Mechanical Systems and Signal Processing 32 (2012) 5968

Contents lists available at SciVerse ScienceDirect

Mechanical Systems and Signal Processing

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

Optimisation study of a vehicle bumper subsystem with

fuzzy parameters
L. Farkas a,n, D. Moens b, S. Donders a, D. Vandepitte c
a
b
c

LMS International, Interleuvenlaan 68, B-3001 Leuven, Belgium

De Nayer Instituut, K.U. Leuven Association, J. De Nayerlaan 5, B-2860 Sint Katelijne Waver, Belgium
Department of Mechanical Engineering, K.U. Leuven, Celestijnenlaan 300, B-3001 Leuven, Belgium

a r t i c l e i n f o

abstract

Article history:
Received 10 May 2011
Received in revised form
9 November 2011
Accepted 16 November 2011
Available online 15 January 2012

This paper deals with the design and optimisation for crashworthiness of a vehicle
bumper subsystem, which is a key scenario for vehicle component design. The
automotive manufacturers and suppliers have to nd optimal design solutions for such
subsystems that comply with the conicting requirements of the regulatory bodies
regarding functional performance (safety and repairability) and regarding the environmental impact (mass). For the bumper design challenge, an integrated methodology for
multi-attribute design engineering of mechanical structures is set up. The integrated
process captures the various tasks that are usually performed manually, this way
facilitating the automated design iterations for optimisation. Subsequently, an optimisation process is applied that takes the effect of parametric uncertainties into account,
such that the system level of failure possibility is acceptable. This optimisation process
is referred to as possibility-based design optimisation and integrates the fuzzy FE
analysis applied for the uncertainty treatment in crash simulations. This process is the
counterpart of the reliability-based design optimisation used in a probabilistic context
with statistically dened parameters (variabilities).
& 2011 Elsevier Ltd. All rights reserved.

Keywords:
Uncertainty
Fuzzy
Crashworthiness
Design optimisation
PBDO

1. Introduction
In a mechanical product development process, during the different virtual design stages, performance attributes are
optimised and reliability, robustness and safety are evaluated. As computer simulations are given a more central role in the
decision-making process, physical trustworthiness of the numerical modelling process becomes crucial. The credibility of
the virtual modelling process can be increased by introducing model parametric non-determinism into the numerical
simulations. Model parametric non-determinism represents different model parameters such as material properties,
geometrical tolerances that carry some physical variation (variabilities) or represent some information deciency
(uncertainties). The application of the fuzzy nite element method in early design stage enables the analyst to assess
the effect of different model uncertainties on the product performance. Moreover the combined application of the fuzzy
Finite Element (FE) concept with optimisation leads to an optimised product for which the performance is guaranteed in
the presence of uncertainties. Such uncertainties can arise from the not yet known production process that affects the

Corresponding author.
E-mail address: laszlo.farkas@lmsintl.com (L. Farkas).

0888-3270/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2011.11.014

60

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

geometry in terms of tolerances, or such uncertainties can be accounted for material imperfections that occur in the real
physical product.
During the past decades passive safety is treated as an attribute with increased importance. Bumper systems play an
important role in the energy management of vehicles during low-speed accidents. Optimisation technology and
automation enable efcient balancing between different performance attributes. Optimisation is typically done without
considering the different model uncertainties such as properties subject to tolerances, environmental effects, non-uniform
material properties, etc. The possibility-based design optimisation process that implies fuzzy nite element analysis
allows the designer to include the effect of non-deterministic model properties in the design performance optimisation.
The bumper subsystem optimisation based on this rationale will result in product performance that is guaranteed in the
presence of uncertainties already from the early design stages. In this context the possibility-based design optimisation
together with the fuzzy nite element method is applied in conjunction with the automated design process. The bumper
optimisaton study in this paper will compare two optimisation scenarios. The rst scenario is based on a classical crisp
optimisation that is compared with the second optimisation scenario considering model uncertainties.
This paper presents the automated process for optimisation in Section 2. This process is the basis for the optimisation
with fuzzy uncertainties presented in Section 4. The presented methodologies are demonstrated in the optimisation study
of a vehicle bumper subsystem (Section 5).

