06PP192

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Modelling of a vehicle windshield with realistic uncertainty

R. d’Ippolito1 , U. Tabak2 , M. De Munck2 , S. Donders1 , D. Moens2 , D. Vandepitte2


1 LMS International, CAE Division, Interleuvenlaan 68, 3001 Leuven, Belgium
2 K.U.Leuven, Department of Mechanical Engineering, Division PMA,

Celestijnenlaan 300 B, B-3001, Heverlee, Belgium


e-mail: roberto.dippolito@lms.be, umut.tabak@mech.kuleuven.be

Abstract
In automotive industry, design decisions are increasingly based on virtual prototypes, such as numerical
Finite Element (FE) models. This reduces the need for expensive physical prototypes, so that design cost
and time-to-market can be reduced. These FE models are deterministic, implicitly assuming that all design
parameters are precisely known and that the manufacturing process produces identical structures. This is
typically not valid. Uncertainty exists in the manufacturing process, in the physical and geometrical proper-
ties and in the environmental conditions. A single deterministic analysis is not sufficient to fully understand
their effect on the structural and vibro-acoustic performance. Including possibilistic models of design vari-
ables into the mechanical simulation process can lead to an improved design process so as to guarantee the
robustness of the design with a fuzzy approach.
In this paper, the benefits of this approach are demonstrated on an industrial windshield structure. Measure-
ments of the dynamic response of the windshield have been collected in free-free conditions for nominally
identical windshields at different temperatures. A validated FE model in the frequency range of interest has
been used. Uncertainty in its physical and geometrical parameters has been obtained from test data and has
been introduced in the numerical model in order to assess the effects on the dynamic response. The outcome
of the analysis consists of a possibilistic model of the frequency response function, given the the uncertainty
in the input parameters. This analysis not only provides a better insight into the effects of uncertainty in the
dynamic response prediction, but also provides sensitivity measures of the design parameters on the final
performance of the structure. In addition, guidelines to further improve structural designs and manufacturing
processes are also obtained.

1 Introduction

The use of Finite Element analyses in the early design stages gives the analysts valuable information on
their design in a virtual development environment. This information can be used to optimise the design
and increase the reliability of the model. Different techniques are available to achieve this goal and their
application is accomplished by the combination of user defined routines that fully automate the analysis
process.
These tools are indispensable components in today’s competent industrial world. While conducting various
analyses on different engineering structures, the problem of uncertainty emerges, awaiting the designer to
deal with it. The presence of uncertainty can be an inherent property of the design variable or may be caused
by a production process that is applied on the structures. Independently on the origin of the uncertainty, the
problem needs to be handled to gain a better insight on the behaviour of the system.
Inclusion of uncertainty into the engineering analysis can be accomplished by the use of probabilistic or
other non-deterministic tools. This study focuses on the application of a non-probabilistic technique, the
Fuzzy Finite Element (FFE) method, to analyse the effect of uncertainties on the system behaviour by using
classical finite element analysis [1]. The method is basically an extension of the Interval Finite Element (IFE)
analysis. The aim of the IFE method is to calculate the range of the result of a classical deterministic FE
analysis taking into account that some uncertain model properties vary within an interval. The FFE method
analyses the influence on the IFE result of the interval widths defined on the uncertain model properties. The
fuzzy procedure as such becomes a tool for interval sensitivity analysis [2]. The fuzzy concept is applied to
represent parameter uncertainty in a numerical FE model.
In order to demonstrate the advantages of the FFE approach, this method has been applied to investigate the
effect of several real life uncertainties on the dynamic behaviour of a windshield model. Consequently, the
FFE method is used as the numerical analysis tool in the conducted analysis. One of the important aspects
of the study is the full automation of the analysis process with the combination of commercial packages and
user defined routines. In this context, Section 2 will briefly outline the characteristics of the windshield and
its uncertainties. Section 3 gives a brief outline of the fuzzy analysis method. Section 4 the integration of
commercial packages and user-defined routines that has been realized to produce the desired results. Section
5 shows the results of the analysis in terms of the FRF responses of the windshield model. The paper is
concluded in Section 6.

