Mathematical and Computer Modelling 53 (2011) 1798–1809

Contents lists available at ScienceDirect

Mathematical and Computer Modelling

journal homepage: www.elsevier.com/locate/mcm

Structural damage detection using an efficient correlation-based index

and a modified genetic algorithm
M. Nobahari a , S.M. Seyedpoor b,∗
a
Department of Civil Engineering, Islamic Azad University, Neyshabur Branch, Neyshabur, Iran
b
Department of Civil Engineering, Shomal University, Amol, Iran

article info abstract

Article history: An efficient optimization procedure is proposed to detect multiple damage in structural
Received 1 March 2010 systems. Natural frequency changes of a structure are considered as a criterion for damage
Received in revised form 29 December 2010 presence. In order to evaluate the required natural frequencies, a finite element analysis
Accepted 29 December 2010
(FEA) is utilized. A modified genetic algorithm (MGA) with two new operators (health
and simulator operators) is presented to accurately detect the locations and extent of the
Keywords:
eventual damage. An efficient correlation-based index (ECBI) as the objective function for
Structural damage detection
Natural frequency
the optimization algorithm is also introduced. The numerical results of two benchmark
Finite element analysis examples considering the measurement noise demonstrate the computational advantages
Modified genetic algorithm of the proposed method to precisely determine the sites and the extent of multiple
Efficient correlation-based index structural damage.
© 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Many structural systems may experience some local damage during their functional age. In order to enhance the
efficiency of the structures, it is necessary to properly identify the damage sites and their extent, and then rehabilitate
them. Therefore, the damage identification problem has recently attracted significant attention. Many methods have been
introduced to correctly determine the location and extent of structural damage. One type of method is based on employing
optimization algorithms to solve damage identification problems. In a study, Koh and Dyke [1] have determined the
location and extent of multiple damage by iteratively searching for a combination of structural responses that maximizes
a correlation coefficient named the multiple damage location assurance criterion (MDLAC) via a genetic algorithm (GA).
Damage detection by a hybrid technique consisting of a real-parameter genetic algorithm and grey relation analysis has been
presented by He and Hwang [2]. They used first a grey relation analysis to exclude impossible damage locations such that
the number of design variables could be reduced. Second, a real-parameter genetic algorithm was combined with simulated
annealing for finding the actual damage. A fault diagnosis method based on a GA and a model of the damaged structure has
been proposed by Vakil-Baghmisheh et al. [3]. The damage identification of a beam-like structure has been formulated as an
optimization problem, and a GA has been employed to find the damage location and extent by minimizing the cost function
which is based on the difference of measured and calculated natural frequencies. A two-stage method of determining the
location and extent of multiple structural damage by using an information fusion technique and a GA has been presented
by Guo and Li [4]. In the first stage, the damage is localized by using evidence theory, and then a micro-search genetic
algorithm (MSGA) has been proposed to determine the damage extent. A two-stage method for determining the location

∗ Corresponding author. Tel.: +98 121 2203726; fax: +98 121 2203726.
E-mail address: [email protected] (S.M. Seyedpoor).

0895-7177/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mcm.2010.12.058
 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809 1799

and extent of multiple structural damage by combining the adaptive neuro-fuzzy inference system (ANFIS) and particle
swarm optimization (PSO) has been proposed by Fallahian and Seyedpoor [5]. A multi-stage particle swarm optimization
(MSPSO) for accurately detecting structural damage has been introduced by Seyedpoor [6]. In another study, Begambre
and Laier [7] proposed a hybrid particle swarm optimization–simplex algorithm (PSOS) for structural damage identification
using frequency domain data.
In this study, a modified genetic algorithm (MGA) is introduced to identify multiple structural damage. This modified
version of the GA has two new tools, namely a health operator (HO) and a simulator operator (SO), in order to recognize the
actual damage correctly and rapidly. An efficient correlation-based index (ECBI) is also proposed as the objective function
for the optimization algorithm. Two illustrative test examples are considered to show the performance of the method.
Numerical results demonstrate that the combination of the MGA and the ECBI can provide a robust tool for determining
the sites and extent of multiple structural damage precisely and quickly.

