Applied Mathematical Modelling 35 (2011) 5872–5883

Contents lists available at ScienceDirect

Applied Mathematical Modelling

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

A model updating approach based on design points for unknown

structural parameters
Hasan Basri Basßağa ⇑, Temel Türker, Alemdar Bayraktar
Karadeniz Technical University, Department of Civil Engineering, 61080 Trabzon, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: Finite element analysis has become an essential tool to estimate structural responses under
Received 7 December 2010 static and dynamic loads. However, there are a lot of uncertainties in structural properties.
Received in revised form 8 May 2011 For this reason, in many cases, the outcomes of the theoretical and experimental modal
Accepted 16 May 2011
analyses do not match. Therefore, the analytical models of the structures need to be
Available online 27 May 2011
updated according to the experimental test results. The commonly used method to get
parameters for model updating is experimental modal analysis which provides structural
Keywords:
dynamic characteristic (natural frequencies, mode shapes and modal damping ratio). There
Design points
Modal parameters
are many methods available for the updating process. This study addresses an updating
Model updating algorithm to modify the numerical models by using the design points for unknown struc-
Unknown structural properties tural properties. The proposed method aims to minimize the difference between the ana-
lytical and experimental natural frequencies by updating uncertain parameters for each
mode and combine them to get an optimum solution. The algorithm is tested on a column
and a 2D frame models. These models are investigated by taking the connection rigidity
and elasticity modulus as unknown parameters. It is observed that the proposed algorithm
gives better results for unknown structural properties compared to the initial values.
Ó 2011 Elsevier Inc. All rights reserved.

1. Introduction

The most widespread approach for analytical modeling in engineering design is the finite element (FE) method. Modern
computers which are capable of processing large matrix problems at high speed have enabled the construction of large and
sophisticated analytical models. In the civil engineering area, the capacity of finite element models to predict vibration
modes of structures is limited by inadequate modeling of joints, variations in the material properties, and other model errors
which might be improved by updating. FE model updating is a process of making sure that finite element analysis results
from models better reflect the measured data than the initial models [1].
The updating procedure of a FE model uses measurement data as exact reference data in order to update the selected
uncertain parameters by minimizing the differences between experimentally and analytically obtained resonance
frequencies and mode shapes. This process is conducted in four steps. In the first step, the domain which is time, frequency
or time-frequency is chosen. The second step is to determine the part of the models that are thought to have been modeled
incorrectly. The third task is to formulate the objective function as the difference between the experimental and analytical
natural frequencies. Lastly, an optimization method is implemented to identify parameters that minimize this function.
Several methods of structural model updating have been proposed and the topic is still under active study in various
areas. Most of these studies centered on approaches such as the optimal matrix updating [2], sensitivity-based parameter

⇑ Corresponding author. Tel.: +90 462 377 40 19; fax: +90 462 377 26 06.
E-mail addresses: hasanbb69@hotmail.com (H.B. Basßağa), temelturker@ktu.edu.tr (T. Türker), alemdar@ktu.edu.tr (A. Bayraktar).

0307-904X/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2011.05.041
 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883 5873

estimation [3,4], eigen-structure assignment algorithms [5] and neural-networks updating methods [6]. Zhang et al. [7] pro-
posed an improved sensitivity-based parameter updating method. Marwala [8] used the response surface method for finite
element model updating. The updated parameters of the finite element model were calculated using genetic algorithms by
optimizing the surface response equation. The proposed method was tested on an unsymmetrical H-shaped structure. It was
observed that the proposed method gave the updated natural frequencies and mode shapes that were of the same order of
accuracy as those given by simulated annealing and genetic algorithm. Steenackers and Guillaume [9] presented an auto-
mated model updating procedure using one updating parameter. The proposed procedure was then extended to multiple
updating parameters by taking into account the measurement uncertainty. A FE model for the test plate was created and
used as a function of one or more modeling parameters. The experimental results on the simple test case were discussed,
the automated model updating technique was described in detail and updated parameters from the FE model were com-
pared to the original ones. It was presented the theory of sensitivity-based model updating with a special focus on the prop-
erties of the solution that result from the combination of optimization of the response prediction with a priori information
about the uncertain parameters [10]. Besides this study, the authors investigated the use of second order information in
model updating [11]. It was expressed that there exists a direct link between regularisation approaches used in model updat-
ing and the second order derivatives originating from the extended part of the Taylor series approximation of the model data
quadratic distance measure. The proposed approach based on the use of the Newton–Raphson model updating scheme was
effectively applied in two case studies.
This study denotes updating a finite element model by using data acquired from a physical vibration test. For this aim, an
updating algorithm to the numerical models is proposed by using the design points for unknown structural properties. The
proposed algorithm has an iterative solution to make the difference between analytical and experimental natural frequen-
cies a minimum. A column and a 2D steel models are selected as numerical examples to illustrate the efficiency of the
algorithm.

