IABSE Conference Guangzhou 2016 May 8-11, 2016

Bridges and Structures Sustainability - Seeking Intelligent Solutions

Numerical Investigation on the Effects of Local Damage to the Dynamic

Properties of Buildings Using Limited Vibration Data

Kristian CARINGAL Andres ORETA

Graduate Student Professor
De La Salle University De La Salle University
Manila, Philippines Manila, Philippines
kristian_caringal@yahoo.com andres.oreta@dlsu.edu.ph
Kristian Caringal, born 1994, is Andres Oreta, born in 1961,
currently a graduate student of received his doctor of engineering
civil engineering from De La degree from Nagoya University,
Salle University Manila, Japan. He is currently a Professor
Philippines. at De La Salle University
Manila, Philippines.

Summary
The challenge in structural health monitoring and damage detection is how to use the limited number
of sensors on a building to assess the condition of the structure at any point in time. A case study
simulating application of different levels of isolated local damage to the different floors were done to
determine the corresponding dynamic properties and to observe the sensitivity of the dynamic
properties to local changes using a several shear building models. Dynamic condensation was also
applied to the model to portray limited vibration data. The results from the condensed 2 DOF model
resulting into the two lowest natural frequencies in the structure can only be used in detecting the
damage from the first unto the fifth floor in the building. With that, a change of 5% in the frequency
can mean up to a 50% local damage to the stiffness in a single floor. These findings can be used to
estimate the damage present in a building and provide warning signals for the building owner.
Keywords: Structural Health Monitoring, Modal Analysis, Damage Detection, High-rise building.

1. Introduction
Structural health monitoring can be started as soon as structure construction is completed and can be
ended at its failure. Due to costly and time consuming inspections when a structure is damaged, many
researchers are engaged in studying and developing different techniques on how to use and maximize
data from recorded accelerographs and apply it to different types of structures as part of advance data
processing and interpretation in structural health monitoring. Sensors or instrumentation of buildings
is very helpful in rapid condition assessment as it provides a real time monitoring of the structure [1].
In addition, the study of response records from instrumented structures can be helpful in identifying
performance problems of a structure and eventually will help engineers in formulating a solution to
the problem [2]. According to [3], specific regulations and guidelines on installing accelerograph
have been published in some codes in different countries. Although instrumentation of these
accelerographs on every floor of a structure is impossible because of its high cost. So the usual
optimum or reasonable installation guideline taken is to place it in three locations.
Numerous researchers have tried do develop and improve the way of detecting damage in a structure.
A common way for most of researchers in starting their process is by having a finite element model
of the undamaged state then comparing it to the damaged state. In the study of [4], he was able to
estimate the severity of the damage by matrix disassembly technique. The approach was a typical
damage detection method - localizing the damage then quantifying it. A similar approach was done
by [5] as they developed a damage detection method and tested it in trying to detect and quantify
member damage of a 10 story steel frame by introducing three different cases of structural changes:
 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

single member stiffness reduction alone, single member stiffness reduction along with mass reduction,
and multiple member damage. They compared the residual force vector to the damaged DOF and
found out that the effect of the damage is minimal to the natural frequency of the system. Some spikes
in their comparison were present and filtering out false signals is eminent but if accurate frequency
response measurements are available, it can be easily identified which members are actually damaged.
The use of accelerometer or other kinds of instrumentations can be done to monitor the changes of
dynamic properties such as frequency and period of a building through time with ambient condition
or a post-event evaluation. Relationship between structural and dynamic properties can be
significantly used to analyze the health of a structure. The relationship between natural frequency and
stiffness is a direct proportionality. A decrease in the natural frequency of a certain element will
correspond to decrease in the stiffness property. Given the basic theory of relationship between
natural frequency and stiffness having a direct proportionality, [6] says that a change of 5% in the
natural frequency is eminent to consider an element to be damaged with confidence. However as
discussed further in his study, a change in frequency alone is not enough to conclude that there is a
damage existing in an element since there are frequency shifts due to changes in ambient conditions.
Discussed in this paper is a distinct relationship between the changes in dynamic properties and local
damage through trends in different cases of deviations, which will eventually lead to identifying a
damaged region in the structure given a change in dynamic properties of the building. Numerical
study was used to investigate the effects of local damage on the dynamic properties of buildings using
the simulated vibration data that can be located at three points.

