Arch Computat Methods Eng (2016) 23:515–533

DOI 10.1007/s11831-015-9150-3

ORIGINAL PAPER

Structural Dynamic Model Updating Techniques: A State

of the Art Review
Shankar Sehgal1 • Harmesh Kumar1

Received: 21 March 2015 / Accepted: 27 March 2015 / Published online: 4 April 2015
 CIMNE, Barcelona, Spain 2015

Abstract This paper presents a review of structural collapsed just due to their own weight if tested under the
dynamic model updating techniques. Starting with a tu- effect of gravity. Thin and light weight products have lot
torial introduction of basic concepts of model updating, more tendencies to vibrate than their thick and heavy
the paper reviews direct and iterative techniques of model weight counterparts. Excessive vibrations can even result
updating along with their applications to real life systems. in pre-mature failure of products, for example, whether it
The main objective of this paper is to review the most is the suspension of an automobile, wing of an aircraft,
widely applied model updating techniques so that begin- the printed-circuit-board installed in a spaceship, blades
ners as well as practising engineers can appreciate, choose of an air-cooler, or the compact-disc of a computer etc.
and then utilize the most suitable model updating tech- On the other hand, consumers of today’s world desire for
nique for their customized application. Another objective non-vibrating and silent functioning of such products.
is to highlight the current issues, applications and obser- Thus it becomes very important for engineers to under-
vations for further advancements in the field of model stand the vibration behavior of structures through their
updating. dynamic analysis. Dynamic analysis aims at understand-
ing, evaluating, analyzing and modifying (if required) the
structural dynamic behavior which can be represented by
1 Introduction many terms such as natural frequencies, eigenvalues,
eigenvectors, damping ratios, Frequency Response Func-
Use of thin and light-weight products in modern day tions (FRFs) etc. The dynamic analysis of structures can
machines and structures is increasing day by day. be done through either experimental route or by using
Therefore better dynamic testing and analysis tools are theoretical approach [77, 33].
becoming the urgent need of hour. In the automotive, The theoretical route involves the formation of an ana-
aircraft and spaceship engines, there is an ever existing lytical model of the system either using a classical method
demand of attaining better fuel economy; which can be [46] or through Finite Element (FE) method [91]. The
met to a good extent by using thin products as well as application of classical method is generally limited to
with the use of light weight materials such as aluminium simple systems only, while FE method is preferred for real
and plastics composites instead of the conventionally used life complex systems. However FE method is not able to
heavy weight materials such as steels. Particularly in the predict the dynamic responses of structures with complete
case of satellites, some parts are so thin that they can get accuracy due to the presence of certain errors in the FE
model. Such errors are inherent in an FE model due to
following reasons:
& Harmesh Kumar
[email protected] 1. Faulty boundary conditions
2. Incorrect values of material properties
1
Department of Mechanical Engineering, University Institute 3. Discretization of continuum or a poor quality mesh
of Engineering and Technology, Panjab University, 4. Difficulty in modeling complex real life shapes
Chandigarh 160014, India

123
516 S. Sehgal, H. Kumar

5. Assumptions for simplification purpose such as con- 2 Finite Element Model Updating (FEMU)
sidering damping on a linear basis instead of non-linear Procedure
6. Incorrect modeling of joints
7. Use of rounding off methods in computations 2.1 Finite Element Method
Thus there is a need to correct an FE model so that its
In FE method, a complex continuous region of a structure
vibration behavior matches with the actual dynamic re-
is discretized into simple geometric shapes called finite
sponse obtained experimentally as shown in Fig. 1. The
elements. Figure 2 shows a cantilever beam of length
procedure used to update the model is called FE model
910 mm, width 50 mm and thickness 5 mm. The cantilever
updating (FEMU) [39]. In FEMU techniques, the ex-
beam is divided into 60 finite elements. Each element has
perimental and FE responses are first compared and cor-
two nodes. At each node, two degrees of freedom are
related so as to ensure that the FE model under
measured, out of which one is the displacement in y-di-
consideration qualifies for further updating procedure and a
rection and the other is the rotation about z-axis. Both the
conceptually new FE model is not required. A number of
degrees of freedom of node number ‘1’ are fully
graphical as well as numerical comparison and correlation
constrained.
techniques are used for this purpose. Further to avoid the
The finite elements can be axial elements, torque ele-
incompatibility in the sizes of experimental and FE data
ments, beam bending elements, thin plate bending ele-
sets the use of size compatibility techniques is required. In
ments, thick plate bending elements, etc. Approximate
FEMU, experimental results are considered as targets and
results of displacement shapes and stress fields can be
the inputs of FE model are adjusted in such a way that the
obtained for such finite elements using shape functions.
outputs of FE model have a better match with their ex-
Continuity across element boundaries can be maintained
perimental counterparts.
using either displacement or energy approaches. The dis-
A number of researchers have proposed the various
placement approach makes use of equilibrium, com-
structural model updating techniques both by classical ap-
patibility and the constitutive laws; while the energy
proach and FE route. While going through the available lit-
approach is based on the principal of virtual work. In en-
erature on this area, a need is felt to summarize all the results
ergy approach, the internal work is equated to external
and conclusions made by different researchers. Therefore,
work. For dynamic analysis purpose, each element needs to
this paper is an attempt to provide a review of major research
be expressed in form of elemental mass, stiffness and
activities carried out in the model updating field.
damping matrices. Equations (1) and (2) represent the
In this paper, first the basic theoretical issues related to
elemental mass and stiffness matrices for a beam element.
FEMU procedure have been discussed. After which, a state 2 3
of the art review of direct and iterative techniques of 78 22a 27 13
FEMU has been presented. Application areas of FEMU qAa 6
6 22a 8a
2
13a 6a2 7 7
½me  ¼ 4 ð1Þ
have also been discussed. The final part of the paper dis- 105 27 13a 78 22a 5
cusses the current problems and future directions for 13a 6a2 22a 8a2
FEMU related research. 2 3
3 3a 3 3a
EI 6 3a 4a2 3a 2a2 7
½ke  ¼ 3 6 7 ð2Þ
2a 4 3 3a 3 3a 5
2 2
3a 2a 3a 4a
where me , q, A, a, ke , E and I are respectively the element
mass matrix, density, area of cross-section of element, half-
length of element, element stiffness matrix, Young’s
modulus of elasticity and moment of inertia of cross-sec-
tion of beam element. Subsequently, the individual ele-
ments are assembled to form their global counterparts,
which are also jointly known as system matrices. These
system matrices along with certain boundary conditions are
used to formulate a set of governing equations, which are
then processed on a computer to evaluate dynamic char-
acteristics of the system.
Fig. 1 Comparison of experimental, initial analytical and updated From above discussion it can be concluded that the
analytical results dynamic response of the structure depends upon a number

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 517

Fig. 2 FE model of a cantilever

beam structure

of parameters. These parameters are called input pa-

rameters which affect the dynamic response parameters.
These input parameters may be belonging to material,
structural, finite element or computational categories. The
various input parameters are assembled in the form of
cause and effect diagram and are shown in Fig. 3.
Dynamic response is generally measured in terms of
eigenvalues, eigenvectors, and FRFs. The frequency at
which a structure naturally vibrates once it is set into
motion is called its natural frequency. Square of the natural
frequency is known as eigenvalue. Eigenvector or mode
Fig. 4 Modeshapes of cantilever beam structure. a First, second and
shape of a particular mode defines the displacement con- third modeshapes, b Fourth, fifth and sixth modeshapes
figuration of structure at corresponding natural frequency.
First six modeshapes of a cantilever beam structure are as
experimental results. Former type of results are generated
drawn in Fig. 4. Extreme and intermediate positions of
by varying some parameters of the FE model thereby
different points of a cantilever beam are drawn in Fig. 5.
producing a perturbed FE model, whose simulation results
FRF (i, j) is defined as ratio of harmonic response at lo-
are assumed as experimental results and are surely different
cation ‘i’ and the harmonic excitation force at location ‘j’. If
from the dynamic results of original FE model. Real life
the measured response is in the form displacement, then
experimental results can be generated through modal
corresponding FRF is called as receptance (or admittance or
testing.
dynamic compliance or dynamic flexibility) FRF. Other-
A representative experimental set-up for finite element
wise, velocity or acceleration response signals can be used to
model updating of a structure is shown in Fig. 7. One end
produce mobility or accelerance FRF respectively. Accel-
of the test structure can be fixed to a non-vibrating base
erance FRF is also sometimes written as inertance FRF.
through welding, bolting or riveting etc. thereby resulting
Further any FRF is called as a point or transient FRF de-
in grounded support condition. Test structure can also be
pending upon whether the response and force are measured at
supported on very soft springs or light elastic bands, which
same or different locations respectively. Point and transient
leads to a free support condition. In such situation, it is
Receptance FRF have been drawn in Fig. 6a, b respectively.
important to attach the suspension as close as possible to
nodal points of the particular mode. Another method of
2.2 Experimental Method
supporting the structure is in situ type, where the test
structure is connected to some other component which
In FEMU, the experimental results can be obtained by two
presents a non-rigid attachment. Input to the test set-up
ways: simulated experimental results and real life
generally includes a short impulse or, a sine wave excita-
tion over a frequency range of interest, or a white noise.
This force input is provided by either a shaker, or an im-
pulse hammer thereby resulting in forced vibration and free
vibration cases respectively. Structure can also be excited
by suddenly releasing it from a deformed position. Selec-
tion of any particular type of shaker depends upon its
frequency range, acceleration, rms (root-mean-square) ve-
locity, rms displacement, maximum load, operating tem-
perature range, power supply, overall dimensions and
mass. Impulse hammer is selected based upon its force
range, frequency range, sensitivity, time constant, operat-
ing temperature range, power supply, mass, diameter and
Fig. 3 Cause and effect diagram length of head. Output of the experimental set-up is

