Professional Documents
Culture Documents
Finite-Element Modelling and Updating of Laser Spot Weld Joints in A Top-Hat Structure For Dynamic Analysis
Finite-Element Modelling and Updating of Laser Spot Weld Joints in A Top-Hat Structure For Dynamic Analysis
net/publication/258176714
Article in ARCHIVE Proceedings of the Institution of Mechanical Engineers Part C Journal of Mechanical Engineering Science 1989-1996 (vols 203-210) · April 2010
DOI: 10.1243/09544062JMES1787
CITATIONS READS
35 263
6 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Huajiang Ouyang on 03 June 2014.
Finite-element modelling and updating of laser spot weld joints in a top-hat structure for dynamic
analysis
N A Husain, H H Khodaparast, A Snaylam, S James, G Dearden and H Ouyang
Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science 2010 224:
851
DOI: 10.1243/09544062JMES1787
Published by:
http://www.sagepublications.com
On behalf of:
Additional services and information for Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering
Science can be found at:
Subscriptions: http://pic.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://pic.sagepub.com/content/224/4/851.refs.html
What is This?
The manuscript was received on 16 June 2009 and was accepted after revision for publication on 2 October 2009.
DOI: 10.1243/09544062JMES1787
Abstract: Spot welds made by resistance spot welding are used extensively in automotive engi-
neering. However, owing to increasing demands in the use of advanced and lightweight materials,
laser welding has become a popular alternative for producing spot welds. Because of the complex-
ity and uncertainties of laser welds and thus formed structures, the finite-element (FE) modelling
of the welds for dynamic analysis is a research issue. This article first outlines some of the exist-
ing modelling works of spot welds. Then, a hat-plate structure used for this study is described
and its FE representations are explained. The welds are modelled using CWELD elements in
MSC/NASTRAN and their feasibility for representing laser spot welds is investigated. Numerical
results for the initial FE model differ considerably from that of their experimental counterparts;
hence, a model updating procedure is carried out to minimize the discrepancy between the two
sets of results. In this work, the updating is posed as an optimization problem and is performed
using the structural optimization capability (SOL 200) in MSC/NASTRAN. Two stages of updating
are conducted, that is (a) updating FE models of individual components and (b) updating an
FE model of the welded structure. Crucial steps in updating are explained. It is found that by
selecting the right updating parameters, the CWELD element can be used to represent laser spot
welds with good accuracy.
Keywords: finite-element modelling, model updating, modal testing, laser welding, spot weld
joints, CWELD
complex issue. This is mainly because of the existence of the bolted joint connections and included several
of many local effects that are not taken into account by examples of static and dynamic friction models.
FE modelling when predicting frequencies and modes. Another highly recommended review paper on bolted
Obviously, there are many works done on weld mod- joints by Ibrahim and Pettit [10] focused on prob-
elling. However, it appears that most of the studies lems relating to the dynamic behaviour of structures
have concentrated on spot welds made by the RSW with bolted joints, including 516 references. Joint
[2–4]. Modelling works related to welds made by LW uncertainties and relaxation were discussed and iden-
have focused mainly on simulating the welding pro- tification of (linear and non-linear) joint properties
cess itself [5–7] and only a small number of works have was a major topic of that paper. Design issues of fully
been done on fatigue estimation of laser welds, such as and partially restrained joints, sensitivity analysis to
reference [8]. To the authors’ best knowledge, there is variations of joint parameters, and fatigue prediction
no reported work on modelling the dynamic behaviour for metallic and composite joints were also covered.
of structures with laser welds. Section 2 of this arti- Oldfield et al. [11] investigated a bolted joint under
cle summarizes several modelling works on commonly harmonic loading by using a detailed ABAQUS FE
used joints, i.e. bolted and welded joints. model to obtain the necessary dynamic information
In this study, a set of nine nominally identical of the joint and then utilized Jenkins elements and a
hat-plate structures are investigated. Each structure Bouc–Wen model to simplify the bolted-joint model.
contains 20 spot welds made by LW along their It was demonstrated that the dynamic properties of
flanges. Development of the FE model of the structures the joint were obtained successfully by the simplified
involves three stages: models with a greatly reduced computational effort.