2. Structural damage detection

Structural damage detection techniques can be generally classified into two main categories [8]. They are dynamic and
static identification methods, requiring dynamic and static test data, respectively. Furthermore, dynamic identification
methods have shown their advantages in comparison with static ones. Among the dynamic data, the natural frequencies of
a structure can be considered to be valuable data. Determining the level of correlation between the measured and predicted
natural frequencies can provide a simple tool for finding the location and extent of structural damage [9]. The parameter
vector used for evaluating the correlation coefficients consists of the ratios of the first nf natural frequency changes 1F due
to structural damage, i.e.
Fh − Fd
1F = , (1)
Fh
where Fh and Fd denote the natural frequency vectors of the healthy and damaged structure, respectively. Similarly, the
corresponding parameter vector predicted from an analytical model can be defined as

Fh − F (X )
δ F (X ) = , (2)
Fh
where F (X ) is a natural frequency vector that can be predicted from an analytic model and X T = {x1 , . . . , xi , . . . , xn }
represents a damage variable vector containing the damage extent (xi , i = 1, . . . , n) of all n structural elements.
Given a pair of parameter vectors, one can estimate the level of correlation in several ways. An efficient way is to evaluate
a correlation-based index, termed the multiple damage location assurance criterion (MDLAC), expressed in the following
form [9]:

1F .δ F (X )2
 T 
MDLAC (X ) = . (3)
(1F T .1F )(δ F T (X ).δ F (X ))
The MDLAC compares two frequency change vectors, one obtained from the tested structure and the other from an
analytical model of the structure. The MDLAC varies from a minimum value 0 to a maximum value 1. It will be maximal when
the vector of analytical frequencies becomes identical to the frequency vector of the damaged structure, that is, F (X ) = Fd .
This concept can be utilized to find a set of damage variables maximizing the MDLAC using an optimization algorithm:

Find X T = {x1 , x2 , . . . , xn }
Maximize : w(X ) = MDLAC (X ) (4)
xi ∈ Rd , i = 1, . . . , n,

where Rd is a given set of discrete values so that the damage extent xi (i = 1, . . . , n) can take values only from this set. Also,
w is an objective function that should be maximized.
Due to the occurrence of damage in a structural element, the element stiffness decreases. Therefore, in many studies
related to the damage identification problem, damage has been simulated by decreasing one of the stiffness parameters of
the element such as the elasticity modulus (E), cross-sectional area (A), moment of inertia (I), and so on. In this study, the
damage variables are defined via a relative reduction of the elasticity modulus of an element as
E − Ei
xi = , i = 1, . . . , n, (5)
E
where E is the initial modulus of elasticity and Ei is the final modulus of elasticity of the ith element.
In this study, before solving Eq. (4) by an optimization method, an initial study was made on the performance of the
MDLAC as an objective function, and for brevity only the final outcomes are provided. The MDLAC as an objective function
for the optimization algorithm is very sensitive to damaged elements while its sensitivity to healthy elements is low. In the
other words, it can find the true locations of the damaged elements; however, it may find a healthy element as a damaged
1800 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809

one. Therefore, in this study a new function having a high sensitivity to healthy elements is presented as
nf
1 − min(fxi , fdi )
obj(X ) = , (6)
nf i=1 max(fxi , fdi )
where fxi and fdi are the ith components of vectors F (X ) and Fd , respectively.
The obj(X ) function can rapidly find the locations of healthy elements when compared to the MDLAC ; however, it is very
probable that it finds a damaged element as a healthy one. Therefore, in this study, a combinational function of Eqs. (3) and
(6), called here the efficient correlation-based index (ECBI), is proposed as
1
ECBI (X ) = (MDLAC (X ) + obj(X )). (7)
2
In the test examples, the performance of the ECBI for damage identification is assessed in comparison with that of the
MDLAC .

3. The modified genetic algorithm

The selection of an efficient algorithm for solving the damage optimization problem is a critical issue. Needing fewer
structural analyses for achieving the global optimum without trapping into local optima must be the main characteristic
of the algorithm. In this study, a modified version of genetic algorithm working with discrete design variables is proposed
to properly solve the damage problem. In this section, the simple genetic algorithm (SGA) is first described, and then the
modified genetic algorithm (MGA) is explained.

3.1. Simple genetic algorithm

The genetic algorithm is a probabilistic search algorithm inspired by the evolutionary laws of living creatures [10]. In this
algorithm, information regarding an optimization problem, such as design variables, is coded into a genetic string known
as an individual (chromosome). Each of these individuals has an associated fitness value, which is usually determined by
the objective function to be maximized or minimized. The genetic algorithm proceeds by taking an initial population, which
is comprised of different individuals with different fitness. The initial population can be randomly generated or it can be
selected by a supervisory method [11]. All individuals are decoded to evaluate their fitness function values. Then, the best
individuals of the current generation (parents) are selected to produce the new generation (offspring) having better fitness
by crossover and mutation operators. The previous individuals are replaced by the new individuals, and this process will be
continued until the convergence criteria are met. Finally, the best individual of the last generation according to its fitness
function is selected as the optimal solution.
The SGA has been shown to be capable of solving various optimization problems via some basic concepts and operators
including coding, initial population, decoding, selection, crossover and mutation [11–13]. However, it has been revealed
from some research works [1,4,14] that the SGA cannot solve damage identification problems properly. Therefore, in this
study two new operators are introduced to enhance the performance of the SGA.

