You are on page 1of 31

ABSTRACT

• The main purpose of this seminar is to investigate verification, validation and


variability issues applied to an industrial component: a car windscreen.
• To focus attention on study of dynamics of acoustic windscreen under free-free
conditions.
• A numerical study performed on a simplified model in free or clamped conditions.
The verification stage, including convergence studies, concludes that the multilayer
shell model approach is valid at low temperature, when polymers are relatively
stiff.
• Finally, the main result is found as a solid model must be used for the windscreen
to correctly reproduce the physics at different temperatures.
• A validation stage, involving numerical and experimental results, has been
performed to evaluate the predictive capability of the developed numerical
models.
INTRODUCTION
• The main purpose of this paper is to investigate verification, validation and
variability issues applied to an industrial component : a car windscreen.
• component is a sandwich structure made of glass and polymers. Its contribution to
the vibro acoustic car performances is quite significant.
• Two main objectives can be identified for this paper. The first objective is to focus
the attention on the study of the dynamics of an acoustic windscreen in free–free.
• As to the second objective, an application of the verification and validation
methodology is presented to assess the capability of finite element models to
predict the natural frequencies of the acoustic windscreen, in presence of intra
variability due to temperature variations.
• Indeed, intra variability of glue and polymers’ elastic properties leads to intra
variability of the dynamic behaviour of the windscreen.
Definations and terminology
• Uncertainty and variability topics are certainly linked to the concept of robustness.
• Taguchi etal. [8] define robustness as ‘‘the state where the technology ,product, or
process performance is minimally sensitive to factors causing variability (either in manu-
facturing or user’s environment) and aging at the lowest unit manufacturing cost’’.
• Intra variability is the intrinsic variation of the behaviour of a single system in response
to changes in environmental conditions. Inter variability is the intrinsic variation of the
behaviour of nominally identical systems considered in the same environmental
conditions.
• The intra variability is often defined as test-to-test variability [3,5] and the inter
variability as scatter or unit-to-unit variability [1,2]. However, in literature, the test-to-
test variability expression is often confusing because it generally refers to repeatability
or reproducibility tests, without significant changes in the environmental conditions. On
contrary, intra variability specifically refers to the effects of environmental changes (for
example the temperature).
• Overall variability stands for the combination of intra variability and inter variability.The
aim of robust design is to master this over all variability to guarantee a good quality
level.
The verification and validation methodology

• V&V is an approach in two steps highlighted in Fig. 1. Starting from the physical reality,
two types of models are identified: a reference mechanical model and finite element
models .
• The reference mechanical model results from the partial differential equations related to
the equilibrium equations, from the con- stitutive laws and from all the initial or boundary
conditions.
• Fig. 1 shows that the verification phase deals with only numerical aspects .Roache [18]
defines the verification as ‘‘solving the equations right’’. Indeed, the objective of the
verification stage is to compare the highly accurate solution of the reference mechanical
model.
• In the best case, the highly accurate solution is exact and comes from analytical
developments. In real- world applications, the analytical developments are rarely
available and this highly accurate solution is numerically obtained with a very fine finite
element model. The verification process concludes when a satisfactory convergence of a
finite element model solution to the accurate solution of the reference mechanical model
is achieved.
• Fig. 1 also shows that the validation stage deals with the comparison between the
numerical solutions of the finite element model identified during the verification
phase and the physical reality. Indeed, in opposition to the verification stage
definition,Roache [18] defines the validation as ‘‘ solving the right equations’’. The
validation test is satisfactory if the difference between the experimental results and
the numerical results obtained with a sufficiently fine mesh is considered as
acceptable .
• A validation metric is a measure of the differences between numerical and
experimental results. When the variability of some parameters is significant, the
metrics used to compare the numerical and experimental results must also take this
variability into account.
• An increasing number of papers were recently published on the V&V issue, namely
applied to vibration or acoustics for automotive industry. Moeller et al. [19] proposed
validation metrics for the assessment of frequency response functions (FRFs).
• V&V was also applied by Lardeur et al. [21] to develop predictive finite
element modeling techniques for the vibration behaviour of galvanized
assemblies. Lardeur et al. [22] applied the V&V methodology to assess
the capability of finite element models to predict the natural frequencies
of an acoustic windscreen, in presence of intra variability.
Windscreen in free–free conditions
• For industrial applications, the final objective is to develop finite element models
in real boundary conditions.