123
518 S. Sehgal, H. Kumar

Impact Hammer
Force Signal

Accelerometer

Data Acceleration
Acquisition Signal
System
FRF Data
Modal Data Updated
FE Model
Modal Analysis FEMU System
System

Fig. 7 Representative experimental set-up for FEMU

based upon their characteristics such as mass, acceleration

range, threshold acceleration, sensitivity, frequency range,
and temperature sensitivity, operating temperature range
and power supply. Signal from accelerometer is then
conditioned using a signal conditioner, which is selected
depending upon its sensor excitation current, sensor exci-
tation voltage, frequency range, output voltage, gain,
power supply, operating temperature range, overall di-
mensions and mass. Conditioned signal is then passed on to
Fast Fourier Transform analyzers so that the time domain
signal can be converted to frequency domain. Such fre-
quency domain data is then analyzed on a computer to
identify the experimental dynamic response of the test
structure. Experimental dynamic response is then com-
pared with its FE counterpart to examine any errors and
Fig. 5 Extreme and intermediate positions drawn using solid and
dashed lines respectively. a First mode, b second mode, c third mode, also to ascertain sufficient degree of correlation between
d fourth mode, e fifth mode, f sixth mode the two as discussed in next section.

2.3 Comparison and Correlation Techniques

generally in the form of a function of acceleration, which is
measured at a number of points using sensors like ac- Before applying any FEMU technique, the experimental
celerometers etc. Selection of such sensors is generally and FE data sets need to be compared so as to ensure

Fig. 6 Receptance FRF. a Point

FRF, b transient FRF

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 519

existence of some correlation between experimental and through modal testing as shown in Table 1. If the per-
FE responses and also to determine that whether it is worth centage errors are small, one can update the FE model of
to update the proposed FE model or a completely new the structure in order to minimize such errors. But if the
model is required. These techniques include comparison of errors are very large, say, more than 200 %, then it is not
FRFs, natural frequencies and mode shapes (Modal Scale advised to update the model rather a new formulation of the
Factor (MSF), Modal Assurance Criterion (MAC), Nor- FE model is required.
malised MAC and Coordinate MAC (COMAC) etc.).
These techniques are discussed as:- 2.3.3 Comparison of Modeshapes

2.3.1 Comparison of FRFs Mode shape portrays the pattern or configuration in which
the structure vibrates at any particular natural frequency.
A simple method of comparing experimental and FE re- Mode shapes are inherent properties of a structure. These
sults is to plot experimental and FE FRFs on a single graph do not depend on the forces or loads acting on the structure.
as shown in Fig. 8. A visual comparison of FRFs is done to These will change if the material properties (mass, stiff-
find out presence of any correlation between the ex- ness, damping properties), or boundary conditions (sup-
perimental and FE results. For example, Fig. 8a shows a ports) of the structure change. A simple method for
clear correlation between the experimental and FE results. comparing mode shapes is through overlaying technique as
This is the case where FE model updating is required to shown in Fig. 8. By this method, the difference between
increase the correlation further, while Fig. 8b shows the experimental and FE mode shape is drawn on a graph. In
case of total mismatch between the measured and predicted this method the difference between two corresponding
results. Under such conditions, instead of FE model up- mode shapes should approach zero. Main limitation of such
dating, a conceptually new FE model is required. graphical methods is that they are not so supportive if the
correlation process is to be automated. For automatic cor-
2.3.2 Comparison of Natural Frequencies relation of modes, we need some quantitative (numerical)
measure of correlation, which can be easily implemented
It is important to compare the natural frequencies obtained through a computer program. One such measure is MSF,
from FE analysis of the structure with the ones acquired which is slope of the best straight line through the modes
plotted on an x–y graph with experimental mode on one
axis and analytical on other [2]. Desired value of MSF is
unity for good correlation. MSF between experimental and
analytical mode shapes can be obtained using Eqs. (3) and
(4); where {[X}i and {[A}j represent the ith experimental
and jth analytical mode shape respectively; while super-
script ‘T’ denotes the transpose of the corresponding
vector.
  T
f;X gi f;A gj
MSF f;X gi ; f;A gj ¼ T ð3Þ
f;A gj f;A gj
  f;A gT f;X g
j i
MSF f;A gj ; f;X gi ¼ T ð4Þ
f;X gi f;X gi
Main drawback of MSF is that it does not provide any
information regarding the scatter of the x–y plots. To avoid
this problem, it is recommended to use MAC. This crite-
rion is also known as mode shape correlation coefficient
[72]. It can be calculated using Eq. (5).
 2
   T 
 f;X gi f;A gj 
MAC f;X gi ; f;A gj ¼     ð5Þ
T T
f;X gi f;X gi f;A gj f;A gj

Fig. 8 Comparison of FRF plots. a Good correlation, b poor MAC is a measure of scatter of points from the straight line
correlation correlation. MAC value equal to unity means perfect

123
520 S. Sehgal, H. Kumar

Table 1 Comparison of natural Mode no. Experimental (Hz) FE (Hz) Difference in frequencies (%)
frequencies
1 24.1 24.3 0.8
2 76.1 73.1 3.9
3 119.3 116.1 2.7
4 158.0 153.9 2.6
5 184.1 180.7 1.8
6 211.6 200.5 5.2

correlation, while a MAC value equal to zero means no PL  2


l¼1 ð;X Þkl ð;A Þkl
correlation between two modes. Figure 9a shows a typical COMACðkÞ ¼ PL 2 P L 2
ð7Þ
MAC plot between perfectly correlated experimental and l¼1 ð;X Þkl l¼1 ð;A Þkl
FE modes. It is seen that all the diagonal elements are equal COMAC should be calculated only after finding ‘L’ num-
to unity and off-diagonal elements are nil, which means ber of the correlated mode pairs through MAC or NMAC.
perfect correlation between experimental and FE modes. A COMAC value close to ‘1’ indicates good correlation at a
poor correlation case is shown in Fig. 9b, where none of particular coordinate say ‘k’. The COMAC plots showing
the diagonal element is unity; and off-diagonal elements good and poor correlation are drawn in Fig. 10a, b
are also not zero. respectively.
One major limitation of MAC is that it cannot identify MAC or its variants consider mode shape matching at
the systematic deviations, which can be overcome to some only discrete coordinates. If one wants to consider the
extent by combining MAC with the best fit straight line mode shape correlation on a continuous basis, then image
plots described earlier. Another limitation of MAC is that it processing and pattern recognition based Zernike moment
is not a true orthogonality check because of absence of descriptor can be used [107]. Further, a scanning laser vi-
mass or stiffness matrix in the formula of MAC. This brometer can also be used for 3D digital image correlation
limitation can also be avoided by using normalized MAC [48].
(NMAC) [72]. The NMAC is also called as normalized After comparison and correlating the experimental and
cross orthogonality as represented by Eq. (6). FE data sets, next step is to make these compatible with
 
each other so that FEMU can be performed. Compatibility
NMAC f;X gi ; f;A gj
 2 techniques have been explained in next sub section.
 T  ð6Þ
 f;X gi ½ W  f;A g j 
¼   2.4 Size Compatibility Techniques
T T
f;X gi ½Wf;X gi f;A gj ½Wf;A gj
Measured degrees of freedom are generally far lesser than
NMAC includes a weighting matrix ½W, which can be their FE counterparts due to limited number of sensors used
replaced by either mass matrix or stiffness matrix. NMAC in experimental set-up. Moreover some degrees of free-
has a limitation of not describing the spatial distribution of doms are very difficult to be measured (e.g. rotational de-
correlation, for which purpose, one can use Coordinate grees of freedom or those corresponding to the points
MAC (COMAC) as given in Eq. (7) [70] as:- which are physically inaccessible). Thus the size of FE