2. Parameter selection for updating

A finite element model which will be updated requires, in its preparation, the consideration of factors not normally taken
into account in regular model construction. Of these, the choice of updating parameters is the most important. The uncer-
tainties in analytical modeling are classified into two main groups: physical uncertainty and numerical uncertainty. The
physical uncertainties are: (i) boundary and initial conditions, (ii) material properties, (iii) geometry and (iv) loads. The
uncertainty is further increased because many of these properties may vary substantially with temperature, frequency or
load level. The numerical uncertainties are; (i) conceptual modeling, (ii) mathematical modeling, (iii) discretization errors,
(iv) numerical solution, and (v) human mistakes [12].
It should be attempted to assess the confidence which can be attributed to various features of the model. For example, the
main span of a beam, away from the boundaries, might be considered to be modeled with a high level of confidence. Joints
and constraints could be considered to be less accurately modeled, and therefore they are in greater need of updating. The
parameterization of the inaccurate parts of the model is important. The numerical predictions (e.g. natural frequencies and
mode shapes) should be sensitive to small changes in the parameters. Experimental results show that natural frequencies are
often significantly affected by small differences in the construction of joints in nominally identical test pieces [13]. However,
it can be very difficult to find joint parameters to which the analytical predictions are sensitive. If the analytical response is
insensitive to changes in one or several updating parameters, then updating will result in unrealistic values for rest of updat-
ing parameters. The result, in this case, will be an updated model which replicates the measurements but lacks physical
meaning.

3. Model updating

The main goal in the model update is to make the tolerance which is the distance between the frequencies obtained
experimentally and theoretically equal to zero. But, it is a difficult process because of the uncertainties in the structural
parameters such as the elasticity modulus, mass density, boundary conditions, etc. Therefore, it is aimed to minimize the
tolerance. If the tolerance in each mode is shown as toli, it can then be defined as:

toli ¼ minðfie  fit Þ; ð1Þ

e t
where i indicates the each mode, f and f are the frequencies obtained from experimental measurements and theoretical
analysis, respectively. In Eq. (1), the fe is accepted as the exact results, so ft is updated to get the minimum difference. If
the theoretical analysis results are considered as the function of the structural parameters, ft is written as;

f t ¼ gðE; I; A; . . .Þ; ð2Þ

where g() is a function which produces the theoretical frequencies, E is the elasticity modulus, I is the inertia moment, A is
the cross section and ‘. . .’ indicates the other parameters. In many practical problems, this function may not be known
5874 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883

explicitly such as through one or more equations to represent the function. Rather it may be known only implicitly through a
procedure such as a finite element analysis. If Eq. (2) is substituted into Eq. (1), the tolerance is obtained implicitly as:
toli ¼ minðfie  g i ðE; I; A; . . .ÞÞ: ð3Þ
If Eq. (3) is rewritten by making the tolerance equal to zero, the expression becomes:
0 ¼ fie  g i ðE; I; A; . . .Þ: ð4Þ
If the right side of Eq. (4) is defined as a new function, the expression is now written as:
0 ¼ g~i ðE; I; A; . . .Þ: ð5Þ
The model update is concerned with the solution of Eq. (5) for the considered random parameters. In this study, Eq. (5) is
solved by updating the selected parameters using design points which are calculated in the reliability analysis.