Kim et al. [12] studied modelling of a structure with
(a) initial FE modelling and manual tuning;
bolted joints using four kinds of FE models
(b) modal testing;
(c) model updating. (a) a solid bolt model;
(b) a coupled bolt model;
The structure of interest and the development of the (c) a spider bolt model;
initial FE models for its components are described in (d) a no-bolt model.
section 3. Modal testing is explained in section 4 and
this is followed by the FE model updating in section 5. It was found that the relatively simpler FE models (i.e.
Section 6 discusses the modelling and updating works the coupled and spider models) were more effective,
of the welded structure, including some explana- being more than 49 per cent faster and used approxi-
tions of the CWELD element and its application in mately only 20 per cent of the memory compared with
modelling the welded structure. the solid bolt model.
In contrast to bolted joints, weld joints are perma-
nent and spot welds are very commonly used in joining
2 MODELLING OF JOINTS metal sheet-like structures. However, like bolted joints,
reliable evaluations of the behaviour of the welds
Modelling of joints is certainly a subject of special con- are always a concern, since many factors such as
cern and many models have been developed in recent geometrical irregularities, residual stresses, material
years. There are many types of joints used in fastening inhomogeneity, and defects in the welds are difficult
parts together, with bolted and welded joints being the to be incorporated in modelling [13]. A number of
most commonly used ones in structural design. Bolted works on the modelling of spot welds have been pub-
joints are popular, largely because they can be easily lished (for example, see references [4] and [14] to [17])
disassembled, maintained, and/or inspected. How- and most of these were discussed by Palmonella et al.
ever, this type of joints has many complexities (such [18]. They grouped the spot weld models into two cat-
as non-linear frictional behaviour, pretension effect, egories: (a) models for limit capacity analysis, such as
etc.) that are difficult to simulate in FE modelling. references [4] and [14], and (b) models for dynamic
As the behaviour of bolted joints plays a signifi- simulation, as in reference [17]. The first category
cant role in the dynamic characteristics of structures, of models requires a very detailed mesh in order to
the need for developing accurate predictive models of work out a smooth stress field within and around
the joints is very demanding. Several authors, such as a spot weld. Normally, brick elements are used to
Gaul and Nitsche [9], Ibrahim and Pettit [10], Oldfield model welds, while brick or shell elements are used
et al. [11], and Kim et al. [12], covered the modelling to model welded plates. This would need tremendous
work associated with bolted joints. The role of fric- computational effort, which is impractical if there are
tion in bolted joint connections and modelling issues thousands of spot welds.
associated with the frictional behaviour of the joints Deng et al. [4] addressed the underlying three-
was discussed thoroughly by Gaul and Nitsche [9], dimensional (3D) features of the stress field in spot
who cited 134 references. The article addressed some welds. Very detailed FE representations of the spot
approaches for modelling the non-linear behaviour weld nuggets and the surrounding areas were made,
with thousands of solid elements being used in the for model updating. So far, CWELD has been widely
modelling. A fine mesh was employed within the weld used in modelling spot welds produced by RSW and
nugget, whereas a coarse mesh was applied elsewhere. it is uncertain whether the element can well represent
One of the findings was that stress concentration spot welds made by other processes, especially by LW,
appeared near the nugget boundary while the centre of as the construction of the joint in LW is very different
the nugget was mostly stress-free, which indicates that from that in the conventional RSW.
the centre of the nugget does not contribute much to The performance of simplified FE models (i.e.
the load-bearing capability of the spot weld. A simpli- single-bar, spoke-bar, and multiple rigid-bar models)
fied version of this model was presented by Chen and against a detailed 3D (or solid) FE model for rep-
Deng [14]. The solid elements were replaced by the resenting spot welds subjected to five basic loading
shell elements and the performance of the shell ele- situations (i.e. tension, out-of-plane torsion, out-of-
ments in modelling the spot welds was investigated. plane bending, in-plane torsion, and in-plane shear)
In that work, the weld nugget was assumed to be rigid, was studied by Xu and Deng [3]. The simplified models
hence not modelled explicitly, unlike in the 3D solid were implemented with several mesh refinements and
model. Apart from the nugget, the rest of the model their performances were evaluated against the con-
adopted the same mesh as the detailed model. The verged 3D FE model. The article showed that the sim-
study demonstrated that the shell element model was plified models can be quite accurate in stiffness simu-
able to estimate the stress field surrounding the spot lation when subjected to tension, out-of plane torsion,
weld regions. However, a very refined mesh must be and bending. However, the models were not reliable
applied in order to achieve a very good approxima- when they were under in-plane torsion and shear.