3.2. Health operator

The number of flawed elements in a damaged structure is usually much less than the total number of healthy elements. If
this idea is considered in the optimization process, convergence to the actual damage will be accelerated. Therefore, in this
study a new operator, named here the health operator (HO), is applied to each generation of the standard genetic algorithm
for randomly considering some elements of each structure as the healthy elements. The operation of the HO can be clarified
by a simple example as follows. It is assumed that, for identifying the damaged elements in a structure having 20 total
elements, the initial population consists of 30 individuals. Therefore, each chromosome has 20 genes, and then the total
number of genes will be 20 × 30 = 600. If the probability of applying the HO to the problem is 0.08, then a total number of
0.08 × 600 = 48 genes (elements) in each generation should be assumed to be zero (healthy).

3.3. Simulator operator

The simulator operator (SO) is an operator for improving the best solution of the genetic algorithm, obtained up to current
generation; it needs fewer analyses in comparison with an ordinary generation. Using this new operator, the best individuals
of the two preceding generations are compared and a new individual is created synthetically. Therefore, the SO can work
after the third generation. The procedure of the SO is explained here by a simple example as follows.
It is assumed that in an optimization problem each chromosome comprises 12 genes that can take values 0–0.5 by a step
of 0.05; the best chromosomes of the two last generations are shown in Fig. 1. It is also assumed that chromosome 1 is fitter
than chromosome 2. It can be seen that the values of genes 1, 4, 8, 9, 10 and 12 are different and that the total number of
these genes is nj = 6. By using the SO, the new chromosome is simulated by considering some instructions as follows.
1. The value of a gene having identical values in chromosomes 1 and 2 is not changed. So, in the new chromosome, the
values of genes 2, 3, 5, 6, 7 and 11 do not change.
 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809 1801

Fig. 1. The combination of chromosomes 1 and 2 for creating a new chromosome via the SO.

Table 1
Two different damage cases induced in a 20-element beam.
Case 1 Case 2
Element number Damage ratio Element number Damage ratio

15 0.20 12 0.20
16 0.15 13 0.30
17 0.20 15 0.30

2. If the value of a gene in chromosome 1 is less than the value of the corresponding gene of chromosome 2 (genes 1, 8,
9 and 12), the value of the gene by a probability of 1/nj = 1/6 will be reduced by the step size of our problem (for
this example the step size is equal to 0.05). It should be noted that, if the gene has a minimum value (0), its value is not
changed (gene 9).
3. If the value of a gene in chromosome 1 is more than the value of the corresponding gene of chromosome 2 (genes 4
and 10), the value of the gene by a probability of 1/nj = 1/6 will be increased by the step size of our problem (for this
example it is equal to 0.05). It should be noted that, if the gene has a maximum value (0.5), its value is not changed (gene
10).
By considering the notes above, the new chromosome produced is as shown in Fig. 1. It is observed that the values of
genes 1 and 8 are the same as those of chromosome 1, and they are not decreased because the probability of decreasing is
1/6. Now, the new chromosome is evaluated: if it is fitter than chromosome 1, chromosome 2 is replaced by chromosome 1
and chromosome 1 is replaced by the new chromosome. This operation is continued until the new chromosome simulated
is not fitter than chromosome 1, so at this stage an individual of the current generation randomly selected is replaced by
chromosome 1. Then, a new optimization iteration is performed to produce a new generation. If in the new generation a
better chromosome than those of previous generation is generated the SO will also be active; otherwise, the worst individual
of the current generation is replaced by the best individual of the last generations (elitism selection).
The flowchart of the SGA employing the HO and SO, called here the MGA, is shown in Fig. 2.

4. Test examples

In order to show the capabilities of the proposed approach for identifying multiple structural damage, two illustrative
test examples are considered. The first example is a 20-element cantilevered beam discussed in detail and the second one
is a 31-bar truss element discussed in brief.