2 Application case

In order to demonstrate the effective application of the FFE method, a numerical model of an acoustic
windshield has been used. The considered acoustic windshield is a sandwich structure composed of 5 layers:
2 outer layers of glass, 2 intermediate layers of a standard polymer (also called PVB) and 1 inner layer of
an acoustic polymer (also called AP). The sandwich structure is represented in Figure 1. The function of the
acoustic polymer is to act as a damping element on frequencies that operate in the acoustic range within the
cavity of a vehicle.
The physical characteristics of the windshield have
been accurately studied by Renault [3] [4] and
the dynamic characteristics of the windshield have
been measured in the framework of an extensive
experimental modal analysis campaign. Since the
polymer layers can be generally subject to vari-
ability in the material properties, the campaign (in
free-free conditions) has been carried out on three
nominally identical acoustic windshield at 5 dif-
ferent temperatures, representative of the operat-
ing conditions of the components. From the data
collected during the modal analysis campaign, a
characterisation of the inter and intra variability of
the dynamic response has been assessed by direct
exploitation of the Frequency Response Functions
(FRF) and by the estimation of the modal param-
eters [5][6]. In this context, the inter variability
refers to the variation of the observed parameters Figure 1: Acoustic Windshield Finite Element Mesh
among the three windshields at the same tempera-
ture (constant environment), while the intra variability refers to the variation of the dynamic response but
at different temperatures (changing environment). In particular, the evolution of the natural frequencies and
the modal damping coefficients as a function of the temperature and the nominally identical windshields
have been identified. The analysis of this data showed that a significant intra variability is present due to the
intermediate and inner polymer layers. As expected, the visco-elastic properties of the polymer layers are
particularly sensible to environmental changes, thus influencing the acoustic performance. More details on
the modal analysis campaign can be found in [4].
Figure 2: Detail of the finite element layers of the windshield

The data collected during the experimental campaign has been used to validate the Finite Element model
of the windshield structure, which is used to predict the dynamic behaviour. Details on the validation and
verification process can be found in [3]. The objective of our analysis is to assess the effect of uncertainty
in various physical and geometrical properties of the windshield and predicting the possible variation of the
dynamic response under uncertainty. Thus the use of a verified and validated FE model is essential.
Using the data collected by Renault in its experimental campaign, a solid FE model of five layers has been
created (Figure 2). For this model a set of 5 uncertain parameters has been selected. Their nominal values
and ranges of variation are reported in Table 1 and Table 2. The nominal values for the Young’s modulus of
the PVB and AP polymer are reported at a reference temperature of 25o C.

Parameter Nominal Value


Young’s Modulus of the PVB 108 MPa
Young’s Modulus of the AP 7 MPa
Sandwich thickness 4.96 mm
Inner Polymer layers thickness 0.67 mm
Bend in the centre of the windshield 0.0 mm

Table 1: Nominal values

Parameter Interval of variation


Young’s Modulus of the PVB [20:800] MPa
Young’s Modulus of the AP [7:50] MPa
Sandwich thickness [4.94:4.98] mm
Inner Polymer layers thickness [0.60:0.76] mm
Bend in the centre of the windshield [-3.0:+3.0] mm

Table 2: Uncertain parameters

It is worth noting that the thicknesses of the polymer and sandwich layers are not split down to the thicknesses
of the single layers. So no hypothesis can be made on the single layer thickness but some observations
can be made on the limits of the glass thickness and thus of the polymer layers. An explanation of the
characterisation used is given in Section 4.1.
3 The Fuzzy Finite Element method

3.1 Fuzzy Sets and Fuzzy FE Analysis

In order to enable the application of the fuzzy concept in the context of the Finite Element Analysis, a fuzzy
finite element (FFE) method has been developed [1, 7, 8, 9]. It is based on the concept of fuzzy sets as
defined by Zadeh [10] for the representation of uncertain model properties.
A fuzzy set can be interpreted as an extension of a classical set. Where a classical set clearly distinguishes
between members and non-members of the set, the fuzzy set introduces a degree of membership represented
by the membership function, as illustrated in Figure 3.
µx̃ (x) µỹ (y)
“ x̃ is around 50 ” “ ỹ is most probably above 50 ”
1 1

0 0
45 50 55 X 50 55 Y

Figure 3: Examples of fuzzy sets with their corresponding linguistic description

This concept allows membership values different from zero and one, which enables the representation of a
value that is only to a certain degree member of the set. The concept of fuzzy numbers is best suited to model
linguistic uncertainty in the model. It is well suited to incorporate all available subjective information, and it
does not attach any probabilistic meaning to the numerical representation.
The aim of the FFE analysis is to calculate the membership function of an output property of a finite element
calculation, based on the fuzzy descriptions of the input uncertainties. The FFE analysis does not give an
objective reliability measure as it is based on explicit subjective information. Therefore, it should be regarded
as a tool for early design validation and optimisation rather than an objective reliability analysis method.