tion, which again is not always practical when there Although FE modelling of conventional spot welds
are thousands of welds. has been carried out by a number of researchers, there
On the other hand, the second category of models has been no work of modelling the spot welds made by
uses very coarse meshes and allows the application of LW for structural dynamic analysis, to the authors’ best
much simpler models with significantly fewer degrees knowledge. It is still uncertain whether the existing
of freedom (DOFs). This type of models simulates the spot weld models can be employed for modelling the
stiffness characteristics of the spot welds and their laser spot welds. This is mainly because the construc-
contributions towards the behaviour of the structure. tion of the laser spot welds is dissimilar to that of the
Salvini et al. [16] developed an FE model of spot welds spot welds made by RSW. The aim of this article there-
that took into consideration the structural behaviour fore is to investigate the feasibility of utilizing one of
of the regions surrounding the welds, which are nor- the mostly used elements for spot weld modelling (i.e.
mally not accounted for in the search for simpler spot CWELD) for the dynamics prediction of the spot welds
weld models. The model was composed of a number made by LW. The CWELD element and its application
of 3D beam elements oriented in the radial direction in this work are described further in section 6.1.
and was so formulated such that it could be easily
employed in complex structures with many spot welds.
The authors also demonstrated an important increase 3 HAT-PLATE COMPONENTS AND FE
in accuracy in their developed model. REPRESENTATIONS
Another spot weld model was proposed by Heiserer
et al. [15]. The model, known as Area Contact Model 2 A hat-plate structure shown in Fig. 1 is used in this
(ACM2), was developed using a brick (i.e. HEXA) ele- work. The structure, which consists of a flat plate and
ment in MSC/NASTRAN and the connections to the a formed hat-like shell (or ‘top-hat’) joined together
joined surfaces were via rigid (i.e. RBE3) elements, also by spot welds at the flanges, is designed to represent
available in MSC/NASTRAN. ACM2 is a coarse spot common structures used in the construction of a car
weld model and it allows non-congruent meshes to be BIW. The spot welds, which are produced by LW, are
employed. This model can be employed for both limit 5 mm in diameter and 60 mm apart in the longitudinal
capacity and dynamics analysis, and also can be used
for subsequent analysis such as model updating.
Alternatively, the CWELD element developed by
MSC/NASTRAN can be employed for the predic-
tion of dynamics behaviour, which is presented in
MSC/NASTRAN documentations and explained in
detail in reference [17]. This element also offers the
prospect of generating weld joints that are indepen-
dent of the mesh, which allows a big reduction in
the computational effort and the model development
time. Like ACM2, CWELD allows the physical proper-
ties of the weld to be considered, which is important Fig. 1 The hat-plate structure
Fig. 3 FE models of (a) the flat plate and (b) the top-hat
Table 1 Experimental, initial, and updated natural frequencies for flat plate and top-hat components
I II III IV V VI VII
Component Mode Experiment (Hz) Initial (Hz) Error (%) = |(II–I)/I| MAC Updated (Hz) Error (%) = |(V–I)/I| MAC
Fig. 4 Test set-up for (a) the flat plates and (b) the top-hats
[19, 20]. Modal testing with the free–free boundary [26–28], and the optimization algorithm (SOL 200) of
condition was conducted on each specimen from both the FE code NASTRAN is used to perform the updating.
sets of components. Figure 4 shows the experimental
set-up for (a) the flat plates and (b) the top-hats. The
flat plates were tested using one hammer point and 5.1 Formulation of model updating procedure
two measurement points (Fig. 4(a)), while the top- The equation of motion for undamped free vibration
hats were tested using one hammer point and five of a structure can be expressed as
measurement points (Fig. 4(b)), with only the first five
modes measured from both sets of tests. The locations (−λM + K)u = 0 (1)
of the impact and measurement points were carefully
chosen such that they are not located near any nodal where M and K are the n × n mass and stiffness
points. Furthermore, the mass of each accelerometer matrices of the structure, and u is the n×1 modal dis-
was only 1.6 g, while the mass of the flat plate and placement vector (with n being the number of DOFs
the top-hat was 0.7 and 1.1 kg, respectively. Hence, the of the whole structure). λ = ω2 is the eigenvalue and ω
effect of mass loading was very minimal. is the natural frequency of the structure.
