to 2 Floor
West Side
2 to 1 Floor
West side

WTN

WTA

PSD

MTB

MAL

SMD

LAN

BBR

NOR

N03

N12

(10/01/87)

(10/04/87)

(10/03/88)

(06/12/89)

(01/19/89)

(06/28/91)

(06/28/92)

(06/28/92)

(01/17/94)

(03/20/94)

(12/06/94)

12

18

145

62

64

60

59

75

63

57

366

159

56

116

31

45

25

25

36

29

29

161

242

37

49

17

19

16

18

24

17

18

136

36

18

15

10

44

21

107

62

71

61

59

65

59

58

559

301

67

72

33

34

26

29

37

26

37

424

407

35

42

18

19

16

18

23

21

21

60

23

14

20

33

15

155

65

88

61

60

76

72

67

415

186

61

32

41

24

28

42

29

44

409

183

39

155

65

88

61

60

76

72

67

559

407

67

31

37

28

29

36

30

32

239

144

31

th

3 to 4
4 to 6
6 to 8
8 to 13

NS
NS
NS
NS

Roof to 6 Floor
East Side
th

rd

rd

nd

nd

st

6 to 3 Floor
East Side
3 to 2 Floor
East Side
2 to 1 Floor
East side
th

9 to 10
10 to 11
11 to 12
12 to 16

EW
EW
EW
EW

Roof to 6 Floor
East Side
th

rd

rd

nd

nd

st

6 to 3 Floor
East Side
3 to 2 Floor
East Side
2 to 1 Floor
East side

Global Maximum
High Frequency Index (mEHf)

>0

> 100

> 200

Global Average
High Frequency Index (aEHf)

>0

> 50

> 100

> 400

67

> 150

97

800
800
700
600
500
400
300
200
100

VN7SH
Building

IMV

N12

N03

NOR

BBR

LAN

SMD

MTB

MAL

PSD

WTA

0
WTN

Global Maximum High Freq. Energy Index (mEHf x 10-12)

1000

ICS
Building

Figure 42 - Comparison of global maximum high frequency energy index


assessed from the drift records for the VN7SH and the ICS building

98

250

200

150

100

50

VN7SH
Building

IMV

N12

N03

NOR

BBR

LAN

SMD

MTB

MAL

PSD

WTA

0
WTN

Global Average High Freq. Energy Index (aEHf x 10-12)

300

ICS
Building

Figure 43 - Comparison of global average high frequency energy index


assessed from the drift records for the VN7SH and the ICS building

99

1000
IMP

Global Maximum (mEHf x 10-12 )

800

mEHf = aEHf
600

NOR

N03
400

200

WTN
Others

0
0

200

400

600

800

1000

Global Average ( aEHf x 10-12 )

Figure 44 - Comparison between global average and global maximum high


frequency energy index for the analyzed events in both buildings

100

Results in figure 42 show that the global maximum high frequency energy index is directly related with the
intensity of damage reached in the building. On the other hand, results from figure 43 gives an idea of how
much damage occurred in the building. It turns out, then, that damage was the most drastic in the ICS
building during the 1979, Imperial Valley earthquake (figure 42), but damage was the most spread in the
structure during the 1994, Northridge earthquake (figure 43).
Figure 44 relates both indexes together. The hypothetical case in which damage is as severe as spread in the
building, i.e. all the pair of signals from which drifts are estimated would lead to the same maximum and
average value is represented in figure 44 by the diagonal line (mEHf = aEHf). As a result, closer to this line
spreader the damage in the building. This representation better explain the issue described above. Damage in
the ICS building during the recorded earthquake was bigger in terms of how severe it was locally for the
columns in the east side of the structure, but damage in the VN7SH building during the 1994, Northridge
earthquake had a relevance presence in more places along the structure, thus closer to line mEHf = aEHf.

INDEX OF DAMAGE INTENSITY AND ITS RELATION WITH OTHER PARAMETERS

It was proposed to assume that roughly speaking, the intensity of the damage occurred in a building, could be
described in agreement with the scale defined in table 1 as the Index of Damage Intensity (IDI). Table 6
shows the assignation of this index for each one of the cases here analyzed according with the observation of
damaged occurred in the buildings and in concordance with the results so far obtained. The assignation
shown in this table also agrees with the classification showed in figure 36.
The importance of doing this rests on the objective of finding significant relations with the information that is
obtained from the analysis of the signals. In order to achieve on this purpose, values obtained in each one of
the cases for the level global maximum level of damage (Lmax) and the global maximum energy (mEHf)
indexes have been also included in table 6.
Figure 45 relates the IDI with Lmax and figure 46 does it with mEHf. In both cases, an arbitrarily defined line
of tendency has been also included. Analysis of figure 45 reveals that there is a limit value of Lmax  43
above which damage occurs with an intensity proportional to the value of Lmax. On the other hand, figure 46
suggests that there is a linearly proportional relation between the proposed classification of IDI and values of
mEHf found for each of the cases. Given this, the next logical step is to directly relate both Lmax and mEHf
without involving the arbitrarily defined scale of intensity IDI. This relation is shown in figure 47.