Fig. 9 Modes correlation using

MAC. a Good correlation,
b poor correlation

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 521

Fig. 10 Modes correlation

using COMAC. a Good
correlation, b poor correlation

modal matrices is generally not compatible with their ex- analytical model so that the size of the FE model matrices
perimental counterparts, while in FEMU one-to-one cor- can be reduced and brought closer to their experimental
respondence between the two results sets is required; which counterparts. A simple method of model reduction is to
can be achieved by either expanding the experimental re- eliminate those degrees of freedom (DOF) which are not
sults or by reducing the FE model as explained in following available in experimental data. Model thus obtained is
sections. known as reduced model. Main drawback of this method is
that mass and stiffness terms related to the eliminated DOF
2.4.1 Coordinate Expansion are completely lost and nowhere compensated [33].
Another method of model reduction is to produce a
Measured data sets can be expanded so that they are of the condensed model which will represent the entire structure
same size as their FE counterparts. Such a process is called completely but approximately. In this method a condensed
as coordinate expansion. A simple method of coordinate model is obtained by transforming the original model as
expansion is to substitute the unmeasured coordinates by given in Eq. (9); where ‘n’ and ‘N’ are the sizes of reduced
their FE counterparts. This method is computationally very and finite element model based eigen-vectors respectively.
efficient. Main disadvantage of this method is that some- f;reduced gn1 ¼ ½TnN f;FE gN1 ð9Þ
times it leads to erroneous solutions during FEMU. To
avoid this problem, some transformation matrix based co- One method of obtaining condensed model is through static
ordinate expansion should be done as used in Eq. (8); reduction [45], in which size of system matrices is reduced by
where ‘N’ and ‘n’ are the sizes of expanded and measured neglecting inertia and stiffness terms associated with un-
eigen-vectors respectively. measured DOFs. This method is generally known as Guyan
  reduction method. Main drawback of static reduction method
;expanded N1 ¼ ½T Nn f;measured gn1 ð8Þ
is that it does not reproduce any of the eigenvalues or eigen-
For an undamped system, the transformation matrix ½T can vectors of the original full FE model. Another method of
be obtained by writing the governing equations of the sys- model reduction is improved reduced system (IRS) method
tem in partitioned matrices form using Kidder’s method which can closely reproduce the eigenvalues and eigenvectors
[60]. In Kidder’s method, sub-matrices are related to mea- of the original FE model [88, 40]. If reduced model is required
sured and unmeasured eigenvectors. For a damped system, to have eigenvalues and eigenvectors exactly same as that of
partitioned matrices based method can be extended to deal original FE model then SEREP should be used [89, 97].
with complex measured modes [47]. Another method of
coordinate expansion is based on the concept that unmea-
sured eigenvectors can be expressed as a linear combination 3 Review of FEMU Techniques
of measured eigenvectors [90]. System equivalent reduction
and expansion processes (SEREP) [89], Curve fitting [110] The FEMU techniques can be broadly classified into two
and FE eigenvectors in conjunction with MAC matrix based categories:-
methods are also used in coordinate expansion [71].
• direct (non-iterative) techniques and
• iterative techniques
2.4.2 Model Reduction
Direct techniques can provide the solution to model
Model reduction process is basically the inverse of the updating problem in just a single step; hence these are
expansion process. This process is generally applied to an computationally very efficient, and divergence related

123
522 S. Sehgal, H. Kumar

problems do not occur. Another important feature of direct orthogonality constraints. During same time, Baruch also
techniques is that they reproduce the measured data ex- developed a direct method in which the FE stiffness matrix
actly. Therefore measurement noise and spurious modes and FE eigenvectors are updated in such a manner that
are also reproduced by such techniques. Thereby, these some weighted norm of the difference between the updated
techniques required very good quality modal testing and and analytical stiffness matrices is minimized using La-
analysis procedures. Direct techniques compute a closed- grange multipliers [10]. In this method, the FE mass matrix
form solution for system matrices using the structural is assumed correct and hence not updated. Later in 1979,
equations of motion and orthogonality properties of modes. Berman adopted the mathematical approach of Baruch [10]
These techniques are also called matrix techniques, be- and proposed a direct method for updating the mass matrix
cause such techniques found the solution in the form of [14]. This method also included an additional constraint to
updated system matrices by solving a set of matrix equa- preserve the symmetry of updated matrices. These tech-
tions. The main drawback of direct techniques is that up- niques [11, 10, 14] are also called as techniques of refer-
dated mass and stiffness matrices may not be symmetric ence because one of the three quantities (eigenvectors,
and positive definite. It becomes very difficult to under- mass and stiffness matrices) is assumed to be reference and
stand such kind of system matrices on a physical basis. the other two are updated.
Dynamic response of FE model of any structure is a In 1983, Berman and Nagy proposed another method in
function of a number of parameters, as explained in which the basic approach of the method of Baruch [10] and
Sect. 2.1. The iterative techniques compute updated values the method of Berman [14] were combined to update both
of material and structural parameters of FE model in such a mass and stiffness matrices of the system in a sequential
way that during each iteration mismatch between ex- manner [15]. Method of Berman and Nagy has been used
perimental and FE response is reduced. These techniques by various researchers in a number of applications. Ap-
are also called as gradient based techniques. Iterations are plication of this method was also studied by Modak et al.
stopped when the values of updating parameters stop [83] for dynamic design of a fixed–fixed beam and an
converging or the error function is reduced to tolerable F-structure. Bais et al. [9] applied the direct techniques of
level. Error function is generally a non-linear function of Baruch [10] and also of Berman and Nagy [15] for dy-
experimental and FE responses such as eigenvalues, namic design of a drilling machine. Dhandole and Modak
eigenvectors or FRFs. Iterative techniques result in only [30] performed the FE model updating of vibro-acoustic
symmetric and positive definite updated system matrices, cavities using the direct method of Berman and Nagy [15].
which can be easily understood on a physical basis. But They investigated a simulated example of a two dimen-
these techniques require a number of iterations before ar- sional rectangular cavity with a flexible surface having
riving at the final result. Thus iterative techniques are structural modeling errors related to the material property,
computationally less efficient and divergence related geometry and boundary conditions; under incomplete and
problems can also arise during iterations. The technology noisy simulated experimental data. Effectiveness of the
and research developments reported for both direct and method was compared with an iterative method on the
iterative techniques are discussed in following sub- basis of accuracy of prediction of vibro-acoustic natural
sections. frequencies and the frequency responses both inside as well
as outside the frequency range taken during updating. It
3.1 Direct Techniques of FEMU was concluded in the study that the direct technique re-
sulted in an accurate prediction of the vibro-acoustic nat-
Some researchers have developed direct techniques which ural frequencies and the response inside the frequency
consider adjustment of elements of system matrices on a range considered during model updating. However, beyond
mathematical basis rather than a physical basis. A few the updating frequency range, the predictions based on the
researchers have also worked for development of such di- direct updated vibro-acoustic models are not as accurate as
rect techniques, which can consider adjustment of physical given by the iterative technique. Further the direct tech-
properties of system matrices. The direct techniques are nique seems to be quite versatile compared to its iterative
further classified into two categories, viz. matrix elements counterpart for dealing with complex cavities. This is due
adjustment based direct techniques and physical property to the reason that for complex cavities, complete knowl-
adjustment based direct techniques. edge of modeling inaccuracies and selection of updating
Foremost, in 1978, Baruch and Bar-Itzhack developed a parameters becomes very difficult. Recently, Modak [81]
direct method of FEMU by assuming FE mass matrix to be also developed a direct vibro-acoustic model updating
correct [11]. They updated the FE eigenvectors and FE technique using modal test data. This technique can be
stiffness matrix by minimizing the weighted norm of dif- used for updating the vibro-acoustic FE model of such
ference between measured and FE eigenvectors subject to systems that involve an elastic structure enclosing a