4. Developed model updating method

4.1. Calculation of design point

The reliability analysis of structures deals with the calculation of the failure probability under a limit condition. This con-
dition is defined by a function named the limit state function or performance function. Design point which is on the limit
state function is calculated by an iterative solution to get the shortest distance from the origin of reduced variables to the
function. The design point is defined in the normal standard space as:
z ¼ minfkzkjgðzÞ ¼ 0g: ð6Þ
In Fig. 1, the design point, the shortest distance (b), sensitivity factors (a) and the limit state function are demonstrated in the
normalized space.
The design point and b can be calculated with different reliability methods [14–16]. In this study, the algorithm which is
developed by Zhang and Der Kiureghian [17] is used to calculate the design point. For this aim, the reliability analysis pro-
gram, FERUM [17–19], is updated for implicit limit state functions using ANSYS [20] finite element program.
The details about the calculation stages of the design point are given below:

1. Define a distribution with mean values and standard deviations for each random parameter.
2. Select an initial point (generally taken mean points) and transform it into a point in the standard normal space:

zi ¼ U1 ðF X ðxi ÞÞ; ð7Þ

where U() is the standard normal cumulative distribution function (CDF) and F() is the marginal CDF of the random
variable.
3. Transform vector zi to xi in the original space (skip this step in the first iteration).
4. Compute the value of the limit state function using ANSYS finite element program and store it as g0 at the first iteration.
5. Compute the gradient vector of the limit state function:
 
@g  @g 
rZg ¼  ¼  ½J1 ; ð8Þ
@Z z @X x

Z1

Tangent α
Design Point

z1*

β
Limit State Function

Z2
z *2

Fig. 1. Demonstration of the design point.

 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883 5875

where J denotes the Jacobian matrix and calculated by:

2 @Z @Z 1
3
@X
1
 @X n
6 1 7
6 .. 7
½J ¼ 6 ... ..
. . 7 : ð9Þ
4 5
@Z n @Z n
@X 1
 @X n

6. Compute the sensitivity factors:

rZg
a¼ : ð10Þ
krZgk
7. Check convergence according to two criteria:
)
j gg0i j < e1
: ð11Þ
Z  aT Z a < e2
8. Take a new step if convergence is not achieved:
Z kþ1 ¼ Z k þ kk dk ; ð12Þ
where dk and kk are step direction vector and step size, respectively, and defined as:
 
g
dk ¼ aT Z k þ a  Zk ; ð13Þ
krZg k k
9
kk ¼ arg minfmðZ k þ kdk Þ  mðZ k Þg >
>
=
mðZÞ ¼ 12 kZk2 þ cjgðZÞj ; ð14Þ
>
>
c> kZk ;
krZgk

where m is merit function.

9. Repeat 3–8 steps until convergence is achieved.

4.2. The algorithm of the developed method

The main stages of the developed algorithm are ‘‘applying the calculation of the design point algorithm for each mode’’,
‘‘determining the modal participation factors (MPFs)’’ and ‘‘the calculation of the uncertain parameters by using design point
and MPFs’’. The steps of the algorithm are given below and shown in the flow chart in Fig. 2.

1. Determine the functions implicitly for each mode (i = 1, 2, 3, . . .) according to Eq. (4).
2. Apply the design point algorithm for the functions obtained in step 1 separately.
3. Get the design points for the selected parameters in each mode.
4. Determine the MPFs for each mode.
5. Multiply each selected variable by the MPFs and then sum up them.

5. Application

The proposed algorithm is tested on the column model and 2D frame model. For this purpose, the steps given below are
applied to both models:

1. Analytical modal analysis is performed using the ANSYS finite element program.
2. The experimental modal analysis is carried out.
3. The differences in natural frequencies obtained experimentally and analytically are compared with each other.
4. Model updating is implemented by the proposed algorithm to get the uncertain parameters.
5. The differences between the measured, initial and updated frequencies are compared with each other.

5.1. Example 1: column model

5.1.1. Analytical modal analysis

The column model is created using a rectangular shape steel beam. A large number of finite elements are used in the mod-
el to represent the real behavior of the column. The material and geometrical properties of the column model are given in
Table 1.
The model is named as the initial finite element model because there are not any updated parameters. The initial finite
element model of the column is shown in Fig. 3.
5876 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883

Fig. 2. The flow chart of developed model updating method.