101

Table 6 - Indexes of damage intensity (IDI) assigned


to the analyzed cases and other parameters
IDI

Lmax

mEHf x 10-12

Whittier Narrows

61

155

Whittier Narrows aftershock

32

65

Pasadena

31

88

Montebello

19

61

Malibu

16

60

Sierra Madre

35

76

Landers

19

72

Big Bear

14

67

Northridge

94

559

Northridge aftershock (392)

78

407

Northridge aftershock (436)

43

67

Imperial Valley

116

909

Building

VN7SH

ICS

Earthquake

Analysis of the results of Lmax and mEHf combined on figure 47 suggests that damage occurs if the following
limits are exceeded.
Lmax 55
mEHf 70 1012

Nevertheless, these are just preliminary values based in the results for the two cases analyzed here.

102

IMP

Index of Damage Intensity

NOR

N03

WTN

Others
1

0
0

20

40

60

80

100

120

140

Global Maximum Level of Damage (Lmax)

Figure 45 - Comparison between global maximum level of damage


and the damage intensity indexes in the cases studied

103

IMP
5

Index of Damage Intensity

NOR
4

N03
3

WTN
2

Others

0
0

200

400

600

800

1000

Global Maximum High Frequency Energy (mEHf x 10-12 )

Figure 46 - Comparison between global maximum high frequency energy


and the damage intensity indexes in the cases studied

104

1000

Global Maximum High Frequency Energy (mEHf x 10-12 )

IMP

800

600

NOR

N03
400

200

WTN
Others

0
0

20

40

60

80

100

120

140

Global Maximum Level of Damage (Lmax)

Figure 47 - Comparison between global maximum level of damage


and the global maximum high frequency energy indexes in the cases studied

105

FINAL COMMENTS

After studying the presence of damage in the two different buildings during different earthquakes, as well as
its time of occurrence, location, magnitude and intensity, some other aspects on which the results here
presented would have effects should be discussed. In what follows, the relevance of the identification of
damage as here covered, will be confront with the needs of other interesting fields of study.

RELEVANCE FOR SYSTEM IDENTIFICATION AND SOIL-STRUCTURE INTERACTION

System identification techniques have widely study the change of structural properties such as the modal
frequencies, critical damping ratios or modal shapes. Nevertheless, it is a constant difficulty to accurately
assign the causes of the change in these properties because, besides the damage, other very complex
phenomena, like the effects of soil-structure interaction or the contribution of non-structural elements are also
involved in the behavior of the system.
Even though the procedure here presented does not yet solve all the queries, the possibility of identifying the
occurrence of damage and the ability of quantify it by means of simple indexes easily related with the damage
observed in the structure as well as being able of locating it in time and have a rough idea of its location in the
structure, opens a wide range of possibilities.
Just to mention one of these possibilities, in the case of the study of how the fundamental frequency of a
structure changes during an earthquake, it would be possible to evaluate whether damage occurred or not, and
then, being able to assign the change on frequency to likelier causes. It would be possible to study the
oscillation of the frequency out of the time interval when damage occurred and thus discreet the causes and
assign better levels of influence of the different factors in the behavior of the structural systems. And even for
those cases in which soil-structure interaction is not a predominant component, the study of damage
occurrence confronted to the performance of structures within linear and non-linear ranges could be better
determined, thus its influence in the properties of the structure, typically estimated using parametric and nonparametric system identification techniques.

RELEVANCE FOR STRUCTURAL HEALTH MONITORING AND FUTURE APPLICATION

Contrary to the complex theories developed in structural health monitoring or beyond the simple monitoring
of changing structural properties, which is really becoming a matter of common knowledge, the methods

106

presented here to identify the occurrence of damage using discrete wavelet transform for just the first level of
decomposition, and the proposed indexes for quantify it and relate it with the damaged observed in the
structure, constitutes a significant contribution for realistic application of health monitoring projects.
According to the different results presented in this report it was found that some non-detected minor damage
occurred during the 1987, Whittier Narrows earthquake near the same zones in which more severe damage
took place during the 1994, Northridge main event and its March aftershock. Besides, it was found that at
these same points, even though the defined limits of damage were no exceeded, maximum values of high
frequency associated energy were consistently present in the same areas. This major finding clearly indicates
the potential capabilities of the proposed method, not only for identifying the occurrence of damage, but also
for easily monitoring it without the need of complicate theories hardly applicable or affordable by public
agencies. And, eventually, a case like the VN7SH building during the 1994, Northridge earthquake could
have been prevented.
Futurist Remote Online Damage Monitoring Systems like the one sketched in figure 48 represent next
challenges in realistically applicable health monitoring programs. In future, reasonable well instrumented
structures should be network integrated under the surveillance of a Damage Control Center for real time
damage occurrence monitoring. This kind of control center would not be only on charge of long time
monitoring the instrumented structures but also, after an earthquake, could be of useful help for coordinating
immediate help or technical surveys.
Nevertheless, more work is needed on testing the here proposed methods and parameters for other cases in
order to improve a damage classification scale and its relation with the detectable properties from the wavelet
analysis of the records.