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 523

medium, like air. By using this technique, mass and stiff- development of direct techniques which aim at reducing
ness matrix of both the structural as well as the acoustic the mismatch between experimental and FE responses by
parts of the model can be updated by preserving their adjusting the physical properties of the FE model.
symmetry. In 2007, Hu et al.[49] proposed the cross-model-cross-
The method of Berman and Nagy [15] was further ex- mode method (CMCM) in which adjustment of physical
tended by Ceasar [19] by including additional constraints properties of the system matrices is carried out to update
related to preservation of the total mass of system and the the mass and stiffness matrices simultaneously. Satisfac-
interface forces. Matrix perturbation concept was used by tory performance of the method was demonstrated by ap-
Chen et al. [26] to simultaneously update mass and stiff- plying it for simulated examples of a shear building model
ness matrices of the system. Based upon first order ap- and a three-dimensional frame structural model. During the
proximation Sidhu and Ewins [99] developed a direct demonstration, some preset error coefficients were intro-
method using error matrix approach. This method was duced into the FE model of the structures. Then the CMCM
useful for only small modeling errors. Kabe [57] suggested method was used to estimate the correction factors of
a direct method which could retain the bandwidth of sys- FEMU. Main disadvantage of the method is that it requires
tem matrices by identifying the null elements of original the measurement of spatially complete modal data, which
system matrices and constraining them to remain zero is generally not possible for large-scale real life structures.
during FEMU. Further, in 2011, substructure energy approach (SEA)
Further, a range of direct updating techniques were based physical property adjustment type direct method was
formulated by Caesar [20] considering different objective developed by Fang et al. [34]. Main benefit of this method
functions, constraints and whether mass matrix or stiffness is that it can handle spatially incomplete experimental
matrix is updated first. Later, Lim [73] proposed a FEMU modal data also as opposed to the CMCM method. In this
method in which sub-matrix-scaling factors are used as method the complete system is divided into several sub
updating parameters. Due to the use of sub-matrix-scaling systems. Instead of updating the complete system, only the
factors number of unknowns is reduced appreciably. Smith critical sub systems are identified and updated using a set
and Beattie suggested a quasi-Newton method for stiffness of linear simultaneous equations deduced from the energy
updating which preserve structural connectivity and can functional of substructure models and substructure modes.
also handle noisy modal data [101, 13]. Application of this For validation purpose, they applied the method for up-
method was studied later by Ramamurti and Rao [95] by dating the models of a mass-spring system, a two-dimen-
using different finite elements. They updated the FE sional and a three-dimensional lattice structure using
models of a rectangular plate with holes using 3-D plate simulated experimental data. Results obtained by Fang
elements; a crane using 3-D beam elements and a rectan- et al., for numerical example of a three-dimensional lattice
gular plate using brick elements. Bucher and Barun [17] structure showed that the method worked satisfactorily for
developed modal data based direct method having the ca- FEMU purpose. Jacquelin et al. [53] developed a direct
pability to deal with partially known experimental eigen- probabilistic model updating technique that takes into ac-
solutions. The method was verified by applying it to nu- count the uncertainties related to experimental results by
merical examples of a clamped beam and a serial spring- using the random matrix approach. Recently, Jiang et al.
mass system. Farhat and Hemez [37] proposed a FEMU [56] reduced the model updating process to the problem of
method using an element-by-element sensitivity method- the best approximation type. This technique is successfully
ology and demonstrated the potential of the method by used for numerical examples of undamped systems. How-
using several simulation examples. Later, Friswell et al. ever the technique needs to be tested further for damped
[41] developed a method for simultaneous updating of simulated systems as well as for actual experimental
damping and stiffness matrices assuming the mass matrix results.
to be correct. Singular value decomposition and matrix Major contributions related to direct techniques of
approximation technique based method was proposed by FEMU have been presented in a brief chronological format
Xiamin [111]. This method is particularly useful if the in Table 2.
mismatch between experimental and FE responses is very
large. Recently, Carvalho et al. [21] developed a method 3.2 Iterative Techniques of FEMU
which can identify and prevent the reproduction of spuri-
ous modes into updated results. The direct techniques Algorithm for iterative techniques of FEMU is drawn in
discussed so far mainly aim at removing the mismatch Fig. 11. In such techniques an error function is minimized
between experimental and FE responses, without bothering iteratively to find the updating parameters or correction
for any adjustment of physical properties of system ma- factors. These techniques originated in 1974, when Collins
trices. Presently research is also oriented towards et al. proposed the eigendata sensitivity based iterative

123
524 S. Sehgal, H. Kumar

Table 2 Major contributions in direct techniques of FEMU

References Major Contribution

Baruch and Bar-Itzhack [11] Updating of stiffness matrix

Berman [14] Updating of mass matrix
Berman and Nagy [15] Updating stiffness, mass matrices in sequential manner
Ceasar [19] Improvement of Berman and Nagy method by preservation of total mass and interface forces
Chen et al. [26] Use of matrix perturbation concept for simultaneous updating of mass and stiffness matrices
Sidhu and Ewins [99] Error matrix approach based direct updating technique to handle small modeling errors
Kabe [57] Constraint on retaining bandwidth of system matrices
Caesar [20] Formulation of a number of direct techniques
Lim [73] Sub-matrix scaling factors based updating technique
Smith and Beattie [101] Updating stiffness matrix with fix structural connectivity
Bucher and Braun [17] Updating for incomplete experimental modal data
Farhat and Hemez [37] Element-by-element sensitivity methodology based updating technique
Friswell et al. [41] Simultaneous updating of damping and stiffness matrices
Ramamurti and Rao [95] FEMU using different finite elements
Xiamin [111] Matrix approximation technique based FEMU method
Modak et al. [83] Dynamic design of fixed–fixed beam and F-structure using the method of
Berman and Nagy
Bais et al. [9] Dynamic design of drilling machine using method of Baruch and also Berman and Nagy
Carvalho [21] FEMU method which can identify and prevent the reproduction of spurious
modes into updated results
Hu et al. [49] Physical property adjustment based cross-model-cross-mode method
Dhandole and Modak [30] FEMU of vibro-acoustic cavities using the method of Berman and Nagy
Dhandole and Modak [31] Application of method of Berman and Nagy in vibro-acustics field
Fang et al. [34] Substructure energy approach based physical property adjustment type direct method
Jacquelin et al. [53] Development of probabilistic model updating
Jiang et al. [56] Development of direct updating technique using best approximation method
Modak [81] Updating of vibroacoustic finite element models using modal test data

method also known as inverse eigensensitivity method modal extraction to be performed on measured data thereby
(IESM) [27]. Later, Chen and Garba [25] used matrix eliminating the chances of errors due to modal analysis
perturbation technique for iteratively computing the being transmitted to updated results. Application of RFM
eigensolution and eigendata sensitivities. This method was for dynamic design of a fixed–fixed beam and an F-struc-
further improved by Kim et al. [62] by including second ture was discussed by Modak et al. [83]. Effectiveness of
order sensitivities. Further the convergence of this method IESM and RFM has been compared by Imregun et al. [52],
was improved by Lin et al. [75] by employing both the FE Modak et al. [84]. For incomplete experimental data case
and the experimental modal data for evaluating sensitivity (where experimental eigenvector is not completely known)
coefficients. This also helped in application of the method with no noise, RFM works better than IESM, while latter
for cases of large errors. performs better in the presence of noise particularly when
In IESM, modal data such as eigenvalues, eigenvectors the updating range covers a greater number of modes.
and damping ratios are used to form an error function. Moreover, in RFM, if the number and location of testing
Modal data is obtained through modal analysis of measured frequencies are not selected properly then the method may
FRFs. If modal analysis is not done carefully then the ex- not be able to converge as reported by Modak et al. [84]
tracted modal data may contain certain errors, which is based upon their comparative study using simulated ex-
further transmitted to the results of FEMU. To avoid this perimental data.
problem one can use the measured FRFs directly for model While formulating the basic RFM, Lin and Ewins [76]
updating using Response Function Method (RFM) as pro- had not taken into account damping. Later, Arora et al. [3,
posed by Lin and Ewins [76]. RFM does not require any 4] proposed two techniques of extending the basic RFM in

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 525

Fig. 11 Algorithm for iterative

techniques of FEMU

order to widen the applicability of the method by consid- identification [5]. Arora et al. [6] also worked on the use of
ering damping. First proposed technique was to combine complex parameters based RFM of FEMU for dynamic
RFM with damping identification method of Pilkey [92]. design purposes. Arora [7] also compared the accuracy of
This technique involved a two step procedure in which first basic RFM [76] against the direct method Bernam and
step involved updating of only mass and stiffness matrices Nagy [15] and found that the performance of basic RFM
while damping matrix was obtained in second step by using was superior to the direct method in terms of the accuracy
the mass and stiffness matrices found in first step. in prediction of FRFs. Recently, in 2012, the basic RFM
Second proposed technique by Arora et al. [4] was to [76] was further extended by Pradhan and Modak [93] to
consider the model parameters and system matrices in a develop a normal RFM that was based on the estimates of
complex form so as to deal with complex modes of damped normal FRFs computed by using only stiffness and mass
structures. Later they compared these two techniques and matrices. Normal RFM was later used by Pradhan and
found that complex parameter based FEMU technique Modak [94] for damping matrix identification using FRF
gives better results than the one with damping experimental data.