2. Damage Detection
To be able to represent the changes in local damage in the structure, the building model used is a
shear-building model. The intact structure is modeled using Finite Element Modeling (FEM) to be
able to precisely determine the properties of the building. Local damage will then be applied to this
undamaged condition to be able to generate the properties of the damaged structure.
In evaluating the damage present in the structure, an equation of motion for damped case given n
Degree of Freedom (DOF) is given by
+ + = ( ) (1)

Where M and K are n x n mass and stiffness matrices, and u are n x 1 acceleration and displacement
vector, respectively and f (t) is given as n x 1 force vector. Under free vibration, equation (1) can be
transformed into
( ) =0 (2)
Where is the ith mode eigenvalue or the natural frequency i and i is the eigenvector or the
normalized mode shape with respect to the i . Equations (1) and (2) can be used to theoretically
solve for the natural frequencies of a structure given its mass and stiffness properties. On the other
hand, an experimental way of solving the same problem can be done using the vibration data from
the structure given an input motion. The method used in this paper for the experimental solution is
Newmarks direct integration in addition to applying a Fast Fourier Transform (FFT) to the obtained
acceleration data in solving for the dominant frequency.
To be able to simplify the model assuming that only three points of acceleration response is available,
the modified Condensed Model Identification Recovery with Dynamic Condensation (CMIR-DC)
equation proposed by [7] will be used. However, since the acceleration at the bottom floor will most
likely correspond to the ground motion vibration, a 2DOF lumped mass model will represent better
the structure depicting limited vibration data points.
 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

The standard equation of motion is given in equation (3) where in subscript s denotes the secondary
DOFs to be condensed while subscript p denotes the remaining primary DOFs in the FE model.
0
+ = (3)
0
Under free vibration, equation (3) can be transformed into a form similar to equation (2):
0
=

(4)
0
Given this, the condensed mass matrix and condensed stiffness matrix the system can be
calculated as follows:
!
= (5)
" =#
" + (6)
Where can be determined from Gauss-Jordan Elimination of secondary DOFS
Eventually, a relationship between the applied local damage to each floor can be used to determine
the changes in the dynamic properties of the structure. The damage detection model is shown in
Figure 1.

Fig 1: Damage detection model

3. Case Study
As suggested by most instrumentation installation guidelines, the location of the three minimum
acceleration is to be placed in the top, middle and bottom. However, accelerometer at the bottom will
mostly have readings with respect to the ground movement while the accelerometer placed at the top
may not be able to have efficient readings to determine the dynamic properties of every story in the
building. It was investigated whether placing the 2nd accelerometer at the middle would really provide
the sufficient data needed for structural health monitoring. Four cases of shear building models were
inspected: Case 1 15 story shear building, Case 2 20 story shear building, Case 3 25 story shear
building, Case 4 30 story shear building. As an example, case 2 will be discussed for determining
the optimum location for the 2nd accelerometer and for detecting damage in the structure. The
structural properties of the model are shown in Table 1.
 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

Table 1: 20 story shear building structural properties

Case 2 - 20 story shear building
k1 (N/mm) 200000 m1 (kg) 1500 c1 100 As part of the data processing, a blast
k2 (N/mm) 200000 m2 (kg) 1500 c2 100 force was applied to the structure as a
k3 (N/mm) 200000 m3 (kg) 1500 c3 100 dynamic loading with an intention of
k4 (N/mm) 200000 m4 (kg) 1500 c4 100 inducing vibration data in the
k5 (N/mm) 200000 m5 (kg) 1500 c5 100 structure numerically. This
k6 (N/mm) 180000 m6 (kg) 3000 c6 200 application is not to be treated as a
k7 (N/mm) 180000 m7 (kg) 3000 c7 200 base excitation. The plot of loading is
shown in Figure 2. The time domain
k8 (N/mm) 180000 m8 (kg) 3000 c8 200
for the whole structure given 20 floors
k9 (N/mm) 180000 m9 (kg) 3000 c9 200
is generated through Newmarks
k10 (N/mm) 160000 m10 (kg) 4500 c10 300
direct integration. The resulting data
k11 (N/mm) 160000 m11 (kg) 4500 c11 300
is shown in Figure 3.
k12 (N/mm) 160000 m12 (kg) 4500 c12 300
k13 (N/mm) 160000 m13 (kg) 4500 c13 300 Moreover, The number of vibration
k14 (N/mm) 150000 m14 (kg) 6000 c14 400 data will be equal to the number of
k15 (N/mm) 150000 m15 (kg) 6000 c15 400 floors of the shear building model as
k16 (N/mm) 150000 m16 (kg) 6000 c16 400 one set of vibration data is available
k17 (N/mm) 150000 m17 (kg) 6000 c17 400 per floor. This is portrayed in Figure
k18 (N/mm) 130000 m18 (kg) 7500 c18 500 4.
k19 (N/mm) 130000 m19 (kg) 7500 c19 500
k20 (N/mm) 130000 m20 (kg) 7500 c20 500