The vibration responses from both tests were mea- When the vector θ of structural parameters (such
sured using a 12-channel LMS data acquisition system. as the Young’s modulus) is changed, the vector λ of
Natural frequencies and mode shapes were extracted m eigenvalues normally changes as well. Approxima-
from the measured FRF by using a PolyMAX curve- tions used in NASTRAN are based on simple first-order
fitting procedure from the LMS Test.Lab system. The Taylor series expansion and the general form of this
average experimental natural frequencies for both expansion for λ is
components are computed and tabulated in Table 1
(column I). For the flat plate, discrepancies of the pre- λi+1 = λi + Si (δθ) (2)
dicted natural frequencies from the measured data are
<1 per cent in bending modes (i.e. modes 1, 2, and In equation (2), Si is an m × n sensitivity matrix at ith
4) but higher errors in torsion modes (i.e. modes 3 iteration, which denotes the rates of change of the
and 5) are obtained. On the other hand, the errors structural eigenvalues, λi , with respect to changes in
of the top-hat are considerably high (with the maxi- parameters, δθ, which can be expressed as [21]
mum of approximately 4.9 per cent) for the first three
modes but reasonably good (with maximum error of ∂λi ∂K ∂M
1.2 per cent) for the two remaining modes, as can Si = = u Ti − λi ui (3)
∂θ ∂θ ∂θ
be seen in Table 1 (column III). Hence, FE model
updating has to be carried out in order to identify It should be noted that any modifications made to the
the uncertain parameters and therefore improve the system parameters could affect the modal properties
results. of the system. Therefore, the parameters and modal
properties involved in the updating process must be
selected properly. Selection of the right parameters
5 UPDATING OF THE MODELS and modal properties for the updating procedure are
briefly explained in the next sections.
Model updating is a viable approach to improve the An objective function based on residuals between
correlation between FE models and the experimental the experimental modal data (e.g. natural frequencies,
data by minimizing the differences between results mode shapes, etc.) and their predictions is set for min-
from the two approaches [21–25]. In this article, imization in the updating procedure. The procedure
updating is cast as a structural optimization problem continues until convergence is accomplished when
Table 2 Updating results for the flat plate and the top-hat
I II
Component Parameter Initial value Updated value Changes (%) = |(II–I)/I|
the difference between values of the objective func- 81 GPa, respectively. The Young’s modulus is allowed
tion J from consecutive iterations is sufficiently small. to vary from 185 to 220 GPa, while the shear modu-
In this work, the objective function is constructed on lus is given only a small variation from 80 to 84 GPa.
the basis of eigenvalue residuals, given by The thicknesses of the top-hat are given the same ini-
2 tial values of 1.45 mm, with lower and upper bounds
n
λj of 1.20 and 1.60 mm, respectively. During updating,
J= exp − 1 (4) the parameters are normalized in the form of θ /θ0
j=1
λj
so that the initial normalized parameter values are
exp unity. Hence, the sensitivities become (δλ/δθ )θ 0 , where
where λj is the jth experimental eigenvalue and λj
δ denotes a small increment.
is the jth eigenvalue predicted by the FE model. It is
important to note that equation (4) only holds if the
measured eigenvalue and its predicted counterpart
5.3 Updating results
are paired correctly, and therefore it is vital to ensure
that the experimental and numerical data relate to the Updating is done by minimizing the objective
same mode. In addition, it is generally preferable to function, as explained in section 5.1, and performed
use a larger number of experimental modal proper- on the basis of the first five measured frequencies.
ties in the updating process. Obviously, this would be Table 1 (column V) shows the updating results for
more difficult, but the updated model would be more the flat plate and the top-hat, while Table 2 shows
predictive than by using only a few modal data. the changes in updating parameters from the initial
values. Figures 5 and 6 show the changes of the updat-
ing parameters for both components from the initial
5.2 Parameter selection
normalized values of unity to convergence.
Selecting the updating parameters is an important Convergence is obtained very quickly for both com-
aspect of the FE model updating process [21, 23, ponents and the updated natural frequencies are
26] and any parameters for material and geometric significantly better, especially for the top-hat where
properties (such as area, inertia, thickness, diame- errors of less than 1.5 per cent are achieved. All the
ter, density, Young’s modulus, etc.) can be considered. natural frequencies of the flat plate are improved and
However, the parameters selected should be justi- the errors for the torsion modes are reduced by more
fied by engineering understanding of the structure than 1 per cent. Modal assurance criterion (MAC) anal-
[25, 29–31] and the number of parameters should be ysis was performed on both components based on the
kept to a minimum to avoid ill-conditioning problems measured points mentioned in the previous section.
