Professional Documents
Culture Documents
2018 - IALCCE 2018 - Parametric Identification and Damage Assessment of Real Buildings
2018 - IALCCE 2018 - Parametric Identification and Damage Assessment of Real Buildings
F.K. Huang
Department of Water Resources and Environmental Engineering, Tamkang University,
New Taipei City, Taiwan
ABSTRACT A Frequency-Domain recursive hybrid Genetic Algorithm was developed by the author in 2015.
The time history of the measurement is divided into a series of time intervals, and then the model of equivalent
linear system is employed to identify the modal parameters of the system for each time interval. The differential
equation can be transformed into the frequency domain by finite Fourier transform and the response in the
frequency domain can be solved by algebraic equations instead of differentials equations for each time interval.
In this paper, this new identification strategy is applied to the identification of two real buildings to monitor the
change of modal parameters. The parameters of these structures are identified using the response records of the
structure under the attack of earthquakes of different intensities. By employing the Maximum Softening Index
and Approximate Story Damage Index, we can determine the damage states of these buildings.
245
domain is developed. The differential equation can
be transformed into the frequency domain by finite
Fourier transform. The numerical time for solving the
algebraic equations can be reduced tremendously com-
pared to the time for solving the differential equation.
The proposed algorithm has been explored by com-
paring the results of the predicted response with the
measured response for both the simulated SDOF linear
system and the simulated MDOF linear system with or
without noise contamination (Wang & Huang, 2015).
In this paper, this new identification strategy will first
be applied to the identification of the office build-
ing of National Center for Research on Earthquake
Engineering (NCREE) located in Taipei subjected to
4 sets of earthquake including the Chi-Chi earthquake.
Then, the algorithm will also be applied to the iden-
tification of building of the Shih Hsien Elementary Figure 1. Structure of hybrid GA.
School located in Chai-Yi City. Based on the identifi-
cation results, damage assessment will be performed displacement and velocity for each time interval as
thereafter. well. The differential equation can be transformed into
the frequency domain by finite Fourier transform and
the response in the frequency domain can be solved by
2 HYBRID GENETIC ALGORITHM algebraic equations instead of differentials equations.
Genetic algorithm is a stochastic search technique 3.1 Identification model of a MDOF system
based on natural selection and genetics, developed
by Holland (1962). Genetic algorithms model natural In this section, we will develop the identification
processes, such as selection, recombination, mutation, model of a multiple degree of freedom system with
migration, and competition. The algorithms work on multiple outputs or measurements. The motion equa-
populations of individuals instead of a single solu- tion of a MDOF system subjected to a single excitation
tion. In this way, the search is performed in a parallel can be transformed into modal equation through mode
manner. However, better results can be obtained by superposition as
introducing multiple subpopulations. Every subpopu-
lation evolves over a few generation isolated (like the
single population GA) before one or more individu-
where usm is the modal displacement in mode m at the
als are exchanged between subpopulation using the
sth DOF, and Psm the effective participation factor in
mechanisms of migration and competition. The mul-
mode m at the sth DOF associated with the ground
tipopulation GA models the evolution of a species in
motion üg
a way more similar to nature than single population
GAs are inherently slow and are poor at hill-
climbing. In order to compensate the computational
inefficiency in hill-climbing when the solution yielded
by GA approaches the optimal value, a local search
operator, the Gauss-Newton method is merged to the in which [M ] is the mass matrix, {φm } the mode shape
GA strategy. This method is performed after com- in mode m, and {l} the ground influence coefficient
pleting the evolution process of every 10 genera- matrix with elements 0 and 1. The finite Fourier trans-
tions. Accordingly, a new hybridization of a GA with form of relative response of usm , u̇sm , and üsm over a
Gauss-Newton method is formed. Figure 1 shows the finite record length of T or each non-overlapping time
structure for such a hybrid GA. interval can be defined as
3 FREQUENCY-DOMAIN RECURSIVE
HYBRID GENETIC ALGORITHM
246
T
where Usm (ω) is the finite Fourier transform of the
modal displacement in mode m at the sth DOF. Trans-
forming equation (1) into frequency domain by using
the terms defined in equation (3∼5), yields
T
where Usm (ω), the Fourier transform of the relative
displacement, is denoted as Usm (ω) for simplicity.