4.1. Twenty-element cantilevered beam

The finite element model of a cantilevered steel beam with 20 elements leading to 40 degrees of freedom, shown in
Fig. 3, is considered as the first example [4]. The length, height, and width of the beam are 1.0, 0.01, and 0.01 m, respectively.
The mass density is 7860 kg/m3 and the elasticity modulus is 210 GPa. In this example, the first ten natural frequencies
are used for damage detection. For this example, two damage cases, listed in Table 1, are studied. Here, small damage is
mainly considered, and the stiffness reduction caused by the damage is smaller than 50% [4]. The parameters of the GA are
selected here from the experience of our previous work [11], the guidelines of other works found in the literature [1,14] and
a trial and error method as follows: the probability of crossover pc is 1.0, the probability of mutation pm is 0.002, the health
parameter hp is 0.08, the maximum number of generations ng is 1000, and the population size ps is 20. The convergence of
the algorithm is met when the objective function reaches 0.999999 or the maximum number of generations is attained. The
machine used in this research is a personal computer with a coreTM 2 Duo 2 GHz CPU and a 2 GB memory.
With the mentioned conditions, the damage is identified using the following methods and the results are compared.
(1) SGA with the MDLAC function (SGA + MDLAC).
(2) SGA with the ECBI function (SGA + ECBI).
(3) MGA with the MDLAC function (MGA + MDLAC).
(4) MGA with the ECBI function (MGA + ECBI).
1802 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809

Fig. 2. The flowchart of the MGA.

Fig. 3. The finite element model of a cantilevered beam having 20 elements.

 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809 1803

Table 2a
The damage ratios of the 20-element beam for case 1 using the SGA with the MDLAC.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1 0.05 0.35 0.35 0.40 0.10 20 000

2 0.1 0.00 0.40 0.45 0.50 0.15 20 000
3 0.1 0.00 0.15 0.35 0.50 0.50 0.20 20 000
4 0.15 0.35 0.40 0.50 0.15 20 000
5 0.10 0.40 0.40 0.50 0.15 20 000
6 0.15 0.35 0.45 0.50 0.10 20 000
7 0.05 0.40 0.40 0.50 0.15 20 000
8 0.05 0.35 0.40 0.45 0.15 20 000
9 0.05 0.40 0.40 0.50 0.15 20 000
10 0.10 0.40 0.40 0.50 0.15 20 000
Average 0 0 0.02 0 0 0.07 0.015 0 0 0 0 0 0 0 0.375 0.415 0.485 0 0 0.145 20 000
Actual 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.20 0.15 0.20 0 0 0 –
damage

Table 2b
The damage ratios of the 20-element beam for case 1 using the SGA with the ECBI.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1 0.25 0.15 0.20 20 000

2 0.20 0.20 0.20 0.15 20 000
3 0.1 0.30 0.25 0.30 0.15 20 000
4 0.05 0.15 0.15 0.20 0.10 20 000
5 0.20 0.20 0.20 0.15 20 000
6 0.1 0.25 0.15 0.25 20 000
7 0.1 0.20 0.15 0.20 20 000
8 0.20 0.15 0.20 0.15 20 000
9 0.1 0.25 0.25 0.30 20 000
10 0.20 0.15 0.20 20 000
Average 0.01 0.01 0 0 0.01 0.005 0.01 0 0 0 0 0 0 0 0.22 0.18 0.225 0 0 0.07 20 000
Actual 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2 0.15 0.2 0 0 0 –
damage

Table 2c
The damage ratios of the 20-element beam for case 1 using the MGA with the MDLAC.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1 0.05 0.40 0.40 0.45 0.10 20 024

2 0.05 0.40 0.40 0.45 0.05 20 030
3 0.20 0.15 0.20 2 742
4 0.05 0.40 0.40 0.45 0.15 20 023
5 0.05 0.40 0.40 0.45 0.15 20 020
6 0.05 0.40 0.40 0.45 0.15 20 034
7 0.05 0.40 0.40 0.45 0.15 20 025
8 0.05 0.40 0.40 0.45 0.15 20 023
9 0.20 0.15 0.20 7 085
10 0.20 0.15 0.20 17 318
Average 0 0 0 0 0 0.035 0 0 0 0 0 0 0 0 0.34 0.32 0.375 0 0 0 16 732
Actual 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.20 0.15 0.20 0 0 0 –
damage