Fig 2: Plot of Blast Force Applied

Blast
TimeForce
Domain Fig 3: Time Domain of First Floor
Acceleration (m/s2)

15 0.001
Force (N)

10
0.0005
5
0 0
0 0 0.5 501 1.5
100 2 150
-0.0005 Time Time
(s) (s)

Fig 4: Sample Shear building model with vibration data

 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

By applying Fast Fourier Transform (FFT)

Frequency Domain to the time domain data, the frequency
2.50E-01 domain can be obtained. An example is
shown in Figure 5. The peak frequency
2.00E-01
readings were then obtained and were
FFT Magnitude

1.50E-01 converted into rad/s. The frequency

readings from the vibration data
1.00E-01 (experimental) were compared to the
5.00E-02 frequency results through modal analysis
(theoretical). In the analysis done, the
0.00E+00 experimental data are expected to be
0 0.5 1 1.5 2
correct if it is equal to the theoretical data
FFT Frequency (Hz) rounded off within two decimal places.
Fig 5: Frequency Domain of
First Floor

Several damage cases represented by change in the first floor stiffness were simulated in the shear
building model. A decrease in 10 to 70% on K was made. Multiple simulations were also done by
varying the position of accelerometer starting from the bottom to the top most floor in the building
model. This is to verify the capacity of a simulated accelerometer in reading correct frequency data.
Overall in the four cases analyzed, the 2nd accelerometer placed at the of the total height of the
building provides the most consistent results in obtaining correct frequency readings from the
vibration data as compared to the maximum correct readings obtained at any single floor. This
analysis result was also validated statistically by plotting a Box and Whisker Plot as shown in Figure
6. The statistical data also shows that the accelerometer placed at the of the total height of the
building has a closer mean to the maximum correct frequency reading at any floor in the structure. It
also has the lowest standard deviation among all the positions analyzed.

Fig 6: Box & Whisker Plot of 20 story frequency readings

Moreover, it was also observed that using the 2nd and 3rd accelerometer readings simultaneously can
greatly increase the number of correct frequency readings captured by the vibration data in both the
damaged and undamaged state of the building. As shown in Figure 7, combining the two points of
measurements of vibration data can exceed the maximum readings from any single floor.
 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

14

# of frequency readings
12
10 0.75h
8
6
4 top
2
0 max from any single
floor
0.75h + top

Damage at first floor

Fig 7: Frequency Readings from 2nd and 3rd Accelerometer

The same 20 story shear building model was used to investigate the effect of local damage to the
dynamic properties of building by dynamic condensation and modal analysis. In the condensation
process, the 1st and 10th Degree of Freedom (DOF) was considered as the primary while the rest were
considered as slaves. After running the analysis, it was found out that w1 = 0.493 and w2 = 1.4324.
These values are considered to be the natural frequencies of the undamaged state of the shear-building
model. Simulations were done by applying different isolated damage severity specifically ranging
from 10% to 90% damage to each of the columns in the model. Combination of two damaged columns
was not investigated in this paper.
The frequency readings acquired from the vibration data were compared to the natural frequencies
obtained through modal analysis. There were two ways done in acquiring the frequency readings
from the vibration data. The first is by using the findpeaks function of MATLAB in finding the
highest FFT magnitude available in the FFT data while the second is manually searching for the
highest values in the FFT magnitude data and looking for its corresponding FFT frequency in every
floor. In the first method presented using findpeaks function, an average of only 14 accurate peaks
out of the 20 available peaks were captured with an error of 0.07%. On the other hand, the manual
selection can provide accurate readings with no error although it was found out that the two lowest
frequencies are not obtainable through this method. Since the obtained two natural frequencies
through dynamic condensation are the two lowest frequencies and are not present in the vibration
data, the data used for comparison done in the analysis are from findpeaks method.
The two natural frequencies obtained from the undamaged state in the condensed model were
compared to the isolated damage column simulated at every floor in the shear-building model. It was
observed that the changes are consistent only up to the 5th floor as shown in Figure 8. This is because
the frequencies observed are only the two of the lowest theoretical natural frequencies. On the other
hand, the data on the upper floors show a lot of inconsistencies, which is why relationship is hard to
establish. There are some floors especially near the top most that does not generate peak frequencies
especially given high damage severity. Furthermore, there is little to none changes in the frequencies
in the upper floors even with excessive local damage. Such occurrence defies the theory of the basic
relationship of structural to dynamic property of a structure. Given this, it was inferred that the lower
modes of frequencies from the condensed data are not to be used in establishing a relationship in the
measured dynamic properties of the upper floors.
 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