The parameters vsm and dsm are the differences in the
mth modal velocity and displacement at the sth DOF
between the beginning and the end of the record seg-
ment of duration T . The measured response in this
paper is the relative acceleration and the associated
Fourier transform of üsm can be derived as
Combining the above modal contributions will yield Figure 2. Structure of frequency-domain recursive hybrid
the following measurement expression for the trans- GA in the for MDOF system.
formed relative acceleration
into non-overlapping time intervals or time windows
by setting the number of time instants or sampling
points, na, in the beginning. In order to account for the
effect of the difference of velocity and displacement
between the beginning and the end of the each inter-
val, vsm and dsm are also implemented as the parameters
In order to optimize the system, error function in to be identified in addition to the system parameters
term of transformed response for each measurement 2
2ξm ωm , ωm and Psm . Then the model of equivalent lin-
is defined in such a way that it is quadratic in terms of ear system in the frequency domain and the recursive
the parameters. hybrid GA is employed to identify the parameters of
the system and the differences of displacement and
velocity for each time interval. The response in fre-
quency domain can be computed using equation (7).
The best five individuals or system parameters are
replicated and used as the individuals of the initial
population of next interval. This process will be con-
tinued until the measurements of the rest of the time
where Y (i, j) and V (i, j) are the Fourier transforms of
interval are implemented. Figure 2 shows the pro-
the measured response and identified response for the
cedure of the proposed Frequency-domain recursive
ith sampled frequency with respective to the jth mea-
hybrid genetic algorithm where the red dashed box
surement. The Notations Re and Im represent the real
represented the evolution process of hybrid genetic
and imaginary parts of the response. If there are K
algorithm illustrated in Figure 1.
measurements, the root mean square error function,
denoted as the error index, E.I ., is expressed as a
function of the error indices of the measurements
4 DAMAGE INDICES
247
of structure, energy absorption and cyclic fatigue in Obviously, the value of ASDI(l, j) can be easily
estimating the damage level. obtained based on only one set of modal parameters (ωj
and φ j ). The value of the ASDI(l, j) is shown between
0 (no damage) and 1 (collapse) in terms of the modal
4.1 Maximum softening index parameters of the jth mode.
Dipasquale and Cakmak (1990) define the maximum
softening for the one-dimensional case, where only
the fundamental modal frequency is considered. The 5 IDENTIFICATION AND DAMAGE
index is given by ASSESSMENT OF REAL BUILDINGS
248
Figure 3. Variation of modal frequency in the longitudinal Figure 5. Variation of maximum softening index in the
direction for NCREE building. longitudinal direction for NCREE building.
Figure 4. Variation of modal frequency in the transverse Figure 6. Variation of maximum softening index in the
direction for NCREE building. transverse direction for NCREE building.
249
Figure 9. Variation of ASDI index of the third floor in the Figure 12. Variation of ASDI index of the top floor in the
longitudinal direction for NCREE building. transverse direction for NCREE building.
Figure 10. Variation of ASDI index of the top floor in the Figure 13. Variation of maximum softening index in the
longitudinal direction for NCREE building. longitudinal direction for SHES building.
Figure 11. Variation of ASDI index of the third floor in the Figure 14. Variation of maximum softening index in the
transverse direction for NCREE building. transverse direction for SHES building.
equation (12). Similarly, ASDI(2,1) reflect the damage Three-component seismographs are installed on the
state of upper level from the third floor to the top floor first floor, third floor and the top floor of the building.
since we can use the effective participation factors of Accelerograms collected during earthquakes can be
the third floor to calculate the term φ in equation utilized to monitor the change of the structural param-
(12). Figures 9 and 10 summarized the values of the eters. Same four sets of strong motion records are
ASDI(1,1) index and ASDI(2,1) index in the longitudi- analyzed. Single-input-multiple-output model is also
nal direction during the four earthquakes considered. utilized to perform the identification of modal parame-
The average value for each earthquake is computed ters of the structure. Again, by monitoring the variation
and plot as a red dashed line in the figure for the of the identified parameters, the damage state of the
following discussion. ASDI(1,1) index and ASDI(2,1) structure will then be evaluated.