• In reality, the windscreen is connected to the body by a glue joint. However, in this
case we are interested in investigating the vibration behaviour of the windscreen in
free–free conditions.

• This first approach in free–free conditions enables the study of the intrinsic
behaviour of this component. It prevents the eventual perturbation due to the
boundary conditions.
• the windscreen design considered in our study consists of a five-layer structure
(Fig. 2). In this configuration, the external layers (1 and 5, blue in Fig. 2) are made
of glass, while the intermediate layer is itself a multilayer structure made of two
different materials: a traditional polyvinyl butyl (PVB) (2 and 4, yellow in Fig. 2) and
a highly dissipating viscoelastic or ‘‘acoustic’’ polymer (3, orange in Fig. 2). Through
this constrained layer damping configuration, the intermediate layers’ materials
are chosen such that a critical ratio between the elasticity modulus of the glass and
that of the polymers exists beyond which part of the dynamic energy of the system
is dissipated through the shear deformation of the viscoelastic layer.
• The above-mentioned multilayer structure and the choice of viscoelastic materials
cause the windscreen to be likely affected by both the intra and the inter variability
The experimental campaign
Description of the tests
• An experimental campaign was carried out to measure the intra variability of the
acoustic windscreen. A windscreen is tested at five temperatures inside a climatic
chamber.
• The temperature range is set from5 degC upto33 degC. The test consists of an
experimental modal analysis of the windscreen in free–free conditions. Since the
research is mainly focused on the dynamic behaviour of the wind screen at low
frequencies ,the 0–200Hz range is considered. The dynamics of the windscreen is
measured through 16piezoelectric accelerometers(Fig. 3), which are fixed on the
structure through a standard waxy.
• The windscreen is excited by a standard impact hammer with a soft tip to
introduce the energy mainly at the lowest frequencies. Three points are selected
as excitation points (points 1, 7, 14 of Fig. 3) to improve the accuracy and reliability
of the post-processing.
• By hitting the structure and measuring the acceleration of the windscreen at each
temperature, a wide dataset of FRFs, coherences and excitation auto powers is
collected. Inparticular,asetof16(points)3 (excitations)5 (temperatures)¼240 FRF sis
made available to identify the dynamics of the system under different
environmental conditions.
• by hitting the structure, the operator is exposed to the risk of externally
introducing some degree of measurement uncertainty in the dataset. Frequency
windowing on input and output acquisition channels, monitoring of autopowers
and coherences traces after every impact excitation guaranteed the correct
execution of the experimental modal analysis.
• In particular, at each reference point, the structure was hitted until five FRFs for
every response point were accepted for being recorded on the hard disk. The
decision whether an FRF was acceptable or not relied upon the assessment of
coherences and autopowers functions. If coherences were above 0.95 within the
whole frequency range under exam and autopowers traces confirmed that the
structure was correctly excited (i.e. no double impacts), then the FRFs were
accepted and recorded. These five FRFs for each response point were then
summed up for averaging out any measurement noise.
• the MDOF modal parameters identification method chosen for processing the
vibration data strongly contributed to minimise local noise possibly introduced on
single measurements. This way, the global dynamic behaviour of the structure was
accurately identified and assumed to be very reliable and robust with respect to
measurement inaccuracies
• By following the above-described testing procedure, it has been proven that the
uncertainties due to a lack of repeatability and reproducibility (test-to-test
variability) are both negligible compared to the intra variability levels described
Modal parameters experimental intra
variability
• The measurement uncertainty has been quantified and is limited to less than 1%.
• the following two metrics to quantify the intra variability of natural frequencies
and damping ratios:
• .
• Fig. 4 shows the results obtained by applying Eq. (1) on the estimated modal parameters
at the five temperatures. In particular, the frequency metric is plotted in blue, while the
damping metric in red. Fig. 4 shows that the first modes vary less than the higher modes
and that the intra variability of natural frequencies and damping ratios increases on
average with the frequency .
• On the ensemble of the first 12 identified modes, a mean frequency intra variability level
of 25% is observed, with a peak value of about 40% for the 11th mode. This intra
variability is much higher than the measurement uncertainty and its level is quite
significant.