Damage severity vs frequency change in w1

0.25

0.2
%change in w1

0.15 K1
K2
0.1 K3
K4
0.05
K5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Damage Severity
Fig 8: Damage Severity vs Frequency Change in w1 of First 5 Floors

Analyzing only the data up to the 5th floor, Table 2 shows the average changes in the dynamic property
of the structure given certain applied local damage. The results are very consistent and can be used
to estimate the damage in the structure given changes in observed natural frequency.

Table 2: Average Frequency Change with Respect to Damage of First 5 Floors

Damage
10% 20% 30% 40% 50% 60% 70% 80% 90%
Severity
w1 % change 0.38% 0.84% 1.43% 2.20% 3.25% 4.78% 7.18% 11.53% 21.79%
w2 % change 0.24% 0.53% 0.89% 1.37% 2.00% 2.90% 4.26% 6.56% 11.12%

Damage severity vs frequency change in w1

0.25
0.2 y = 1.6431x4 - 2.1249x3 + 0.9448x2 - 0.0849x + 0.0016
R = 0.9979
%change in w1

0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-0.05
Damage Severity

Fig 9: Trend Line for K1 damage vs w1 frequency change

An example of the equation that was modeled given the data is shown in Figure 9. A trend line was
created to come up with a polynomial equation in the 4th degree. The coefficient of determination, R2
is 0.9979, which is very close to 1 indicating that the modeled equation fits the data well. Several
equations may be modeled to come up with a relationship between damage and frequency change
although it is limited up to the first 5 floors only due to data consistency with respect to the condensed
frequencies obtained.
 IABSE Conference Guangzhou 2016 May 8-11, 2016
Bridges and Structures Sustainability - Seeking Intelligent Solutions

4. Conclusion
The extent of the use of limited vibration data was explored to determine the effect of local damage
to the dynamic properties of a building was investigated. Numerous shear-building models were
modeled to establish relationship between local damage and natural frequency change. Several key
observations were made in the study:
Contrary to the common suggestion in installing the three minimum accelerometers at the top, middle
and bottom; it was examined that the supposed to be middle accelerometer should be placed at of
the total height of the building. By doing so, it can provide more correct frequency readings as
compared to placing it exactly at the mid height of the building. This was validated by having a case
study of shear building models with 15, 20, 25, and 30 story. It is recommended that a different
approach may be used to higher story building models or other structural systems.
The first two natural modes representing the 2DOF-condensed model were used in detecting the
damage from the first unto the fifth floor in the building. With that, a change of 5% in the frequency
can mean up to a 50% local damage to the stiffness in a single floor. To analyze the response further,
a model was formed in establishing a relationship between the damage severity applied to an isolated
column with respect to its corresponding frequency change by comparing the damaged and
undamaged state of the structure. These findings can be used to estimate the damage present in a
building even with limited vibration data and provide warning signals as part of structural health
monitoring.

5. References
[1] SAFAK, E., & HUDNUT, K. Real-time structural monitoring and damage detection by
acceleration and GPS sensors. 8th US National Conference on Earthquake Engineering. San
Francisco, California, 2006.
[2] CELEBI, M. Seismic instrumentation of buildings (with emphasis on federal buildings),
Special GSA/USGS project. California, 2002.
[3] WANG, G., WANG, W., AAFSHAR, K., & DOJCINOVSKI, D. Seismic instrumentation of
high-rise buildings. Progress in Natural Science 19, 2009, 223-227.
[4] YANG, Q. W. A numerical technique for structural damage detection. Applied Mathematics
and Computation, 215, 2009, 2775-2780.
[5] GE, M., & LUI, E. M. Structural damage identification using system dynamic properties.
Computers and Structures, 83, 2002, 2185-2196.
[6] SALAWU, O. S. Detection of structural damage through changes in frequency: a review.
Engineering Structures, 19, 1997, 718-723.
[7] Koh, C. G., Tee, K. F., & Quek, S. T. Condensed Model Identification and Recovery for
Structural Damage Assessment. Journal of Structural Engineering, 2006, 2018-2026.