107

Instrumented structure
with standard comunication
Instrumented structure
with satellital comunication

Instrumented structure
with wireless comunication

DAMAGE CONTROL CENTER

Signals are received


and drifts between
vertically adjacent
instruments are
obtained

Square coefficients
and averaged normalized coefficients are
obtained

Level of damage is
determined according
with the Lmax and mEHf
indexes results from
prevously colected
data

Actions are taken


according to the
expected damage
from the indexes
results

Figure 48 - Futurist remote online damage monitoring system

108

CONCLUSIONS

This report presented a theoretical background about wavelet analysis and it possibilities in damage detection
from strong motion records and proposed different approaches and parameters to identify not only the
occurrence of damage but also for evaluate the severity of it. The mention technique is used in two
instrumented buildings for which clear evidence of damage during recorded events was available and well
reported.
Results demonstrate that the use of multiresolution analysis with discrete wavelet transform excellently
detects the occurrence of damage and suggest that the most suitable signals for being analyzed are the relative
interstory drifts from vertically adjacent instruments. Analysis of the square coefficients of the first level of
decomposition of drift signals led to valuable results for space and time locating the damage and also led to
very reasonable indicators of the magnitude and intensity of it.
Especial findings in the VN7SH building indicate that non-reported minor damage occurred before major
damage took place, suggesting that the proposed parameters could be useful in damage monitoring
instrumented structures.
Finally, implications of the presented theory on other fields of study are discussed and futurist remote online
damage monitoring system is presented as a look beyond the current tendencies on structural health
monitoring programs.

109

REFERENCES

xxxxxxxxxxxxxx NEED TO CHECK REFERENCES xxxxxxxxxxx


Corbin M, Hera A and Hou Z (2000). Lacating damage regions using wavelet approach. Proceedings of the
Fouteenth Engineering Mechanics Conference, ASCE. Department of Civil Engineering, University of Texas
at Austin. May 22-24.
Gurley K and Kareem A (1999). Applications of wavelet transforms in earthquake, wind and ocean
engineering. Engineering Structures, Vol. 21, pp. 149-167.
Hou Z, Noori M and Amand R St (2000). Wavelet-based approach for structural damage detection. Journal
of Engineering Mechanics, ASCE. Vol. 126, No. 7, July. pp. 677-683.
Kijewski T and Kareem A (2003). Wavelet transforms for system identification in civil engineering.
Computer-Aided Civil and Infrastructural Engineering. Vol. 18. pp. 339-355.
Kim H and Melhem H (2004). Damage detection of structures by wavelet analysis. Engineering Structures.
Vol. 26. pp. 347-362.
Kitada Y (1998). Identification of nonlinear structural dynamic systems using wavelet. Journal of
Engineering Mechanics. Vol. 124, No. 10, October. pp. 1059-1066.
Koji S, Trifunac M D and Anderson J C (1984). A post earthquake response analysis of the Imperial
County Services building in El Centro. University of Southern California, Department of Civil Engineering,
Technical Report No. CE 84-02.
Kumar P and Foufoula-Georgiou E (1994). Wavelet analysis in Geophysics: An introduction. Wavelet in
Geophysics, Foufoula-Georgiou E and Kumar P (eds.), Academic Press. pp. 1-43.
Misiti M, Misiti Y, Oppenheim G and Poggi J M (2000). Wavelet toolbox. For use with Matlab. Users
Guide, Ver. 2, The Math Works Inc.
Ovanesova A V and Surez L E (2004). Applications of wavelet transform to damage detection in frame
structures. Engineering Structures. Vol. 26. pp. 39-49.

110

Rezai M, Rahmatian P and Ventura C (1994). Seismic data analysis of a seven-storey building using
frequency response function and wavelet transform. Proceedings of the NEHRP Conference and Workshop
on Research on the Northridge, California Earthquake of January 17, 1994. Vol. III. pp. 421-428.
Sone A, Yamamoto S, Nakaoka A and Masuda A (1995). Health monitoring system of structures based on
orthonormal wavelet transform. Seismic Engineering, ASME. Vol. 312. pp. 161-167.
Sun Z and Chang C C (2002). Structural damage assessment based on wavelet packet transform. Journal of
Structural Engineering, ASCE. Vol. 128, No. 10, Octuber 1. pp. 1354-1361.
Todorovska M I and Hao T Y (2003). Information granulation and dimensionality reduction of seismic data
vibration monitoring data using orthonormal discrete wavelet transform for possible application to data
mining. University of Southern California, Department of Civil Engineering, Technical Report No. CE 0302.
Vincent H T, Hu S-L H and Hou Z (1999). Damage detection using empirical mode decomposition method
and a comparison with wavelet analysis. Proceedings of the 2nd International Workshop on Structural Health
Monitoring. Stanford University, Stanford, California. September 8-10.
Wang Q amd Deng X (1998). Damage detection with special wavelets. International Journal of Solids and
Structures. Vol. 36. pp. 3443-3468.
xxxxxxxxxxxxxx NEED TO CHECK REFERENCES xxxxxxxxxxx

111

You might also like