In the analysis, the first five natural frequencies are obtained from the initial analytical model with in the 10.281–
583.772 Hz frequency range.

5.1.2. Experimental measurement

The column model is fixed to a rigid surface by welded connection. The measurement of the column model is achieved
using six single axis accelerometers. The bending modes of the column model are measured by placing the accelerometers in
 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883 5877

Table 1
The geometrical and material properties of the column model.

Geometrical properties Height (m) 0.9

Cross section area (m2) 3.0E4
Moment of inertia (m4) 2.50E9
Connection on the foot Fully rigid
Material properties Average material density (kg/m3) 7750
Elasticity modulus (N/m2) 2.06E11

Fig. 3. The initial finite element model of the column.

the transversal direction. The responses of the column model are acquired for ambient vibrations. A 17 channel data acqui-
sition system is used to collect signals coming from the accelerometers. The column model, the connection on the base and
measurement equipment are shown in Fig. 4.
The measurement on the column model is made in the 0–800 Hz frequency range. The responses of the column model are
collected approximately over a 10 min interval. The power spectral density of the response acceleration of the test column
model is given in Fig. 5.
The peaks correspond to the resonant frequencies of the column model. The frequencies of each peak are the natural fre-
quencies. The first six natural frequencies of the column model are measured in the frequency range. The modal damping
ratios are calculated using the method proposed by Brincker et al. [21]. The mode shapes of the column model are extracted

Fig. 4. Some views from the column model.

5878 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883

Fig. 5. The power spectral density of the response acceleration of the test column model.

from real part of the power spectral functions. The experimental natural frequencies of the column model are attained with
in the 10.01–565.90 Hz frequency range.

5.1.3. Comparison of experimental and analytical natural frequencies

It is observed from the experimental measurement and analytical modal analysis of the column model that the natural
frequencies are close to each other. There are some differences between the natural frequencies obtained from the experi-
mental measurement and initial analytical model. These arise from the assumptions made on the initial analytical models. It
is a known fact that the connection at the base is not fully rigid [13]. In addition, the elasticity value is not constant for all
steel. A comparison of the natural frequencies of the models from analytical and experimental modal analyses is given in
Table 2. The results show that on average there is 2.8% error between the frequencies.

5.1.4. Model updating of the initial model by the developed algorithm

The density, elasticity modulus and connection rigidity (CR) at the base of the column can be considered to be uncertain
parameters for the initial analytical model of the column. In this study, the elasticity modulus and the CR at the base are
selected as updating parameters, because the density of the used steel element is determined by a test. Because there is
no information about welded CR at the base of the column in the literature, a parametric study is carried out to determine
a value for the approximate rigidity. For this purpose, the elasticity modulus of the column model is selected to be constant
as given in Table 1 and the CR is taken with mean of 90,000 Nm/rad, standard deviation of 7000 Nm/rad and distribution of
lognormal. After the developed algorithm is applied, the CRs giving the nearest frequency value to the experimental natural
frequencies are determined for the first five modes and given in Table 3.
In the light of the obtained result, the model update is carried out by taking the statistical distributions of the selected
parameters given in Table 4 and the results are demonstrated in Table 5.

5.1.5. Comparison of the results

The comparison of the natural frequencies obtained from initial analytical modal analysis and experimental measure-
ments are given in Table 2 with relative errors before the model update. After the calculation of the uncertain parameters,
the natural frequencies are obtained for these values again and compared with the previous values. The results are shown in
Table 6. The results show that the updated frequencies are approximately the same as the experimental frequencies. The
average error has decreased from 2.8% to 0.259%.

Table 2
The comparison of analytical and experimental natural frequencies of the column model.

Mode Frequency (Hz) Relative error (%)

Measured Initial
1 10.01 10.281 2.7
2 62.90 64.425 2.4
3 175.80 180.352 2.6
4 343.50 353.298 2.9
5 565.90 583.772 3.2
 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883 5879

Table 3
The CRs for each mode of the column model.