123
526 S. Sehgal, H. Kumar

In 2000, Modak et al. [82] developed a constrained excitation force is not known or difficult to measure.
optimization based FEMU technique. They evaluated the Jamshidi and Ashory [55] compared and found that MUBE
performance of the said technique by applying it to a fixed– gives better updating results than RFM for incomplete
fixed beam structure. This technique was computationally experimental data with measurement noise.
more intensive than IESM but was able to address the Further the problem of model updating for a real life
difficulty that might arise due to large difference between structure is quite complex due to presence of non-linearity,
the sensitivities of natural frequencies and mode shapes. damping, measurement errors and large number of updat-
Later, they also used the constrained optimization based ing parameters. Under such conditions, the traditional
FEMU for dynamic design of fixed–fixed beam and model updating techniques may fail to converge or may get
F-structure [85]. Use of this technique for better FE for- stuck in local minima rather than finding a global minimum
mulation of acoustic cavities was done by Dhandole and of the error function. In that kind of situations, it is ad-
Modak [32]. visable to use such optimization techniques which are ca-
Neural network (NN) method has also been applied for pable of finding the global optimum results even for a
FEMU by Atalla and Inman [8] for FE model updating of a complicated optimization problem. Few such promising
flexible frame. In this method, a NN is trained and techniques are Simulated Annealing (SA), Genetic Algo-
validated using FE responses. Thereafter this NN can be rithms (GA) and Particle Swarm Optimization (PSO).
employed to obtain updated physical parameters by taking Levin and Lieven [68] used and compared SA and GA for
experimental responses as inputs. Once the NN model is FEMU and observed that SA based FEMU method per-
properly trained, the NN based calculations are relatively forms better than its GA based counterpart. Success of both
fast compared to conventional optimization techniques SA and GA based FEMU techniques depends upon ap-
regardless of the complexity of the real life structure. NN propriate choice of updating parameters. If updating pa-
based model updating method is also robust to the noise rameters are not selected properly, then both these
present there [67]. Main limitation of this method is that it techniques give unsatisfactory results. Marwala [79] ap-
requires a large number of training data ‘qp’ (where ‘p’ is plied the PSO method in the field of FEMU. PSO is a
the number of updating parameters and ‘q’ is the number of population based stochastic search algorithm derived from
levels or values which any updating parameter can take). social-psychological behavior of biological entities (birds,
However this problem can be circumvented by using an fish, ants, etc.) when they are foraging for resources (food).
orthogonal array method in which the number of training It is particularly useful if the number of updating pa-
data sets are reduced to p(q - 1) ? 1 only [23, 16]. rameters is very large. Further, Mthembu et al. [87] used
Because of the experimental limitations, if only natural PSO for selecting the best model from amongst a number
frequencies (or eigenvalues) are measured then FEMU can of updated structural models.
also be performed by using reduced order characteristic Kwon and Lin [64] suggested a robust FEMU technique
polynomial (ROCP) based method as proposed by Li [69]. using the concept of Taguchi method and were able to
Li applied the ROCP based method for FE model updating obtain good FEMU results. They used both frequency as
of a beam structure elastically constrained at one end. In well as modal data for formulation of the objective func-
this method a polynomial is defined in terms of the mea- tion. This technique was reported to be robust against
sured eigenvalues, FE eigenvalues, FE eigenvectors and various noises because the parameters were updated in
updating parameters. Assuming the measured natural fre- such a way that signal to noise ratio is maximized.
quencies to be the roots of the polynomial, a set of non- It is to be mentioned here that due to their iterative nature,
linear equations are derived. This set of non-linear equa- the iterative techniques seem to be computationally ineffi-
tions is further solved for calculating the values of the cient, particularly, if applied to FE models of large size
updating parameters. structures. Working on these directions Guo and Zhang [44]
Sometimes, it becomes difficult to measure FRFs ac- suggested the use of response surface methodology (RSM)
curately due to small size or delicate nature of test structure based FEMU. In this technique, FE calculations are not re-
[like a hard disk drive (HDD)]. Under such conditions quired at each iteration step, thereby making it computa-
Model Updating using Base Excitation (MUBE) method tionally efficient. In this method, n-dimensional response
can be applied as proposed by Lin and Zhu [74]. They used surfaces are created by taking updating parameters as inputs
MUBE method for FEMU of a cantilever beam and a truss and dynamic FE responses as outputs. Updated values of
structure, and showed that the method works satisfactorily. model parameters are found by using the response surfaces
In this method, base of the structure is excited with an and the measured responses by minimizing the error func-
unknown force input using an electric shaker and the dis- tion. RSM based FEMU is reported to be as accurate as the
placement output measured from the test set-up is used for sensitivity based method, and, moreover it is more robust and
model updating. This method is of great use when computationally efficient than the sensitivity based FEMU.

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 527

Later, Ren and Chen [96] applied RSM as well as sensitivity selection procedure, Kim and Park successfully reduced the
based FEMU techniques to a precast concrete bridge and number of updating parameters to 150; which was further
compared the convergence of objective function by both the drastically reduced to 20 using the second phase of their
techniques. Their study showed that for a given number of suggested procedure.
iterations, the objective function of FEMU is reduced to a Moreover the updating parameters are generally the
lower level by using the RSM based FEMU technique than physical parameters such as thickness, density, modulus of
its sensitivity based counterpart. In traditional RSM based elasticity, damping, Poisson’s ratio etc. of structure. These
FEMU technique, the experimentally measured signals were parameters do not have a single discrete value throughout
first transformed to one or more response parameters such as the body of a real life object. Spatial distribution of such
natural frequencies. This reduced the information in the parameters was considered by Adhikari and Friswell [1] by
training data used for developing response surface models. In expressing the updating parameters as spatially correlated
order to circumvent this problem, time domain based results random fields.
were used by Shahidi and Pakzad [98] for RSM based FEMU If a number of identical test structures are taken, all may
technique. Time domain based technique was helpful in not have same value of a particular material parameter viz.
extracting more information from measured signals and modulus of elasticity, density etc. This variability in
compensate for the error present in the meta-models. Effi- seemingly identical test pieces may arise due to many
ciency of RSM based FEMU technique was further increased sources such as geometric tolerances, manufacturing pro-
by Chakraborty and Sen [22] whilst developing an adaptive cesses etc. Thus each material parameter has a stochastic
RSM based FEMU technique by replacing the least square nature with a mean value and a variance. Such stochastic
method with the moving least square method. nature of material parameters was taken into account dur-
It is important to mention that FE model of a real life ing FEMU by Mares et al. [78] using a simulated example
structure contains a large number of parameters. If all the and obtained quite satisfactory results. In this method, the
parameters are taken as updating parameters, then the mean value was represented by the centre of the scatter
model updating problem becomes too complex and time ellipse while the size and orientation of the ellipse was
consuming. A large number of updating parameters also determined by variance of test data. The updated analytical
result in ill-conditioning of the problem or trapping in scatter ellipse overlays quite closely the experimental
many local minima [61, 39]. To identify and select only a scatter ellipse. Same work was then extended by Motter-
few important updating parameters, Fissette et al. [38] shead et al. [86] by carrying out stochastic FEMU for a set
proposed a force balance method. This method can be used of physical structures.
for error location and then updating parameters are selected The stochastic FEMU based approach is based on
from only the regions of modeling errors. Waters [108] probabilistic models, which require large number of tests to
proposed a modified force balance method in which it is be conducted and also the large volumes of test data; which
assumed that the regions of response errors are not nec- demands for high experimental as well as computational
essarily an indication of modeling error. Recently, Kim and efforts. In order to get rid of the large quantities of test
Park [61] developed an automated parameter selection data, one can opt for the approach that was used by Kho-
procedure. This is a two phase method, wherein, during daparast et al. [59] for performing interval FEMU in which
first phase an updating parameter is assigned to each er- ranges or distributions of updating parameters are found
roneous FE. Then two neighboring updating parameters are rather than just one ‘true’ value of any updating parameter.
merged if their sensitivity (of response w.r.t. parameters) is Table 3 presents the major technological developments and
of same sign thereby reducing the number of updating advancements in the field of iterative techniques of FEMU.
parameters. This is repeated until all the neighboring up-
dating parameters have opposite sign of sensitivity. If the
number of updating parameters is acceptable then the 4 Applications of FEMU Techniques
procedure can be stopped; otherwise one can move on to
the second phase. In second phase those two neighboring FEMU is widely applicable in the field of better FE model
parameters of opposite sign are found and grouped together formulations, damage analysis of structures, non-destruc-
which will result in least reduction in total sensitivity. This tive characterization of material properties and also for
method was validated by applying it to the FE model of dynamic design purposes. With the aim of better FE model
cover of a HDD having 1115 FEs and 6732 dofs. Kim and formulations, FEMU has been used in a number of appli-
Park used the force balance method of Fissette and Ibrahim cations such as aircrafts, satellites, automobiles, nuclear
for error localization purpose and found that out of a total power plants, rotor bearing systems, laser spot welds,
of 1115 FEs only 628 FEs contained the modeling errors. bridges, dams, multi-storey buildings, steel frames, truss
After the application of first stage of automatic parameter structures, acoustics, etc. It also finds applications in