[26]. Therefore, it is necessary to compute the eigen-
value sensitivities beforehand, so that only the most
influential parameters can be chosen.
Having done the sensitivity analysis using SOL 200,
two parameters are selected for the flat plate and four
parameters are chosen for the top-hat, as follows:
(a) the Young’s modulus of the flat plate, Eflatplate ;
(b) the shear modulus of the flat plate, Gflatplate ;
(c) the thickness of the folds for the top-hat, t1 ;
(d) the thickness of the flanges for the top-hat, t2 ;
(e) the thickness of the sidewalls for the top-hat, t3 ;
(f) the thickness of the top for the top-hat, t4 .
The initial values of the Young’s modulus and the Fig. 5 Parameter changes from the initial values of unity
shear modulus of the flat plate are set to 210 and for the flat plate
I II III IV V VI VII
Mode Experiment (Hz) Initial (Hz) Error (%) = |(II–I)/I| MAC Updated (Hz) Error (%) = |(V–I)/I| MAC
I II
Parameter Initial value Updated value Changes (%) = |(II–I)/I|
6.4.1 Updating parameters Therefore, several parameters from the spot welds and
the patch of the welds (such as the diameter, thick-
There are uncertainties in the material properties of nesses, Young’s moduli, densities, and Poisson’s ratios)
the spot welds and the surrounding areas (or patch). are investigated by carrying out the sensitivity analy-
sis. The patch parameters are included because it is
found that the weld parameters alone cannot success-
fully improve the results of the initial model. Of all the
possible parameters, the weld diameter dweld and the
Young’s moduli of the weld and the patch, Eweld and
Epatch , are selected for the updating process.
For the updating procedure, each of the three
parameters has an initial value, as shown in Table 4
(column I) and they are allowed to have some vari-
ations depending on the uncertainty level associated
Fig. 10 Parameter changes from the initial values of with the parameters. The values of the weld parame-
unity for the complete model ters (i.e. dweld and Eweld ) are allowed to vary in a limited
reasonable range only, while the Young’s modulus of Most importantly, it is essential to have an appro-
the patch is allowed to have a very big variation due to priate mesh in the FE model since it will influence
high uncertainties of the patch properties. As men- the size of the patch used in the FE model. In this
tioned in section 6.1, a higher initial value of Epatch work, the patch size is set to be 20 per cent bigger
is assigned to justify the rigidness of the patch in than the nominal diameter of the weld and, based on
comparison to the adjacent bulk materials. the updating results, the Young’s modulus of the patch
should be three times the value of the Young’s modu-
6.4.2 Model updating and its results lus of the weld. These values are recommended for the
dynamic analysis of structures of similar constructions
Updating is carried out on the basis of the first five with many laser spot welds when the CWELD element
measured frequencies from the welded samples. The is used.
natural frequencies and the MAC values of the updated
model are presented in Table 3 (column V and VII),
while the updated values of the parameters are shown ACKNOWLEDGEMENT
in Table 4 (column II). After the updating procedure, it
is found that all the natural frequencies are improved Discussion with Dr Zongjun Lu of Jaguar and Land
significantly and the MAC value for mode 4 is much Rover is gratefully acknowledged. Nurulakmar Abu
better. The updating converges after three iterations, Husain would like to thank the Malaysian Min-
as depicted in Fig. 10, and the mode shapes are shown istry of Higher Education (MOHE) and the Uni-
in Fig. 11. versiti Teknologi Malaysia (UTM) for their support.
The results achieved show that the CWELD param- Hamed Haddad Khodaparast wishes to acknowledge
eters (i.e. dweld and Eweld ) are well within reasonable the support of EC Marie Curie Excellence Project
limits. Attention is given to the Young’s modulus of the ECERTA under the contract number MEXT-CT-2006-
patch (i.e. Epatch ) which is considered as the parameter 042383.
that most affects the behaviour of the model. Based
on the updating results, it is found that Epatch should © Authors 2010
be approximately three times the Young’s modulus of
the weld.
REFERENCES
11 Oldfield, M., Ouyang, H., and Mottershead, J. E. Sim- 27 Jaishi, B. and Ren, W. Finite element model updat-
plified models of bolted joints under harmonic loading. ing based on eigenvalue and strain energy residuals
Comput. Struct., 2005, 84, 25–33. using multiobjective optimisation technique. Mech. Syst.