In order to consider the stochastic nature of the optimization process using various methods, ten independent sample
runs are made for each damage case. The damage identification results of damage case 1 using four different methods are
given in Tables 2a–2d, respectively. The damage identification results of damage case 2 using four different methods are
given in Tables 3a–3d, respectively.
By comparing the damage detection results of the four various methods, some interesting points can be concluded. It
is revealed that the MGA and the ECBI have better performance when compared to the SGA and the MDLAC, respectively.
The ordinary method SGA + MDLAC method cannot identify the actual damage accurately. The results of the method can be
enhanced when each of the proposed MGA or ECBI is utilized instead of the SGA or MDLAC, respectively. All of the numerical
results reported in the tables demonstrate that the best solutions in terms of actual damage identification and the total
number of FEAs required are obtained by means of the MGA employing the ECBI. It is worth mentioning that most of the
different optimization runs made by this method resulted in obtaining the actual damage induced.
1804 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809

Table 2d
The damage ratios of the 20-element beam for case 1 using the MGA with the ECBI.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1 0.2 0.15 0.2 3327.0

2 0.2 0.15 0.2 1254.0
3 0.2 0.15 0.2 1056.0
4 0.2 0.15 0.2 2803.0
5 0.2 0.15 0.2 11 550
6 0.2 0.15 0.2 6094.0
7 0.2 0.15 0.2 8288.0
8 0.2 0.15 0.2 3858.0
9 0.2 0.15 0.2 8107.0
10 0.2 0.15 0.2 4375.0
Average 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2 0.15 0.2 0 0 0 5071.2
Actual damage 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2 0.15 0.2 0 0 0 –

Table 3a
The damage ratios of the 20-element beam for case 2 using the SGA with the MDLAC.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1 0.10 0.25 0.5 0.05 0.50 0.10 20 000

2 0.15 0.25 0.5 0.05 0.50 0.00 20 000
3 0.15 0.25 0.5 0.05 0.50 0.10 20 000
4 0.10 0.25 0.5 0.05 0.50 0.10 20 000
5 0.1 0.15 0.15 0.4 0.05 0.45 0.05 0.00 20 000
6 0.10 0.25 0.5 0.05 0.50 0.15 20 000
7 0.1 0.00 0.25 0.5 0.05 0.50 0.10 20 000
8 0.05 0.10 0.25 0.5 0.05 0.45 0.05 20 000
9 0.10 0.25 0.5 0.05 0.50 0.10 20 000
10 0.10 0.25 0.5 0.05 0.50 0.10 0.00 20 000
Average 0 0.01 0 0 0.01 0.005 0 0 0.105 0 0 0.24 0.49 0.05 0.49 0 0 0.075 0 0.07 20 000
Actual 0 0 0 0 0 0 0 0 0 0 0 0.2 0.3 0 0.3 0 0 0 0 0 –
damage

Table 3b
The damage ratios of the 20-element beam for case 2 using the SGA with the ECBI.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

1 0.20 0.35 0.35 0.15 20 000

2 0.10 0.15 0.30 0.30 20 000
3 0.05 0.15 0.35 0.35 20 000
4 0.20 0.30 0.30 0.10 20 000
5 0.05 0.15 0.30 0.30 20 000
6 0.05 0.15 0.30 0.30 20 000
7 0.05 0.20 0.30 0.25 0.05 20 000
8 0.20 0.35 0.30 20 000
9 0.20 0.30 0.30 18 580
10 0.05 0.15 0.30 0.30 0.1 0.03 20 000
Average 0 0 0 0 0 0.005 0 0 0.03 0 0 0.175 0.315 0 0.305 0 0 0 0.01 0 19 858
Actual 0 0 0 0 0 0 0 0 0 0 0 0.2 0.3 0 0.3 0 0 0 0 0 –
damage

The convergence history of the various methods for cases 1 and 2 are depicted in Figs. 4a and 4b, respectively, where the
objective function (MDLAC or ECBI) versus finite element analyses is shown. In these figures, each curve is related to the
result of a typical optimization run for each method. It should be mentioned that, for cases 1 and 2, the sites and extent of all
damage induced are accurately detected using methods 2 and 4. The convergence history of the various methods achieved
by a single optimization run demonstrates that the combination of the MGA and the ECBI can also provide a robust tool for
identifying the actual damage.
In order to investigate the noise effects on the performance of the proposed method (MGA + ECBI), measurement noise is
considered here by a standard error of ±0.15% for the natural frequencies [4,9]. The damage identification results for damage
cases 1 and 2 considering noise and with ng = 450 are given in Tables 4a–4b, respectively. It can be observed that, when
noise is taken into account in the damage detection problem, the MGA–ECBI method can detect the damage sites and extent
correctly for most of the independent runs. In these cases, the maximum number of generations controls the convergence
of the algorithm.
 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809 1805