123
528 S. Sehgal, H. Kumar

Table 3 Major contributions in iterative techniques of FEMU

References Contribution

Collins et al. [27] Development of inverse eigensensitivity method (IESM)

Chen and Garba [25] Use of matrix perturbation technique for improved computation of sensitivities of IESM
Kim et al. [62] Improvement of IESM by Including second order sensitivities
Fissette et al. [38] Selection of updating parameters using force balance method for FEMU
Lin and Ewins [76] Development of response function method (RFM)
Lammens et al. [65] Application of FEMU in structural optimization of a tennis racket
Imregun et al. [52] Comparison of IESM and RFM based FEMU techniques
Lin et al. [75] Improvement of convergence of IESM
Waters [108] Development of modified force balance method for error localization in FEMU
Levin and Lieven [68] Development and comparison of simulated annealing and genetic algorithm based FEMU techniques
Atalla and Inman [8] Development of neural networks based FEMU technique
Levin and Lieven [67] Robust FEMU technique using neural networks
Cunha and Piranda [28] Application of FEMU in identification of elastic constants of composite materials
Wahab et al. [106] Application of FEMU in damage analysis of reinforced concrete beams
Modak et al. [82] Development of constrained optimization based FEMU technique
Chang et al. [23] Improvement of neural network based FEMU technique using orthogonal array method
Li [69] Reduced order characteristic polynomial (ROCP) method
Sinha et al. [100] Application of FEMU in dynamic design of spacer locations of tubes of nuclear power plants
Modak et al. [83] Application of RFM for dynamic design of a fixed–fixed beam and an F-structure
Modak et al. [84] Comparative study of RFM and IESM
Butkewitsch and Steffen Application of FEMU in design synthesis of a heavy truck side guard
[18]
Bais et al. [9] Application of iterative method for dynamic design of a drilling machine
Guo and Zhang [44] Development of response surface method (RSM) based FEMU technique
Kwon and Lin [63] Development of frequency selection method for FRF based FEMU
Hwang and Kim [51] Improvement of computational aspect of FEMU based damage detection technique
Teughels and Roeck [103] Application of FEMU in damage identification of highway bridge
Modak et al. [85] Application of constrained optimization based FEMU technique for dynamic design of fixed–fixed beam and
F-structure
Marwala [79] Development of particle swarm optimization based FEMU technique
Kwon and Lin [64] Development of Taguchi based FEMU technique
Vepsa et al. [105] Application of FEMU to a feed water pipeline of a nuclear power plant
Zhou and Farquhar [113] Application of FEMU in finding the in vivo material property of wheat stem
Mottershead et al. [86] Application of stochastic FEMU technique to a set of physical structures
Decouvreur et al. [29] Study of effect of dispersion error on acoustic FEMU
Jaishi and Ren [54] Modal flexibility residual based FEMU technique
Lin and Zhu [74] Development of model updating using base excitation (MUBE) method
Gondhalekar et al. [43] Application of FEMU for dynamic design of a complex automobile structure
Kim and Park [61] Development of automated parameter selection procedure for FEMU
Zapico et al. [112] Application of neural networks based FEMU to a steel frame
Jamshidi and Ashory [55] Comparative study of MUBE and RFM based FEMU techniques
Arora et al. [3] Development of damped FEMU technique by combining RFM with a damping identification method
Arora et al. [4] Development of damped FEMU technique by implementing RFM using complex parameters technique
Arora et al. [5] Comparative study of RFM based damped FEMU techniques
Arora et al. [6] Dynamic design of F-structure using complex parameters technique based RFM
Husain et al. [50] Application of FEMU to laser spot welds
Fang and Perera [35] Application of RSM based FEMU in damage analysis of a bridge
Dhandole and Modak [30] Application of direct method of Berman and Nagy for vibro-acoustic FEMU
Weber et al. [109] Consistent regularization of nonlinear FEMU in damage identification

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 529

Table 3 continued
References Contribution

Mthembu et al. [87] Application of particle swarm optimization in selection of updated FE models
Ren and Chen [96] Application of RSM based FEMU technique to a concrete bridge
Adhikari and Friswell [1] Development of spatially distributed FEMU technique using spatially correlated random fields
Dhandole and Modak ( [31] Comparative study of vibro-acoustic FEMU techniques
Chellini et al. [24] Application of FEMU in damage analysis of a steel–concrete composite frame
Arora [7] Comparative study of FEMU techniques
Lepoittevin and Kress [66] Use of FEMU for modal damping prediction
Khodaparast et al. [59] Development of interval FEMU technique using the Kriging predictor
Mottershead et al. [86] Application of sensitivity method for FEMU of helicopter airframe
Goller et al. [42] Application of stochastic FEMU for complex aerospace structures
Bayraktar et al. [12] Application of FEMU to arch dam systems
Khanmirza et al. [58] Application of FEMU to multi-storey shear buildings
Fang and Perera [36] Application of RSM based FEMU in damage identification of reinforced concrete frame
Dhandole and Modak [32] Acoustic FEMU using constrained optimization technique
Pradhan and Modak [93] Development of normal RFM
Pradhan and Modak [94] Application of RFM in damping matrix identification
Shahidi and Pakzad [98] Development of time domain results based model updating
Chakraborty and Sen [22] Development of moving least square method based model updating

damage analysis of rotor blades of helicopters, bridges, Existing FEMU techniques aim towards mapping only
multi-storey framed concrete buildings, reinforced concrete the dynamic behavior of the structure. More efforts are
frames, reinforced concrete beams, etc. It has also been required for developing such FEMU techniques which can
used successfully for non-destructive characterization of predict accurately the dynamic as well as static behavior of
elastic constants of composites, longitudinal moduli of structures.
living wheat, structural damping, etc. FEMU has also been More research is required in the development of realistic
used for dynamic design of automobiles, drilling machines, updated mass matrices by combining the FEMU techniques
spacers of nuclear power plants, F-structures, beams and with digital image processing and x-ray/ultrasonic im-
tennis racket etc. The flow chart shown in Fig. 12 also ages. 3-D digital image processing has been used for
represents the applications of FEMU in a number of mode-correlation purpose, but its use for objective function
disciplines. formulation still awaits more efforts.
Fuzzy logic, neural networks and their combination
have been used, for model formulation purpose, in many
such fields where system parameters are not understood
5 Research Issues in FEMU properly. FEMU is a similar problem where the elements of
large scale system matrices are not known exactly. Thus
There are conflicting opinions of various researchers re- research can also be directed towards the application of
garding use of objective functions and test structure used in such black-box based techniques in FEMU.
FEMU. Future research needs to be directed towards FEMU has been used to infer the in vivo material
standardizing a benchmark objective function and test properties of crops. Still, more research efforts are required
structure, so that these can be used worldwide by different in the direction of application of FEMU for better modeling
researchers for comparing the effectiveness of different and characterization of bones, flesh and plant leaves etc.
FEMU techniques.
It is also a well known fact that damping is present in all
real life structures. But still there are many model updating 6 Conclusions
techniques in which damping effects of a real life structure
have been neglected [8],69, 68, 96, 104, 80, 102]. Such In this paper a review of a number of direct and iterative
techniques need to be investigated so as to increase their techniques of FEMU has been presented. Direct techniques
ability to deal with damped real life systems. find little applications in industry due to the fact that

123
530 S. Sehgal, H. Kumar

Fig. 12 Applications of FEMU

updated FE models of most of the direct techniques are 3. Arora V, Singh SP, Kundra TK (2009) Finite element model up-
difficult to be understood on a physical basis. While a few dating with damping identification. J Sound Vib 324:1111–1123
4. Arora V, Singh SP, Kundra TK (2009) Damped model updating
physical property adjustments based direct techniques are using complex updating parameters. J Sound Vib 320:438–451
also now available in literature; their application to real life 5. Arora V, Singh SP, Kundra TK (2009) Comparative study of
systems needs to be explored further. On the other hand, damped FE model updating methods. Mech Syst Signal Process
iterative techniques have received considerable attention 23:2113–2129
6. Arora V, Singh SP, Kundra TK (2009) Damped FE model up-
from industrial application point of view and have been dating using complex updating parameters: its use for dynamic
applied successfully in a number of fields such as aircrafts, design. J Sound Vib 324:350–364
automobiles, satellites, machine tools and civil structures 7. Arora V (2011) Comparative study of finite element model
etc. Further a number of future research directions have updating methods. J Vib Control 17(13):2023–2029
8. Atalla MJ, Inman DJ (1998) On model updating using neural
been highlighted which can be used for further advance- networks. Mech Syst Signal Process 12(1):135–161
ments in the field of model updating. 9. Bais RS, Gupta AK, Nakra BC, Kundra TK (2004) Studies in
dynamic design of drilling machine using updated finite element
models. Mech Mach Theory 39:1307–1320
10. Baruch M (1978) Optimization procedure to correct stiffness
and flexibility matrices using vibration tests. Am Inst Aeronaut
References stronaut J 16(11):1208–1210
11. Baruch M, Bar-Itzhack IY (1978) Optimal weighted
1. Adhikari S, Friswell MI (2010) Distributed parameter model orthogonalization of measured modes. Am Inst Aeronaut stro-
updating using the Karhunen-Loeve expansion. Mech Syst naut J 16(4):346–351
Signal Process 24:326–339 12. Bayraktar A, Sevim B, Altunisik AC (2011) Finite element
2. Allemang RJ, Brown DL (1982) A correlation coefficient for model updating effects on nonlinear seismic response of arch
modal vector analysis. Proceedings of the first international dam–reservoir–foundation systems. Finite Elem Anal Des
modal analysis conference. Orlando, USA, pp 110–116 47:85–97