deterministic
analysis at the
µx˜1 (x1 )
α4 -level
α4
µỹ (y)
α3 interval analysis at
α2 the α3 -level
α1
µx˜2 (x2 ) interval analysis at
α4 the α2 -level
fuzzy output
α3
α2 interval analysis at
α1 the α1 -level
fuzzy input

Figure 4: Scheme of the numerical procedure to perform a fuzzy FE analysis using 4 α-sublevels

The numerical procedure for the implementation of the FFE method is based on a general strategy for com-
puting the fuzzy result of operations on fuzzy operands, i.e., the α-sublevel technique. This technique subdi-
vides the membership range into a number of α-levels. At each level, the intersection with the membership
function of the input uncertainties results in an interval. Based on these input intervals for all uncertain pa-
rameters, an interval analysis can be performed. This results in an output interval at the considered α-level.
The total fuzzy solution is assembled from the interval results at all sublevels. Figure 4 clarifies this proce-
dure for a function of two triangular fuzzy variables with 4 α-sublevels. Based on this description, it is clear
that the FFE method requires a reliable and efficient implementation of the corresponding interval FE(IFE)
problem.

3.2 Interval Frequency Response Function Analysis

An accurate and efficient hybrid procedure for the Interval Frequency Response Function analysis has been
developed by Moens [11, 12]. The method is based on an interval translation of the deterministic modal
superposition principle. The first step of the procedure consists of a global optimisation step, in which
the input uncertainty intervals of the FE model are translated to the exact intervals for the eigenvalues, the
modal stiffness and the modal mass parameters according to each of the considered modes of the structure,
thereby neutralising all conservatism in the system matrix assembly phase. With the intervals on these modal
parameters, an accurate calculation of the upper and lower FRF bounds is performed by an interval arithmetic
procedure. The final interval FRFs have been proved to contain only a very limited amount of conservatism.
A complete mathematical description of the method can be found in the mentioned references.
The computational cost of an interval FRF analysis using the hybrid procedure is mainly determined by the
first step of this procedure, as it requires six global optimisation procedures for each mode that is taken
into account in the modal superposition. Hence an efficient optimisation procedure taking into account
all uncertain input parameters is primordial. The method for the integration of parametrised geometrical
uncertainties and other FE property uncertainties is described in the following section.

4 Analysis case implementation

In order to conduct the analysis on the windshield model, a computational framework has been set up. The
numerical procedure for the implementation of the FFE method requires the evaluation of the dynamical
response of the structure at specific values of the input parameters, so that a FE modal analysis has to be
performed for each evaluation. To accomplish this task, an automation framework has been created that
modifies the FE model to take the variation of the input parameters into account, as will be explained in
Section 4.1.

4.1 Integration of uncertainties and the morphing scenario

This subsection gives a brief outline of the process to integrate the uncertain input parameters in the fuzzy
analysis. The input parameters of Tables 1 and 2 are used. The fuzzy FE method requires that a deterministic
analysis can be performed at any combination of parameter values within their given interval bounds. Un-
certainties on the geometrical parameters are integrated by using the Morpher module of LMS Virtual.Lab
[13]. In this phase, the FE model of the windshield is morphed using the input parameters that represent
the geometrical uncertainties. Simultaneously, the other model properties are also updated in the FE model.
The output of LMS Virtual.Lab (i.e. the morphed model) becomes the input of F UZZY FRF, the K.U.Leuven
software module to calculate a fuzzy FRF (see Section 4.3). The morph scenario that is used in the analysis
consists of two steps. The first step is to morph the inner layers (PVB, AP, PVB) such that their total thick-
ness equals the parameter value representing the thickness of the inner polymer layers (see Tables 1 and 2).
The second step is to morph the two glass layers such that the total thickness of the windshield equals the
corresponding parameter value. The details of the morphing process can be observed in Figure 5.
Glass
PVB Morphing Morphing
AP
PVB Step 1 Step 2
Glass