index increased to 7.4% and 5.53% under the attack of The maximum softening index of the first mode,
Chi-Chi earthquake and dropped to 1.96% and 1.72% DM ,1 , in the longitudinal direction is plotted in Fig-
under the attack of the 1022 earthquake. These two ure 13. The index is 14.73% during the Chi-Chi
indices became 4.1% and −0.91% under the attack of earthquake and decreased to 10.33% during the 1022
the last earthquake. Figures 11 and 12 summarized the earthquake. The index reduced to 4.82% at the end of
values of the ASDI(1,1) index and ASDI(2,1) index in 614 earthquake. The result in the transverse direction
the transverse direction. These two indices increased to is illustrated in Figure 14. The index of the first mode
15.65% and 11.56% under the attack of the last earth- increased to 19.74% at the end of 1022 earthquake.
quake. This result indicates that the building has slight Finally, this index returned to 11.97%.
damage in this direction. Figures 15–16 summarized the values of the
ASDI(1,1) index and the ASDI(2,1) index in the
longitudinal direction during the four earthquakes
5.2 Identification and damage assessment of Shih considered. The value of the ASDI(1,1) index and
Hsien Elementary building the ASDI(2,1) index increased to 16.96% and 3.09%
under the attack of the 614 earthquake. Figures 17–18
The other building identified here is the six-story rein- summarized the values of the ASDI(1,1) index and
forced concrete building of the Shih Hsien Elemen- ASDI(2,1) index in the transverse direction. These two
tary School (SHES) located in Chai-Yi City, Taiwan. indices increased to 27.07% and 19.18% under the
250
Fourier transform. The numerical time for solving the
algebraic equations can be reduced tremendously com-
pared to the time for solving the differential equation.
The algorithm has been applied to the identification
of buildings of NCREE and Shih Hsien Elemen-
tary School. The damage assessments were performed
based on the identification results:
1. From the identification results of National Earth-
Figure 15. Variation of ASDI index of the third floor in the quake Engineering Research Center office build-
longitudinal direction for SHES building. ing, we can see that the fundamental frequency
in the longitudinal direction during the 614 earth-
quake decreased to about 23.856 rad/sec in the end,
which is about 3% lower than the value 24.6 rad/sec
yielded during the first earthquake. In the trans-
verse direction, the modal frequency decreased
from 23.83 rad/sec to 21.53 rad/sec
2. The maximum softening index DM for the office
building of NCREE was evaluated using the results
of the identified modal frequency. This index in
Figure 16. Variation of ASDI index of the top floor in the the transverse direction is 9.29% after the struck
longitudinal direction for SHES building. of the 614 earthquake. This indicates a quite slight
damage state. The values of the ASDI(1,1) index
and ASDI(2,1)index in the transverse direction.
increased to 15.65% and 11.56% under the attack of
the last earthquake. These values also indicate that
the building has slight damage in this direction.
3. The maximum softening index DM for the SHES
building was also evaluated using the results of the
identified modal frequency. This index in the trans-
verse direction is 11.97% after the struck of the
Figure 17. Variation of ASDI index of the third floor in the last earthquake. This also indicates a quite slight
transverse direction for SHES building. damage state in this direction . The values of the
ASDI(1,1) index and ASDI(2,1)index in the trans-
verse direction. increased to 27.07% and 19.18%
under the attack of the last earthquake. These values
again indicate that the building has slight damage
in this direction.
4. The MAC index seems to be not able to capture the
damage state of the structure since the values are
very close to 1.
251
Wang, G.S. & Lin, H.H. 2005. Structural parameter identi- Wang, G.S. & Huang. F.K. 2015. Frequency-Domain Recu-
fication of torsionally-coupled buildings. Journal of the sive Hybrid GA to the Identification of a Real Build-
Chinese Institute of Civil Engineering and Hydraulic ing. 15th Structural Engineers Congress, Singapore,
Engineering 17(2): 281–291. Singapore.
Wang, G.S. 2009. Application of hybrid genetic algorithm Wang, J.F., Lin, C.C. & Yen. S.M. 2007. A story damage
to system identification. Structural Control and Health index of seismically-excited buildings based on modal
Monitoring 16(2): 125–153. frequency and mode shape. Engineering Structures 29:
Wang, G.S. Tsai, Y.C. & Huang. F.K. 2013. Hybrid genetic 2143–2157.
algorithm in frequency domain to structural dynamic
parameter identification. 11th International Conference
on Structural Safety & Reliability, New York, USA.
252