Table 3c
The damage ratios of the 20-element beam for case 2 using the MGA with the MDLAC.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 0.25 0.4 0.4 0.1 20 025
2 0.25 0.4 0.4 0.1 20 019
3 0.15 0.2 0.2 0.0 20 013
4 0.15 0.2 0.2 0.0 20 024
5 0.25 0.4 0.4 0.1 20 022
6 0.25 0.4 0.4 0.1 20 016
7 0.25 0.4 0.4 0.1 20 021
8 0.25 0.4 0.4 0.1 20 020
9 0.25 0.4 0.4 0.1 20 024
10 0.25 0.4 0.4 0.1 20 016
Average 0 0 0 0 0 0 0 0 0 0 0 0.23 0.36 0 0.36 0 0 0 0 0.08 20 020
Actual damage 0 0 0 0 0 0 0 0 0 0 0 0.2 0.3 0 0.3 0 0 0 0 0 –

Table 3d
The damage ratios of the 20-element beam for case 2 using the MGA with the ECBI.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 0.2 0.30 0.3 7503.0
2 0.2 0.30 0.3 8022.0
3 0.2 0.30 0.3 12 903
4 0.2 0.30 0.3 4002.0
5 0.2 0.30 0.3 10 386
6 0.2 0.30 0.3 2262.0
7 0.2 0.30 0.3 1756.0
8 0.2 0.35 0.4 20 024
9 0.2 0.35 0.4 20 018
10 0.2 0.25 0.3 20 020
Average 0 0 0 0 0 0 0 0 0 0 0 0.2 0.305 0 0.32 0 0 0 0 0 10 690
Actual damage 0 0 0 0 0 0 0 0 0 0 0 0.2 0.3 0 0.3 0 0 0 0 0 –

Fig. 4a. The convergence history of the 20-element beam for case 1 obtained by the four methods.

4.2. Thirty one-bar planar truss

The 31-bar planar truss shown in Fig. 5 selected from [9] is modeled using the conventional finite element method
without internal nodes, leading to 25 degrees of freedom. In this example, the first ten natural frequencies are utilized
for damage detection. The material density and elasticity modulus are 2770 kg/m3 and 70 GPa, respectively. Damage in the
1806 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809

Fig. 4b. The convergence history of the 20-element beam for case 2 obtained by the four methods.

Table 4a
The damage ratios of the 20-element beam for case 1 using the MGA with the ECBI considering noise.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 0.25 0.15 0.2 9028
2 0.25 0.10 0.2 9016
3 0.2 0.15 0.2 9020
4 0.2 0.15 0.2 9024
5 0.2 0.15 0.2 9029
6 0.2 0.15 0.2 9025
7 0.2 0.15 0.2 9026
8 0.2 0.05 0.2 9020
9 0.2 0.15 0.2 9023
10 0.2 0.15 0.2 9015
Average 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.21 0.135 0.2 0 0 0 9022.6
Actual damage 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2 0.15 0.2 0 0 0 –

Table 4b
The damage ratios of the 20-element beam for case 2 using the MGA with the ECBI considering noise.
Sample no. Element numbers FEA
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 0.2 0.3 0.4 9022
2 0.05 0.2 0.25 0.3 9019
3 0.2 0.3 0.3 9029
4 0.2 0.3 0.3 9023
5 0.1 0.05 0.1 0.3 0.3 9021
6 0.2 0.3 0.3 9019
7 0.2 0.3 0.4 9025
8 0.1 0.2 0.2 0.3 9013
9 0.2 0.4 0.4 9011
10 0.2 0.4 0.4 9021
Average 0 0 0 0 0 0 0 0.015 0.01 0.005 0 0.19 0.305 0 0.34 0 0 0 0 0 9020.3
Actual 0 0 0 0 0 0 0 0 0 0 0 0.2 0.3 0 0.3 0 0 0 0 0 –
damage

structure is also simulated as a relative reduction in the elasticity modulus of individual bars. Three different damage cases,
given in Table 5, are induced in the structure, and the MGA employing the ECBI is tested for each case. All optimization
parameters are the same as the first example.
The prediction of damage levels for cases 1–3 without considering noise are given in Tables 6a–6c, respectively. As can
be seen in the tables, the damage identification method proposed here (MGA + ECBI) can accurately detect the damage
 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809 1807

Table 5
Three different damage cases induced in the 31-bar planar truss.
Case 1 Case 2 Case 3
Element number Damage ratio Element number Damage ratio Element number Damage ratio
11 0.25 16 0.30 1 0.30
25 0.15 – – 2 0.20

Fig. 5. A planar truss having 31 elements.