Figure 5: Scheme of the morphing process

4.2 Mesh Morphing

This type of mesh transformation, called mesh morphing [14], consists of a change in the position of the
nodes belonging to a specified mesh, following a given expression. This expression can be mathematical,
such as a matrix, or geometrical, for example the definition of a target shape. The morphing technology -
available in LMS Virtual.Lab [13] - offers very interesting perspectives to apply such mesh modifications in
a user-friendly and application-oriented way [15] [16]. The most commonly used mesh morphing techniques
are freeform morphing and control based approach.
In free form morphing, a set of morphing tools acts directly on the mesh. The approach uses the concept
of control nodes, deformable nodes and fixed nodes. Fixed nodes determine the boundary of the area of the
deformable zone of the mesh. The deformable nodes correspond to the nodes that will be morphed. The
actual displacement of the deformable nodes is determined by the displacement of the control nodes. To
move the control nodes, several transformations can be applied including projection (to surfaces or lines),
translation, rotation, alignment and scaling.
In a control based approach, morphing operations are not carried out directly on the mesh of the existing FE
model. Instead, morphing is applied by means of control blocks. Each control block represents a volume
that envelops a small part of the existing FE mesh. By linking the location of the FE nodes inside each
control block to the location of the corners of the control block, morphing operations applied to the blocks
automatically define the geometry of the morphed FE model.
For the case considered in this paper, the first approach has been selected. The particular complexity of the
mesh made the use of control blocks less suitable. The morphing process had to be repeated a number of
times for each combination of the input parameters. The performance function needs to be evaluated for
each set of mesh morphing transformations, thus requiring each time a complete re-analysis.

4.3 Process Integration

In order to automate the whole procedure, a process integration set-up has been created. The integration has
been divided in two steps: first the integration of the mesh morphing/material updating process and then the
FFE analysis.
The first part was accomplished by using LMS Virtual.Lab [13] together with a Visual Basic Applica-
tion script, provided by the DS CATIA automation framework [17]. This block has been integrated in
the F UZZY FRF software module developed in-house at the PMA department of the Katholieke Univer-
siteit Leuven, providing the necessary problem definition, input/output files, results extraction and post-
processing [18].
An outline of the procedure is shown in Figure 6.
Figure 6: Outline of the integration procedures for the FFE analysis and FE model updating.

5 Results

To illustrate the implementation, two analyses are performed on the windscreen. Each analysis calculates a
direct FRF in a direction perpendicular to the windscreen in a certain point.

−3
10

−4
10
FRFjk

−5
10

−6
10

0 10 20 30 40 50 60 70 80
f [Hz]

Figure 7: Comparison of FRF responses for vertex analysis and Monte Carlo samples

In the first analysis, the windscreen is subject to five interval uncertainties as defined in Table 2. The upper
bound on the response is calculated using the FRF envelope algorithm described in Section 3. The bounds
on the modal parameters are calculated using a vertex analysis. A vertex analysis evaluates the goal function
in every combination of the minima and maxima of the input intervals. For n parameters, it requires 2n goal
function evaluations. Because the goal functions are monotonous with respect to the input parameters, the
vertex analysis calculates the actual minima and maxima. Figure 7 shows the upper bound on the FRF (black
line) together with the 32 (25 ) vertex FRFs and 20 uniformly sampled Monte Carlo FRFs (grey lines). The
calculated upper bound on the response is a close but conservative approximation of the real upper bound.
In the second analysis, the windscreen is subject to only one uncertain parameter: the bending in the centre
of the windscreen. The other four uncertainties defined in the first analysis are held at their reference value
as defined in Table 1. The uncertainty on the bending is described by a fuzzy number with a triangular shape
function as illustrated in Figure 8. Again, the bounds on the modal parameters are calculated using the vertex
method. The fuzzy FRF in Figure 9 shows that, compared to the first analysis (Figure 7), the influence of the
bending on the response is rather limited: it can be seen that the amplitude uncertainty in the first response
peak of the FRF is very small. In the low frequency region, the influence is very small, but around 25 Hz, it
increases quite suddenly.
1

Membership Value
0
−3 0 +3 mm

Figure 8: Fuzzy input for the bend in the centre of the windshield

−3 1.0

0.8

degree of membership
−4
log10(FRFjk)

0.6

−5
0.4

0.2
−6

0.0
0 10 20 30 40 50 60 70 80
frequency [Hz]

Figure 9: Fuzzy FRF response for the fuzzy bend in the centre

6 Conclusions

In this paper, the dynamic behaviour of a windshield with geometrical and material uncertainties has been
analysed. Several observations have been made.
For the windshield, the geometric tolerances have been parametrised as shape design variables using mesh
morphing technology, which enabled to accurately assess their effect on the dynamic performance. The
entire process (mesh morphing, finite element analysis, fuzzy finite element analysis) has been captured in
an automated multi-platform framework.
In summary, the methodology outlined in this paper constitutes an effective general-purpose tool to assess
the effects of uncertainties on the structural dynamic response of a design at a reasonable computational cost.
It identifies the envelope conditions of operation and the sensitivity of the response to uncertainties on the
input parameters. Note that the number of variables and their selection is not limited to the present choice,
as they can be changed and increased in number, depending on the particular application. In fact, it is the
engineer who should ensure that all relevant tolerance and material parameter uncertainties are included in
the analysis.