• With reference to the damping ratio, a variation up to 8% is calculated on the first 12


modes. However, due to the extremely high values of the estimated damping ratios (up
to 10%), a certain degree of prudence in stating absolute conclusions is required since
their reliability is strictly influenced by the accuracy of the estimation methods which
decreases for highly damped systems.
Verification of finite element models
• . In this study, this reference mechanical model is the linear solid elasticity theory.
The choice of a linear theory is a point that can be discussed. Indeed, polymer is a
visco elastic material and its mechanical behaviour may depend on several
parameters. Experimental tests have been carried out to identify the elastic
properties of the polymers at various amplitudes, frequencies and temperatures.
Range considered went from 5 to 50 mm for the excitation amplitude, from 5 to
500 Hz for the excitation frequency and from 5to33 degC for the environmental
temperature.
• By the study , effect of frequency and amplitude are of second order importance
compared to the temperature influence. Therefore, eventual non- linear effects
due to those parameters are not considered at this stage of the research, while
only temperature is directly taken into account.
• Nastran finite element software [29] have been used, respectively, as the pre-
processor (CAD editing and mesh creation) and the solver. A shell element was
appropriately chosen.
• The elastic properties of the materials were identified by the Renault Material
Division, thanks to a viscoanalyzer device dedicated to the vibration
characterization of viscoelastic materi- als.
• Two convergence analyses,for the boundary
temperatures5 and 33 1C,
areperformedforthefirst12frequencies.For fine
meshes along the mean surface,the
differenceobservedbetween the meshes that
contain one element per layer and the meshes
with two elements per layer is always less than
0.1%.As a conclusion, the number of elements per
layer through the thickness has a negligible
effect.The mesh is refined with 15mm element
size(40000 nodes) and checked too.The main
conclusion is that,at both temperatures,
convergence is achieved with the solid approach.
• As a complementary test, for the solid_15_1 mesh, a compar- ison between first
order and second order finite elements is also performed. All the obtained results
are consistent and demonstrate that the solid approach leads to robust and stable
models
• Because the polymer layers, and the acoustic polymer one in particular, are very
thin, the aspect ratio is very high: it varies between 50 and 450.
• since the shell model is a derived approximation of solid model. The process is
prepared through shell model with 60000 nodes at both the temperatures. No
discretization was found through the thickness.
• Fig. 6 shows th efrequencies errors obtained with the shell model, compared with
the reference solid one, for the 5and33 1C temperature conditions. This
comparison is carried out with the shell_15 andsolid_15_1meshes.At5 1C, the
maximal error amongst the first 12 frequencies is only 0.5%.Moreover,the error is
quite stable for all the modes. This demonstrates that in this case, the shell
approach is quite satisfactory
• At33 1C, the errors are much more significant.The error globally gets higher when
the frequency increases. Indeed, for mode 1 the error is only 0.8%, while the
maximal error is about 6% for mode12.The shell approach is not satisfactory
anymore, even if it remains acceptable for the very first modes.This is due to the
fact that transverse shear effects increase with the temperature and the frequency.
• The verification process leads to the conclusion that at low temperature, a shell
approach is valid while a solid approach is necessary at higher temperature. This is
a consequence of the large application domain considered here. As a conclusion,
for the windscreen study, if one wants to use a unique finite element model for
any temperature level, the solid model is required.
validation
• A Modal Assurance Criterion (MAC) calculation has been performed to correlate
the first 12 experi- mental modes to the numerical modes [31,32,33]. The MAC
matrix represents one of the standard metrics used to measure the degree of
linearity of two sets of modal vectors.
• The numerical and experimental modes shapes proved to be very well correlated
at low frequencies where 16 points allowed a very accurate mode shape
observation. MAC values above 0.95 are systematically observed for the first
modes at lower tempera- tures. For higher modes, where the structure
observability decreases, and at higher temperatures, where shear effects play an
ever increasing role, lower MAC values are observed (0.70–0.85). Nevertheless,
the accurate observation of modal vectors clearly shows the coherence of the
experimental and numerical modes.MAC values close to 1.
• Fig. 7 shows the experimental and numerical frequencies, for the first 12
modes at the five temperatures.The general trends experimentally observed
are correctly reproduced by the numer- ical model. All the frequencies
progressively decrease when the temperature increases.This monotonic
behaviour justifies a-posteriori the study of only the two boundary
temperatures during the convergence analysis of the verification phase .
Validation metrics
• The choice of the metrics defines the means used to measure the differences
between the experimental and the numerical results. In presence of variability, a
good finite element model must be able to correctly reproduce the mean
behaviour and the variability level. A natural way to do it is to consider the mean
and standard deviation values of the quantities of interest, in this case the
frequencies.
• Two metrics are proposed here: the mean value metric (MVM) and the standard
deviation metric (SDM), given by
• These validation metrics directly express, in percentage, the difference between
the numerical and the experimental results, in terms of mean value and variability
level. The results obtained with these metrics are given in Fig. 8. The MVM curve
shows that the numerical model is able to accurately predict the mean
experimental results.
• The SDM curve highlights the fact that the standard deviation of all the frequencies
of interest is underestimated by the numerical approach.
• This result is confirmed in Fig. 9, where the variability level in terms of frequency
variation is represented for the first 12 frequencies. The general experimental
trend.s are well reproduced by the numer- ical model.
• Invariability is high in higher modes than for the first ones. Nevertheless, the
numerical variability level is systematically lower than the experimental one. This
discrepancy is certainly due to the mechanical behaviour of polymers.
V & V advantages