Table 6a
The damage ratios of the 31-bar truss for case 1 using the MGA with the ECBI.
Sample no. Element numbers FEA
1 2 ... 10 11 12 13 14 15 16 17 ... 24 25 26 27 28 29 30 31
1 0.25 0.15 16 501
2 0.25 0.15 6045.0
3 0.25 0.15 7419.0
4 0.25 0.15 2181.0
5 0.25 0.15 6130.0
6 0.40 0.25 20 025
7 0.25 0.15 18 403
8 0.25 0.15 3398.0
9 0.25 0.25 7701.0
10 0.25 0.25 1298.0
Average 0 0 0 0 0.265 0 0 0 0 0 0 0 0 0.18 0 0 0 0 0 0 8910.1
Actual 0 0 0 0 0.25 0 0 0 0 0 0 0 0 0.15 0 0 0 0 0 0 –
damage

Table 6b
The damage ratios of the 31-bar truss for case 2 using the MGA with the ECBI.
Sample no. Element numbers FEA
1 2 ... 10 11 12 13 14 15 16 17 ... 24 25 26 27 28 29 30 31
1 0.30 1365.0
2 0.30 4739.0
3 0.30 1907.0
4 0.30 2588.0
5 0.30 4328.0
6 0.30 2090.0
7 0.30 4673.0
8 0.30 4451.0
9 0.30 1290.0
10 0.30 3646.0
Average 0 0 0 0 0 0 0 0 0 0.30 0 0 0 0 0 0 0 0 0 0 3107.7
Actual 0 0 0 0 0 0 0 0 0 0.30 0 0 0 0 0 0 0 0 0 0 –
damage

sites and extent for most of the simulations. It can also be observed that only for one optimization run of case 1 are the
damage extent are not accurately determined. In order to assess the effects of measurement noise on the efficiency of
the method, the damage identification results of cases 1–3 when noise is considered for natural frequencies are given
in Tables 7a–7c, respectively. The numerical results demonstrate the efficiency of the method for locating the damaged
elements.
1808 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809

Table 6c
The damage ratios of the 31-bar truss for case 3 using the MGA with the ECBI.
Sample no. Element numbers FEA
1 2 ... 10 11 12 13 14 15 16 17 ... 24 25 26 27 28 29 30 31

1 0.3 0.2 1 557

2 0.3 0.2 19 974
3 0.3 0.2 9 958
4 0.3 0.2 1 493
5 0.3 0.2 9 507
6 0.3 0.2 3 646
7 0.3 0.2 2 728
8 0.3 0.2 3 758
9 0.3 0.2 8 931
10 0.3 0.2 2 359
Average 0.3 0.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6391.1
Actual 0.3 0.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 –
damage

Table 7a
The damage ratios of the 31-bar truss for case 1 using the MGA with the ECBI considering noise.
Sample no. Element numbers FEA
1 2 ... 10 11 12 13 14 ... 21 22 23 24 25 26 27 28 29 30 31

1 0.20 0.05 0.10 13 028

2 0.40 0.10 0.20 13 020
3 0.20 0.05 0.10 13 029
4 0.40 0.20 0.10 13 043
5 0.40 0.20 0.05 13 021
6 0.40 0.10 0.20 0.05 13 026
7 0.30 0.05 0.15 13 023
8 0.20 0.10 0.10 13 024
9 0.10 0.05 0.10 13 026
10 0.25 0.10 0.05 13 025
Average 0 0 0 0 0.285 0 0 0 0 0.06 0 0 0 0.135 0 0.005 0 0.005 0.015 0 13 027
Actual 0 0 0 0 0.25 0 0 0 0 0 0 0 0 0.15 0 0 0 0 0 0 –
damage

Table 7b
The damage ratios of the 31-bar truss for case 2 using the MGA with the ECBI considering noise.
Sample no. Element numbers FEA
1 2 ... 10 11 12 13 14 15 16 17 ... 20 ... 26 27 28 29 30 31

1 0.25 13 014
2 0.30 13 031
3 0.40 0.05 13 027
4 0.30 13 029
5 0.30 13 019
6 0.30 13 027
7 0.30 13 025
8 0.30 13 017
9 0.40 0.05 13 020
10 0.25 13 028
Average 0 0 0 0 0 0 0 0 0 0.31 0 0 0.005 0 0 0.005 0 0 0 0 13 024
Actual 0 0 0 0 0 0 0 0 0 0.30 0 0 0 0 0 0 0 0 0 0 –
damage