Mode CR (Nm/rad) MPFs (%) CR ⁄ MPF (Nm/rad)

1 9.328E4 66.810 6.23203680E4
2 11.069E4 20.474 2.26626706E4
3 9.319E4 7.004 0.65270276E4
4 8.138 E4 3.556 0.28938728E4
5 6.630 E4 2.155 0.14287650E4
The mean value 9.5832704E4

Table 4
The statistical distributions of random parameters for the column model.

Mean Standard deviation Distribution

Elasticity modulus (N/m2) 2.06E11 1.03E10 LogNormal
CR (Nm/rad) 95,000 4750 LogNormal

Table 5
The updated values for elasticity modulus and CR of the column model.

Mode Elasticity modulus (N/m2) CR (Nm/rad) MPFs (%) Uncertain parameters ⁄ MPF
Elasticity modulus (N/m2) CR (Nm/rad)
1 2.04256E11 94867.34 66.810 1.36464E11 63380.870
2 2.05279E11 94876.89 20.474 0.42028E11 19425.100
3 2.05917E11 94883.26 7.004 0.14422E11 6645.624
4 2.06683E11 94891.46 3.556 0.07349E11 3374.340
5 2.05914E11 94883.35 2.155 0.04437E11 2044.736
The mean values 2.04702E+11 94870.66

Table 6
The comparison of the natural frequencies of the column model before and after the update.

Mode Frequency (Hz) Relative error (%)

Experimental Analytical
Initial Updated Before update After update
1 10.01 10.281 9.981 2.7 0.290
2 62.90 64.425 62.604 2.4 0.471
3 175.80 180.352 175.291 2.6 0.290
4 343.50 353.298 343.027 2.9 0.138
5 565.90 583.772 566.509 3.2 0.108

5.2. Example 2: 2D frame model

5.2.1. Analytical modal analysis

The material properties of the 2D frame model are the same as the column model. So, all properties are given in Table 1.
The span of the frame is considered to be 90 cm. The initial finite element model of the 2D frame is shown in Fig. 6 and the
natural frequencies obtained analytically from the ANSYS finite element program are obtained between 9.371 and
132.046 Hz.

5.2.2. Experimental measurement

The 2D model is created by using steel beams having the same sectional properties as the column model. The bases of the
2D frame model are fixed to rigid surfaces by welded connections. The 2D frame model, the connection of the base and
beam-to-column joints are shown in Fig. 7. The measurement of the 2D frame model is undertaken using 10 single axis
accelerometers. The bending modes of the 2D frame model are measured by placing the accelerometers in the transversal
directions of the beam and columns as shown in Fig. 8.
The measurement of the 2D frame model is done in the 0–200 Hz frequency range. The responses of the 2D frame model
are collected over a period of approximately 10 min. The power spectral density of the 2D frame model is given in Fig. 9.
The experimental natural frequencies of the 2D frame model are obtained between 9.066 and 127.50 Hz.
5880 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883

Fig. 6. The initial finite element model of the 2D frame.

Fig. 7. Some views from the 2D frame model.

Fig. 8. The measurement equipments used in the measurement.

5.2.3. Comparison of experimental and analytical natural frequencies

Similar to the column model results, the natural frequencies obtained analytically and experimentally are close to each
other. The analytical and experimental results of 2D frame model are given in Table 7. It is seen from the table that there is an
error of 3.2% error between the results because of the uncertainties.
 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883 5881

Fig. 9. The power spectral density functions of the 2D frame model.

Table 7
The comparison of the analytical and experimental natural frequencies of the 2D frame model.

Mode Frequency (Hz) Relative error (%)

Measured Initial
1 9.066 9.371 3.4
2 35.89 36.977 3.0
3 58.31 60.320 3.4
4 63.80 65.394 2.5
5 127.50 132.046 3.6

5.2.4. Model updating of the initial model by the developed algorithm

For the 2D frame model, in addition to the uncertainties considered in the column model, the beam-to-column connec-
tion rigidities (BCCRs) are considered as uncertain parameters. A parametric study is carried out for the 2D frame model, sim-
ilar to the column model, to get the BCCR. For this purpose, the elasticity modulus given in Table 1 and CR obtained from the
parametric study of the column model are selected as constants and BCCR is considered a random parameter with mean of
90,000 Nm/rad, standard deviation of 7000 Nm/rad and distribution of lognormal. The BCCRs obtained from developed
method are given in Table 8 for each mode.
After the parametric study, the model updated of the 2D frame model is performed by the developed method. The sta-
tistical distributions of the three random parameters are given in Table 9. The results of the model updating of the 2D frame
model are shown in Table 10.