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 531

13. Beattie CA, Smith SW (1992) Optimal matrix approximants in 36. Fang SE, Perera R (2011) Damage identification by response
structural identification. J Optim Theory Appl 74(1):23–56 surface based model updating using D-optimal design. Mech
14. Berman A (1979) Mass matrix correction using an incomplete set Syst Signal Process 25:717–733
of measured modes. Am Inst Aeronaut stronaut J 17:1147–1148 37. Farhat C, Hemez FM (1993) Updating finite element dynamic
15. Berman A, Nagy EJ (1983) Improvement of a large analytical models using an element-by-element sensitivity methodology.
model using test data. Am Inst Aeronaut stronaut J 21:1168–1173 Am Inst Aeronaut stronaut J 31(9):1702–1711
16. Besterfield DH, Besterfield-Michna C, Besterfield GH, Bester- 38. Fissette E, Stavrinidis C, Ibrahim S (1988) Error location and
field-Sacre M (1995) Total quality management. Prentice-Hall, updating of analytical dynamic models using a force balance
Englewood Cliffs, NJ, USA method. Proceedings of the sixth international modal analysis
17. Bucher I, Braun S (1993) The structural modification inverse prob- conference. Orlando, USA, pp 1063–1070
lem: an exact solution. Mech Syst Signal Process 7(3):217–238 39. Friswell MI, Mottershead JE (1995) Finite element model up-
18. Butkewitsch S, Steffen VJ (2002) Shape optimization, model dating in structural dynamics. Kluwer Academic Publishers,
updating and empirical modeling applied to the design synthesis Dordrecht, The Netherlands
of a heavy truck side guard. Int J Solids Struct 39:4747–4771 40. Friswell MI, Garvey SD, Penny JET (1995) Model reduction
19. Caesar B (1986) Update and identification of dynamic mathe- using dynamic and iterated IRS techniques. J Sound Vib
matical models. Proceedings of the fourth international modal 186(2):311–323
analysis conference. Los Angeles, USA, pp 394–401 41. Friswell MI, Inman DJ, Pilkey DF (1998) The direct updating of
20. Caesar B (1987) Updating system matrices using modal test damping and stiffness matrices. Am Inst Aeronaut stronaut J
data, In: proceedings of the fifth international modal analysis 36(3):491–493
conference, London, UK, pp. 453–459 42. Goller B, Broggi M, Calvi A, Schueller GI (2011) A stochastic
21. Carvalho J, Datta BN, Gupta A, Lagadapati M (2007) A direct model updating technique for complex aerospace structures.
method for model updating with incomplete measured data and Finite Elem Anal Des 47:739–752
without spurious modes. Mech Syst Signal Process 43. Gondhalekar AC, Kundra TK, Modak SV (2007) Dynamic de-
21:2715–2731 sign of a complex automobile structure via finite element model
22. Chakraborty S, Sen A (2014) Adaptive response surface based updating. Proceedings of the 25th international modal analysis
efficient finite element model updating. Finite Elem Anal Des conference. Orlando, USA, pp 16–19
80:33–40 44. Guo Q, Zhang L (2004) Finite element model updating based on
23. Chang CC, Chang TYP, Xu YG (2000) Adaptive neural net- response surface methodology. Proceedings of the 22nd inter-
works for model updating of structures. Smart Mater Struct national modal analysis conference. Michigan, USA,
9:59–68 pp 1990–1997
24. Chellini G, Roeck GD, Nardini L, Salvatore W (2010) Damage 45. Guyan RJ (1965) Reduction of stiffness and mass matrices. Am
analysis of a steel-concrete composite frame by finite element Inst Aeronaut stronaut J 3(2):380–386
model updating. J Constr Steel Res 66:398–411 46. Hartog JPD (1934) Mechanical vibrations. McGraw-Hill Book
25. Chen JC, Garba JA (1980) Analytical model improvement using Company Inc, USA
modal test results. Am Inst Aeronaut stronaut J 18(6):684–690 47. He J, Ewins DJ (1991) Compatibility of measured and predicted
26. Chen JC, Kuo CP, Garba JA (1983) Direct structural parameter vibration modes in model improvement studies. Am Inst Aero-
identification by modal test results. Proceedings of the 24th naut stronaut J 29(5):798–803
structural dynamics and materials conference. California, USA, 48. Helfrick MN, Niezrecki C, Avitabile P, Schmidt T (2011) 3D
pp 44–49 digital image correlation methods for full-field vibration mea-
27. Collins JD, Hart GC, Hasselman TK, Kennedy B (1974) Sta- surement. Mech Syst Signal Process 25:917–927
tistical identification of structures. Am Inst Aeronaut stronaut J 49. Hu SLJ, Li H, Wang S (2007) Cross-model cross-mode method
12(2):185–190 for model updating. Mech Syst Signal Process 21:1690–1703
28. Cunha J, Piranda J (1999) Application of model updating 50. Husain NA, Khodaparast HH, Snaylam A, James S, Sharp M,
techniques in dynamics for the identification of elastic constants Dearden G, Ouyang H (2009) Modal testing and finite element
of composite materials. Compos B Eng 30:79–85 model updating of laser spot welds. J Phys Conf Ser 181:1–8
29. Decouvreur V, Bel ED, Bouillard P (2006) On the effect of the 51. Hwang HY, Kim C (2004) Damage detection in structures using
dispersion error when updating acoustic models. Comput a few frequency response measurements. J Sound Vib 270:1–14
Methods Appl Mech Eng 195:394–405 52. Imregun M, Ewins DJ, Hagiwara I, Ichikawa TA (1994) Com-
30. Dhandole SD, Modak SV (2009) Simulated studies in FE model parison of sensitivity and response function based updating
updating with application to vibro-acoustic analysis of the techniques. Proceedings of 12th international modal analysis
cavities, In: Proceedings of the third international conference on conference. Hawaii, USA, pp 1390–1400
integrity, reliability and failure, Porto/Portugal 53. Jacquelin E, Adhikari S, Friswell MI (2012) A second-moment
31. Dhandole SD, Modak SV (2010) A comparative study of approach for direct probabilistic model updating in structural
methodologies for vibro-acoustic FE model updating of cavities dynamics. Mech Syst Signal Process 29:262–283
using simulated data. Int J Mech Mater Des 6:27–43 54. Jaishi B, Ren WX (2006) Damage detection by finite element
32. Dhandole S, Modak SV (2012) A constrained optimization model updating using modal flexibility residual. J Sound Vib
based method for acoustic finite element model updating of 290:369–387
cavities using pressure response. Appl Math Model 36:399–413 55. Jamshidi E, Ashory MR (2009) Comparative study of RFM and
33. Ewins DJ (2000) Modal testing: theory, practice and application, model updating method using base excitation test data. Pro-
2nd edn. Research Studies Press Limited, England, UK ceedings of the 27th international modal analysis conference.
34. Fang H, Wang TJ, Chen X (2011) Model updating of lattice Orlando, USA, pp 2399–2409
structures: a substructure energy approach. Mech Syst Signal 56. Jiang J, Dai H, Yuan Y (2013) A symmetric generalized inverse
Process 25(5):1469–1484 eigenvalue problem in structural dynamics model updating.
35. Fang SE, Perera R (2009) A response surface methodology Linear Algebra Appl 439:1350–1363
based damage identification technique. Smart Mater Struct 57. Kabe AM (1985) Stiffness matrix adjustment using mode data.
18:1–14 Am Inst Aeronaut stronaut J 23(9):1431–1436