Acknowledgments

The presented methodologies are being studied and developed in the context of the EC research training
network (RTN) MADUSE (Modelling product vAriability and Data Uncertainty in Structural dynamics En-
gineering, see http://www.maduse.org/). The support of the European Commission is gratefully acknowl-
edged. In addition, the support of IWT Vlaanderen for the “Analysis Leads Design” project is gratefully
acknowledged.

References

[1] S.S.Rao and P. Sawyer, Fuzzy Finite Element Approach for the Analysis of Imprecisely Defined Systems,
AIAA Journal, vol.33, no.12, pp. 2364-2370, 1995.

[2] D.Moens and D.Vandepitte, Non-probabilistic approaches for non-deterministic dynamic FE analysis of
imprecisely defined structures, Proc. of ISMA 2004, Leuven, Belgium, pp. 3095-3119, 2004.

[3] P. Lardeur, M. Scionti, Verification and validation of finite element models for the vibrational behaviour
of a windscreen in presence of variability, Proceedings of ISMA 2006, Leuven, Belgium, September
18-20, 2006.

[4] M. Scionti, P. Lardeur, Experimental and numerical study of the intra/inter variability of an acoustic
windscreen, Proceedings of ISMA 2006, Leuven, Belgium, September 18-20, 2006.

[5] LMS International, Cada-X Modal Acquisition Monitor Manual Rev 3.5.F, Leuven, Belgium,
www.lmsintl.com, 2006.

[6] LMS International, Cada-X Modal Analysis Manual Rev 3.5.F, Leuven, Belgium, www.lmsintl.com,
2006.

[7] S.Valliappan and T.D.Pham, Fuzzy Finite Element Analysis of a Foundation on an Elastic Soil Medium,
International Journal for Numerical and Analytical Methods in Geomechanics, vol. 17, pp. 771-789,
1993.

[8] T.M. Wasfy and A.K. Noor, Finite Element Analysis of Flexible Multibody Systems with Fuzzy Parame-
ters, Computer Methods in Applied Mechanics and Engineering, vol. 160, pp. 223-243, 1998.

[9] U.O. Akpan,T.S. Koko, I.R. Orisamolu and B.K. Gallant, Fuzzy Finite Element Analysis of Smart Struc-
tures, Smart Materials and Structures, vol. 10, pp. 273-284, 2001.

[10] L.A.Zadeh, Fuzzy sets, Information and Control, vol. 8, pp. 338-353, 1965.

[11] D. Moens and D. Vandepitte, An interval finite element approach for the calculation of envelope fre-
quency response functions, International Journal for Numerical Methods in Engineering, vol. 61, no. 14.,
pp. 2480-2507, 2004.

[12] D. Moens and D. Vandepitte, Frequency response function analysis of structures with fuzzy modal
damping parameters, Proc. of the 7th AIAA Non-Deterministic Approaches Forum, Austin, 2005.

[13] LMS International, LMS Virtual.Lab, Rev. 5B, November 2005.

[14] H. Van der Auweraer, T. Van Langenhove, M. Brughmans, N. El Masri, T. Olbrechts, Advanced Mesh
Based Design Optimization for Early-Stage Virtual Prototyping, Proc. of ISMA 2004, Leuven, Belgium,
2004.

[15] I. Bosmans, M. Brughmans, P. Brizzi, H. Shiozaki, J.Yanase, Application of Morphing in Concept


Phase of Vehicle Development, Proc. 2005 JSAE Annual Congress, Yokohama, Japan, May 18-20, 2005.

[16] H. Van der Auweraer, T. Van Langenhove, M. Brughmans, I. Bosmans, N. El Masri, S. Donders,
Application of Mesh Morphing Technology in the Concept Phase of Vehicle Development, In Press, Int.J.
of Vehicle Design, issue on Computer Aided Automotive Development.
[17] Dassault Systemes, CATIA V5 R16, Copyright 2006 Dassault Systemes.

[18] M. De Munck, D. Moens, W. Desmet and D. Vandepitte, Optimisation algorithms for non-deterministic
dynamic finite element analysis of imprecisely defined structures, Proc. of the Third European Confer-
ence on Computational Mechanics, ECCM, Lisbon, Portugal, June 2006.

You might also like