5. Conclusions

An optimization-based damage identification method is proposed here to identify multiple damage in structural systems.
For the optimization, a modified genetic algorithm (MGA) including two new operators is proposed. The new operators are
called here the health operator and the simulator operator. An efficient correlation-based index (ECBI) is also introduced as
the objective function for the optimization algorithm. Two illustrative test examples considering and without considering
measurement noise are selected to assess the advantages of the proposed method. Numerical results for various damage
cases with different sample runs demonstrate that the combination of the MGA and the ECBI can create a robust tool for
structural damage detection in terms of finding the actual damage sites and extent, and reducing the total number of finite
element analyses when comparing with the available tools such as SGA–MDLAC.
 M. Nobahari, S.M. Seyedpoor / Mathematical and Computer Modelling 53 (2011) 1798–1809 1809

Table 7c
The damage ratios of the 31-bar truss for case 3 using the MGA with the ECBI considering noise.
Sample no. Element numbers FEA
1 2 ... 10 11 12 13 14 15 16 17 ... 24 25 26 27 28 29 30 31

1 0.25 0.25 13 022

2 0.25 0.2 13 024
3 0.30 0.25 13 018
4 0.30 0.25 13 027
5 0.30 0.25 13 025
6 0.40 0.10 13 019
7 0.40 0.20 13 020
8 0.40 0.10 0.05 13 024
9 0.40 0.10 13 021
10 0.30 0.25 13 018
Average 0.33 0.195 0 0 0 0 0 0 0 0 0 0 0 0 0.005 0 0 0 0 0 13 021.8
Actual 0.3 0.2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 –
damage

Acknowledgements

This research work was supported by Islamic Azad University, Neyshabur Branch. The authors are also grateful to the
reviewers for their constructive suggestions.

References

[1] B.H. Koh, S.J. Dyke, Structural health monitoring for flexible bridge structures using correlation and sensitivity of modal data, Computers & Structures
85 (2007) 117–130.
[2] R.S. He, S.F. Hwang, Damage detection by a hybrid real-parameter genetic algorithm under the assistance of grey relation analysis, Engineering
Applications of Artificial Intelligence 20 (2007) 980–992.
[3] M.T. Vakil-Baghmisheh, M. Peimani, M. Homayoun Sadeghi, M.M. Ettefagh, Crack detection in beam-like structures using genetic algorithms, Applied
Soft Computing 8 (2008) 1150–1160.
[4] H.Y. Guo, Z.L. Li, A two-stage method to identify structural damage sites and extents by using evidence theory and micro-search genetic algorithm,
Mechanical Systems and Signal Processing 23 (2009) 769–782.
[5] S. Fallahian, S.M. Seyedpoor, A two stage method for structural damage identification using an adaptive neuro-fuzzy inference system and particle
swarm optimization, Asian Journal of Civil Engineering (Building and Housing) 11 (2010) 797–810.
[6] S.M. Seyedpoor, Structural damage detection using a multi-stage particle swarm optimization, Advances in Structural Engineering (2011) (in press).
[7] O. Begambre, J.E. Laier, A hybrid particle swarm optimization–simplex algorithm (PSOS) for structural damage identification, Advances in Engineering
Software 40 (2009) 883–891.
[8] X. Wang, N. Hu, H. Fukunaga, Z.H. Yao, Structural damage identification using static test data and changes in frequencies, Engineering Structures 23
(2001) 610–621.
[9] A. Messina, E.J. Williams, T. Contursi, Structural damage detection by a sensitivity and statistical-based method, Journal of Sound and Vibration 216
(1998) 791–808.
[10] D.E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning, Addison-Wesley, Massachusetts, 1989.
[11] S.M. Seyedpoor, S. Gholizadeh, Optimum shape design of arch dams by a combination of simultaneous perturbation stochastic approximation and
genetic algorithm methods, Advances in Structural Engineering 11 (2008) 501–510.
[12] R. Balamurugan, C.V. Ramakrishnan, N. Singh, Performance evaluation of a two stage adaptive genetic algorithm (TSAGA) in structural topology
optimization, Applied Soft Computing 8 (2008) 1607–1624.
[13] E.S. Kameshki, M.P. Saka, Genetic algorithm based optimum design of nonlinear planar steel frames with various semi-rigid connections, Journal of
Constructional Steel Research 59 (2003) 109–134.
[14] H.M. Gomes, N.R.S. Silva, Some comparisons for damage detection on structures using genetic algorithms and modal sensitivity method, Applied
Mathematical Modelling 32 (2008) 2216–2232.