12 Kim, J., Yoon, J.-C., and Kang, B.-S. Finite element anal- Signal Process., 2007, 21, 2295–2317.
ysis and modelling of structure with bolted joints. Appl. 28 Duan, Z., Liu,Y., and Spencer, B. F. Finite element model
Math. Modell., 2007, 31, 895–911. updating of structures using a hybrid optimization tech-
13 Mottershead, J. E., Mares, C., James, S., and Friswell, M. nique. Proc. SPIE, 2005, 5765, 335–344.
I. Stochastic model updating: part 2 – application to a set 29 Mottershead, J. E., Mares, C., Friswell, M. I., and James,
of physical structures. Mech. Syst. Signal Process., 2006, S. Selection and updating of parameters for an alu-
20, 2171–2185. minium space-frame model. Mech. Syst. Signal Process.,
14 Chen, W. and Deng, X. Performance of shell elements 2000, 14(6), 923–944.
in modelling spot-welded joints. Finite Elem. Anal. Des., 30 Ahmadian, H., Gladwell, G. M. L., and Ismail, F. Parame-
2000, 35, 41–57. ter selection strategies in finite element model updating.
15 Heiserer, D., Chargin, M., and Sielaff, J. High perfor- J. Vibr. Acoust., 1997, 119(1), 37–45.
mance, process oriented, weld spot approach. In Pro- 31 Gladwell, G. M. L. and Ahmadian, H. Generic element
ceedings of the First MSC Worldwide Automotive User matrices suitable for finite element model updating.
Conference, Munich, Germany, 1999. Mech. Syst. Signal Process., 1995, 9(6), 601–614.
16 Salvini, P., Vivio, F., and Vullo, V. A spot weld finite ele-
ment for structural modelling. Int. J. Fatigue, 2000, 22,
645–656. APPENDIX
17 Fang, J., Hoff, C., Holman, B., Mueller, F., and Waller-
stein, D. Weld modelling with MSC.Nastran. In Pro-
ceedings of the Second MSC Worldwide Automotive Notation
Conference, MSC Software Corporation, 2000.
18 Palmonella, M., Friswell, M. I., Mottershead, J. E., and
dweld diameter of the spot weld
Lees, A. W. Finite element models of spot welds in struc- E Young’s modulus
tural dynamics: review and updating. Comput. Struct., Eflatplate Young’s modulus of the flat plate
2005, 83, 648–661. Epatch Young’s modulus of the patch
19 Ewins, D. J. Modal testing: theory, practice and appli- Eweld Young’s modulus of the weld
cation, 2nd edition, 2000 (Research Studies Press i number of iterations
Ltd, Taunton, UK). J objective function
20 Avitabile, P. Experimental modal analysis (a sim- K stiffness matrix
ple non-mathematical presentation). Sound Vibr., M mass matrix
2001, 1–15. S sensitivity matrix
21 Friswell, M. I. and Mottershead, J. E. Finite element
t1 thickness of folds
model updating in structural dynamics, 1995 (Kluwer
Academic Publishers, Dordrecht, The Netherlands).
t2 thickness of flanges
22 Friswell, M. I., Inman, D. J., and Pilkey, D. F. The direct t3 thickness of sidewalls
updating of damping and stiffness matrices. AIAA J., t4 thickness of top
1998, 36(2), 491–493. u modal displacement vector
23 Mottershead, J. E. and Friswell, M. I. Model updating
in structural dynamics: a survey. J. Sound Vibr., 1993, δ infinitesimal change
167(2), 347–375. θ structural parameter
24 Kenigsbuch, R. and Halevi, Y. Model updating in struc- θ vector of structural parameters
tural dynamics: a generalised reference basis approach. θ0 initial value of the structural parameter
Mech. Syst. Signal Process., 1998, 12(1), 75–90. λ structural eigenvalues
25 Mares, C., Friswell, M. I., and Mottershead, J. E. Model
λ vector of structural eigenvalues
updating using robust estimation. Mech. Syst. Signal
Process., 2002, 16(1), 169–183.
λexp experimental eigenvalues
26 Kim, G. H. and Park, Y. S. An improved updating param- ρ density
eter selection method and finite element model update ρflatplate density of the flat plate
using multiobjective optimisation technique. Mech. Syst. ν Poisson’s ratio
Signal Process., 2004, 18(1), 59–78. ω natural frequency