5.2.5. Comparison of the results

In Table 7, the natural frequencies obtained from initial analytical modal analysis and experimental measurements and
relative errors are given. In this section, the natural frequencies obtained from the updated parameters are compared with
the previous values. The results are shown in Table 11.
According to the results shown in Table 11, the average error which arises from the differences between updated and
experimental frequencies is 1.24%. This value is much smaller than the average error (3.2%) obtained from the comparison
of initial analytical and experimental frequencies. Especially, for the first mode, the relative error obtained is very small
(0.165%).

Table 8
The BCCRs for each mode of the 2D frame model.

Mode BCCR (Nm/rad) MPFs (%) BCCR ⁄ MPF (Nm/rad)

1 2.197E4 66.32 1.4570504E4
2 0.543E4 14.83 0.0805269E4
3 1.271E4 9.6 0.1220160E4
4 6.062E4 7.33 0.4443446E4
5 0.467E4 1.92 0.0089664E4
The mean value 2.1129043E4
5882 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883

Table 9
The statistical distributions of random parameters for the 2D frame model.

Mean Standard deviation Distribution

Elasticity modulus (N/m2) 2.06E11 1.03E10 LogNormal
CR (Nm/rad) 95000 4750 LogNormal
BCCR (Nm/rad) 21000 1050 LogNormal

Table 10
The updated values for elasticity modulus, CR and BCCR of the 2D frame model.

Mode Elasticity modulus CR BCCR MPFs (%) Uncertain parameters ⁄ MPF

(N/m2) (Nm/rad) (Nm/rad)
Elasticity modulus CR BCCR
(N/m2) (Nm/rad) (Nm/rad)
1 2.064E11 94887.327 20976.072 66.32 1.3689E11 62929.275 13911.331
2 2.001E11 94857.019 20967.471 14.83 0.2968E11 14067.296 3109.476
3 2.031E11 94855.898 20967.410 9.6 0.1950E11 9106.166 2012.871
4 2.159E11 94957.148 21042.344 7.33 0.1583E11 6960.359 1542.404
5 1.982E11 94854.927 20964.020 1.92 0.0381E11 1821.215 402.509
The mean values 2.0567E11 94884.310 20978.591

Table 11
The comparison of the natural frequencies of the 2D frame model before and after the update.

Mode Frequency (Hz) Relative error (%)

Experimental Analytical
Initial Updated Before update After update
1 9.066 9.371 9.051 3.4 0.165
2 35.89 36.977 36.388 3.0 1.388
3 58.31 60.320 58.669 3.4 0.616
4 63.80 65.394 62.399 2.5 2.196
5 127.50 132.046 129.854 3.6 1.846

6. Conclusion

This paper is concerned with an algorithm which is developed for model updating. For this purpose, design points are
used to calculate the random parameters. Two examples which are a column and a 2D frame models are selected to dem-
onstrate the efficiency of the proposed method. The models are investigated under five subtitles: analytical modal analysis,
experimental measurement, comparison of the experimental and initial analytical natural frequencies, application of the
developed model updating method and comparison of the results.
It is observed that there are differences in the natural frequencies obtained from experimental measurement and initial
analytical modal analysis of the model because of the uncertain structural parameters. So, the models are updated using the
proposed model updating algorithm. In the models, the CR and elasticity modulus are taken as uncertainties for the column
model and CR, BCCR and elasticity modulus are taken as uncertainties for the 2D frame model. Because of the lack of infor-
mation about CR and BCCR, a parametric study is performed to determine the approximate values of these parameters.
According to the results of the study, the mean values of each selected parameter are attained to reflect the real condition
of the models in terms of the elasticity value and connection rigidity. The average error in the natural frequencies is de-
creased from 2.8% to 0.259% for the column model and is decreased from 3.2% to 1.24% for the 2D frame model by using
the developed model updating method. In consequence, the proposed algorithm gives better solutions for model updating
compared to the initial values.