123
532 S. Sehgal, H. Kumar

58. Khanmirza E, Khaji N, Majd VJ (2011) Model updating of 81. Modak SV (2014) Direct matrix updating of vibroacoustic finite
multistory shear buildings for simultaneous identification of element models using modal test data. AIAA J 52(7):1–7
mass, stiffness and damping matrices using two different soft- 82. Modak SV, Kundra TK, Nakra BC (2000) Model updating using
computing methods. Expert Syst Appl 38:5320–5329 constrained optimization. Mech Res Commun 27(5):543–551
59. Khodaparast HH, Mottershead JE, Badcock KJ (2011) Interval 83. Modak SV, Kundra TK, Nakra BC (2002a) Prediction of
model updating with irreducible uncertainty using the Kriging dynamic characteristics using updated finite element models. J
predictor. Mech Syst Signal Process 25:1204–1226 Sound Vib 254(3):447–467
60. Kidder RL (1973) Reduction of structural frequency equations. 84. Modak SV, Kundra TK, Nakra BC (2002b) Comparative study
Am Inst Aeronaut stronaut J 11(6):892–902 of model updating methods using simulated experimental data.
61. Kim GH, Park YS (2008) An automated parameter selection Comput Struct 80:437–447
procedure for finite-element model updating and its applications. 85. Modak SV, Kundra TK, Nakra BC (2005) Studies in dynamic
J Sound Vib 309:778–793 design using updated models. J Sound Vib 281:943–964
62. Kim KO, Anderson WJ, Sandstrom RE (1983) Non-linear in- 86. Mottershead JE, Link M, Friswell MI (2011) The sensitivity
verse perturbation method in dynamic analysis. Am Inst Aero- method in finite element model updating: a tutorial. Mech Syst
naut stronaut J 21(9):1310–1316 Signal Process 25:2275–2296. doi:10.1016/j.ymssp.2010.10.012
63. Kwon KS, Lin RM (2004) Frequency selection method for FRF- 87. Mthembu L, Marwala T, Friswell MI, Adhikari S (2010) Finite
based model updating. J Sound Vib 278:285–306 element model selection using particle swarm optimization.
64. Kwon KS, Lin RM (2005) Robust finite element model updating Proceedings of the 28th international modal analysis conference.
using Taguchi method. J Sound Vib 280:77–99 Florida, USA, pp 41–52
65. Lammens S, Heylen W, Sas P (1992) Model updating and 88. O’Challan JC (1989) A procedure for an improved reduced
structural optimization of a tennis racket. Mech Syst Signal system (IRS) model. Proceedings of the seventh international
Process 6(5):461–468 modal analysis conference. Las Vegas, USA, pp 17–21
66. Lepoittevin G, Kress G (2011) Finite element model updating of 89. O’Challan JC, Avitabile P, Riemer R (1989) System equivalent
vibrating structures under free-free boundary conditions for reduction and expansion processes. Proceedings of the seventh
modal damping prediction. Mech Syst Signal Process 25: international modal analysis conference. Las Vegas, USA,
2203–2218 pp 29–37
67. Levin RI, Lieven NAJ (1998) Dynamic finite element model 90. O’Challan JC, Liew IW, Avitabile P, Madden R (1986) An
updating using neural networks. J Sound Vib 210(5):593–607 efficient method of determining rotational degrees of freedom
68. Levin RI, Lieven NAJ (1998) Dynamic finite element model from analytical and experimental modal data. Proceedings of the
updating using simulated annealing and genetic algorithms. fourth international modal analysis conference. Los Angeles,
Mech Syst Signal Process 12(1):91–120 USA, pp 50–58
69. Li WL (2002) A new method for structural model updating and joint 91. Petyt M (1998) Introduction to finite element vibration analysis.
stiffness identification. Mech Syst Signal Process 16(1):155–167 Cambridge University Press, Cambridge, UK
70. Lieven NAJ, Ewins DJ (1988) Spatial correlation of mode 92. Pilkey DF (1998) Computation of damping matrix for finite
shapes, the coordinate modal assurance criterion (COMAC). element model updating, Ph.D. thesis, Virginia Polytechnic In-
Proceedings of the sixth international modal analysis confer- stitute and State University, Virginia
ence. Orlando, USA, pp 690–695 93. Pradhan S, Modak V (2012) Normal response function method
71. Lieven NAJ, Ewins DJ (1990) Expansion of modal data for for mass and stiffness matrix updating using complex FRFs.
correlation. Proceedings of the eighth international modal ana- Mech Syst Signal Process 32:232–250
lysis conference. Orlando, USA, pp 605–609 94. Pradhan S, Modak V (2012) A method for damping matrix
72. Lieven NAJ, Waters TP (1994) Error location using normalized identification using frequency response data. Mech Syst Signal
cross orthogonality. Proceedings of the 12th international modal Process 33:69–82
analysis conference. Honolulu, Hawaii, pp 761–764 95. Ramamurti V, Rao MH (1998) Studies on model updating using
73. Lim TW (1990) Submatrix approach to stiffness matrix cor- different finite elements, the editor. J Sound Vib 211(3):507–512
rection using modal test data. Am Inst Aeronaut stronaut J 96. Ren WX, Chen HB (2010) Finite element model updating in
28(6):1123–1130 structural dynamics by using the response surface method. Eng
74. Lin RM, Zhu J (2007) Finite element model updating using Struct 32:2455–2465
vibration test data under base excitation. J Sound Vib 97. Sastry CVS, Mahapatra DR, Gopalakrishnan S, Ramamurthy TS
303:596–613 (2003) An iterative system equivalent reduction expansion
75. Lin RM, Lim MK, Du H (1995) Improved inverse eigensensi- process for extraction of high frequency response from reduced
tivity method for structural analytical model updating. Trans order finite element model. Comput Methods Appl Mech Eng
ASME J Vib Acoust 117(2):192–198 192:1821–1840
76. Lin RM, Ewins DJ (1990) Model updating using FRF data. 98. Shahidi S, Pakzad S (2014) Generalized response surface model
Proceedings of 15th international modal analysis seminar. KU updating using time domain data, ASCE Journal of Structural
Leuven, Belgium, pp 141–162 Engineering 140(8)
77. Maia NMM, Silva JM (1997) Theoretical and experimental 99. Sidhu J, Ewins DJ (1984) Correlation of finite element and
modal analysis. Research Studies Press Limited, England, UK modal test studies of a practical structure. Proceedings of the
78. Mares C, Mottershead JE, Friswell MI (2006) Stochastic model second international modal analysis conference. Orlando, USA,
updating: part 1- theory and simulated example. Mech Syst pp 756–762
Signal Process 20:1674–1695 100. Sinha JK, Mujumdar PM, Moorthy RIKA (2000) parameter
79. Marwala T (2005) Finite element model updating using particle identification technique for detection of spacer locations in an
swarm optimization. Int J Eng Simul 6(2):25–30 assembly of two coaxial flexible tubes. Nucl Eng Des
80. Marwala T, Sibisi S (2005) Finite element model updating using 196:139–151
Bayesian framework and modal properties. J Aircr 42(1): 101. Smith SW, Beattie CA (1991) Secant-method adjustment for
275–278 structural models. Am Inst Aeronaut stronaut J 29(1):119–126

123
Structural Dynamic Model Updating Techniques: A State of the Art Review 533

102. Steenackers G, Devriendt C, Guillaume P (2007) On the use of 108. Waters TP (1998) A modified force balance method for locating
transmissibility measurements for finite element model updat- errors in dynamic finite element models. Mech Syst Signal
ing. J Sound Vib 303:707–722 Process 12(2):309–317
103. Teughels A, Roeck GD (2004) Structural damage identification 109. Weber B, Paultre P, Proulx J (2009) Consistent regularization of
of the highway bridge Z24 by FE model updating. J Sound Vib nonlinear model updating for damage identification. Mech Syst
278:589–610 Signal Process 23:1965–1985
104. Tu Z, Lu Y (2008) FE model updating using artificial boundary 110. Williams EJ, Green JSA (1990) spatial curve fitting technique
conditions with genetic algorithms. Comput Struct 86:714–727 for estimating rotational degrees of freedom. Proceedings of the
105. Vepsa A, Haapaniemi H, Luukkanen P, Nurkkala P, Saaren- eighth international modal analysis conference. Orlando, USA,
heimo A (2005) Application of finite element model updating to pp 376–381
a feed water pipeline of a nuclear power plant. Nucl Eng Des 111. Xiamin Z (1999) A best matrix approximation method for up-
235:1849–1865 dating the analytical model. J Sound Vib 223(5):759–774
106. Wahab MMA, Roeck GD, Peeters B (1999) Parameterization of 112. Zapico JL, Buelga AG, Gonzalez MP, Alonso R (2008) Finite
damage in reinforced concrete structures using model updating. element model updating of a small steel frame using neural
J Sound Vib 228(4):717–730 networks. Smart Mater Struct 17:1–11
107. Wang W, Mottershead JE, Mares C (2009) Mode-shape recog- 113. Zhou J, Farquhar T (2005) Wheat stem moduli in vivo via ref-
nition and finite element model updating using the Zernike erence basis model updating. J Sound Vib 285:1109–1122
moment descriptor. Mech Syst Signal Process 23:2088–2112

123