• the application of V&V to industrial examples helps to identify the real nature of
problems which limit the correlation quality between numerical and physical
results.
• Consequently, V&V helps to assess and improve the predictive capability of a
numerical approach, in particular with the finite element method.
Conclusion:
• The paper has focused the attention on the experimental and numerical study of the
dynamics of an acoustic windscreen in free–free conditions. An application of the
verification and validation methodology has been presented to assess the capability
of finite element models to predict the natural frequencies of the acoustic
windscreen, in presence of intra variability.
• In free–free conditions the experimental campaign proves that the system is
extremely sensitive to temperature changes. The natural frequencies and the
damping ratios show an intra variability up to, respectively, 40% and 10%. Variability
of the polymers elastic properties due to temperature variations leads to variability of
the frequencies.
• For the verification stage, the main result is that the multilayered shell model
approach is valid at lower temperature, when polymers are relatively stiff. On the
contrary, at higher temperature, polymers are very flexible and shell models lead to
significant errors because transverse shear effects are considerable. Therefore, a solid
finite element model is necessary.
• For the validation stage, two validation metrics have been proposed. They assess the
mean value and standard deviation of the natural frequencies, respectively. The solid
finite element model leads to very satisfactory results for the mean value of the
frequencies. The general trends of the intra variability experimentally observed are
also well reproduced.
References

• [1] L.A. Wood, C.A. Joachim, Variability of interior noise levels in passenger cars,
in: Proceedings of the Conference on Vehicle Noise and Vibration, United
Kingdom, 1984, pp. 197–206. [2] L.A. Wood, C.A. Joachim, Scatter of structure
borne noise in four cylinder motor vehicles, in: Proceedings of the SAE Annual
Exposition and Congress, USA, 1986, ref. 860431. [3] R. Benedict, J. Porter, E.
Geddes, G. Weyeneth, Measurement of acoustic response of automotive cabin
interior, in: Proceedings of the SAE Noise and Vibration Conference and
Exhibition, USA, 1990, ref. 900047. [4] R.S. Kompella, R.J. Bernhard, Variation of
structural-acoustic characteristics of automotive vehicles, Noise Control
Engineering Journal 44 (2) (1996) 93–99. [5] J.A. Cafeo, S. Doggett, D.A.
Feldmaier, R.V. Lust, D.J. Nefske, S.H. Sung, A design-of-experiments approach to
quantifying test-to-test variability for a modal test, in: Proceedings of the 15th
International Modal Analysis Conference, USA, 1997, pp. 598–604. [6] E. Hills, B.
Mace, N. Ferguson, Response statistics of stochastic built-up structures, in:
Proceedings of the ISMA2004 Conference, Leuven

You might also like