Acknowledgements

This research was supported by the TUBITAK and Karadeniz Technical University under Research Grant Nos. 106M038
and 2006.112.001.1, respectively.

References

[1] M.I. Friswell, J.E. Mottershead, Finite Element Model Updating in Structural Dynamics, Kluwer Academic Publishers Group, Norwell, MA, 1995.
[2] A. Berman, E.J. Nagy, Improvement of a large analytical model using test data, AIAA J. 21 (1983) 1168–1173.
 H.B. Basßağa et al. / Applied Mathematical Modelling 35 (2011) 5872–5883 5883

[3] K.K. Denoyer, L.D. Peterson, Method for structural update using dynamically measured static flexibility matrices, AIAA J. 35 (1997) 362–368.
[4] C.P. Fritzen, D. Jennewein, T. Kiefer, Damage detection based on model updating methods, Mech. Syst. Signal Process. 12 (1998) 163–186.
[5] D.C. Zimmerman, M. Widengren, Correcting finite element modes using a symmetric eigenstructure assignment technique, AIAA J. 28 (1990) 1670–
1676.
[6] M.J. Atalla, D.J. Inman, On model updating using neural networks, Mech. Syst. Signal Process. 12 (1998) 135–161.
[7] Q.W. Zhang, C.C. Chang, T.Y.P. Chang, Finite element model updating for structures with parametric constraints, Earthquake Eng. Struct. Dyn. 29 (2000)
927–944.
[8] T. Marwala, Finite element model updating using response surface method, in: Proceedings of the 45th AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics and Materials Conference, Palm Springs, California, USA, 2004.
[9] G. Steenackers, P. Guillaume, Finite element model updating taking into account the uncertainty on the modal parameters estimates, J. Sound Vib. 296
(2006) 919–934.
[10] B. Titurus, M.I. Friswell, Regularisation in model updating, Int. J. Numer. Methods Eng. 75 (2008) 440–478.
[11] B. Titurus, M.I. Friswell, Second order approximations in model updating, in: ISMA 2010, Leuven, Berlin, 2010, pp. 2701–2712.
[12] E. Dascotte, The use of FE model updating and probabilistic analysis for dealing with uncertainty in structural dynamic simulation, in: Presented at the
2003 Japan Modal Analysis Conference (JMAC), Tokyo, Japan, 2003.
[13] T. Türker, M.E. Kartal, A. Bayraktar, M. Muvafık, Assessment of semi-rigid connections in steel structures by modal testing, J. Constr. Steel Res. 65
(2009) 1538–1547.
[14] A.M. Hasofer, N.C. Lind, Exact and invariant second moment code format, J. Eng. Mech. Div. ASCE 100 (1974) 111–121.
[15] R. Rackwitz, B. Fiessler, Structural reliability under combined random loadsequences, Comput. Struct. 9 (1978) 489–494.
[16] K. Breitung, Asymptotic approximation for multi-normal integrals, J. Eng. Mech. ASCE 110 (3) (1984) 357–366.
[17] Y. Zhang, A. Der Kiureghian, Finite Element Reliability Methods for Inelastic Structures, Report No. UCB/SEMM-97/05, Department of Civil and
Environmental Engineering, University of California, Berkeley, 1997.
[18] T. Haukaas, A New Computational Framework for Nonlinear Finite Element Reliability and Sensitivity Analysis, Report No. CE299, Department of Civil
and Environmental Engineering, University of California, Berkeley, 2001.
[19] <http://www.ce.berkeley.edu/FERUM/User_s_Guide/user_s_guide.html> Ferum User’s Guide, 01.01.2009.
[20] ANSYS, Swanson Analysis Systems Inc., Houston PA, USA, 2007.
[21] R. Brincker, C.E. Ventura, P. Andersen, Damping estimation by frequency domain decomposition, in: Proceedings of 19th International Modal Analysis
Conference (IMAC), Orlando, FL, USA, 2001.