2. Integrated methodology
A methodology is developed and presented to support automated balancing between different crash attributes of the
vehicle bumper system. Fig. 1 presents the schematic representation of the bumper optimisation process, starting from
geometric design. The process consists of three main elements. The rst element incorporates design modication and preprocessing in LMS Virtual.Lab [1]. In the second phase, the impact problem is solved with LS-DYNA [2]. The full process of
the crash scenario is then captured in the third element OPTIMUS [3], which allows the process integration and design
optimisation of the sequence in an automated way.

2.1. Integrated solution for geometry based multi-attribute simulation

A key element in this integrated process is LMS Virtual.Lab, which addresses multi-attribute model assembly and
analysis areas to perform end-to-end assessment of a design with respect to multiple performance attributes long before
committing to expensive tooling and physical prototypes. For the vehicle bumper subsystem of interest, engineers can
start from the CAD design, dene a generic assembly model, dene multi-attribute simulation models and meshes, as well
as multiple analysis cases (see Fig. 2). The entire process is fully associative, enabling automated iteration of design and
model changes, which is key towards an efcient optimisation process.

2.2. Process integration and automation for optimisation purpose

In order to automate the entire design procedure from parameter changes to analysis results processing, the above
process has been formalized and integrated. For the present case, the OPTIMUS software package has been used to apply
the selected analysis methodology and to integrate the different analysis tools for parameter pre-processing, mesh
regeneration, crash analysis as well as output extraction and post-processing. The process integration workow has
enabled the automatic execution of the different analysis phases in order to automatically iterate during the optimisation
process and nd the optimal design. Fig. 3 shows the workow of the multi-attribute optimisation process, which has been
captured.

Fig. 1. Schematic representation of the automated process.

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

61

Fig. 2. Integrated solution: from CAD changes to FE models.

Fig. 3. Process integration workow in OPTIMUS.

3. Design exploration and optimisation tools

This section describes different tools that have been adopted for the optimisation application presented in Section 5.
Design of experiments (DOE) is a technique [5] that in a statistics context allows the analysis of correlations or shows the
statistical signicance of an effect, but it is also used for screening purposes or to build meta-models. OPTIMUS provides a
wide range of DOE methods for different kinds of applications, such as factorial designs, Box-Behnken, Latin Hypercube,
Taguchi or Monte Carlo sampling [4]. In the bumper optimisation process, the DOE strategy is used with double purpose:
on the one hand it allows the extraction of global sensitivities or so called degree of inuence (DOI) [7], on the other hand,
the DOE experiments serve as a basis for response surface models (RSM).
In order to identify the most signicant parameters in an optimisation process, a large scale sensitivity analysis is
performed (DOI). Opposed to the generally applied local sensitivity measures based on nite differences, this approach
provides large-scale sensitivity information that is calculated based on DOE. Given that for each parameter i, a specic
output o is available at three different levels (minimum, center, maximum), the variation of the output o with respect to
parameter i is approximated: the large-scale sensitivity is given by VARoi 9D19 9D29 (see Fig. 4). The DOI for each
parameter-output pair is expressed with the following formula:
,
X
o
o
DOIi VARi
VARoi
1
i

The DOI information is used to select a subset of parameters that have strong inuence on the outputs. Parameters with a
minor inuence can be omitted form further analysis. This way, the computational burden on the optimisation is relaxed.
DOE is often used to build a response surface models (RSM) [6,8]: a meta-model is estimated from the experimental data,
to build an approximate functional relationship between the input parameters and the true response. In this context,

62

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

Fig. 4. Variation of output w.r.t. an input parameter.

OPTIMUS offers a whole range of meta-models, from the simple polynomial approximations to more advanced Radial Basis
Functions or Kriging models [4].
In many cases, design engineers must deal with multiple objectives, possibly conicting with each other, so that some
trade-off between the optimality criteria is needed. The multi-objective optimisation problem for the present application
case is transformed to single objective optimisation by optimising the weighted sum of the different objectives. In order to
limit the total computational effort required for a full optimisation process, a hybrid optimisation approach has been used,
taking advantage of DOE and RSM techniques, which is summarized in the following steps:






Design space exploration with DOE

Response surface modelling of the functional performance
Possibility-based design optimisation, based on the response surface model
Validation of the obtained results

For the present paper, given the computational time required for one single execution of the complete analysis, the DOE
approach limits the total computational effort that needs be spent. The optimisation relies on the creation of response
models to considerably speed up the process. To guarantee the validity of the optimum found with the efcient RSM
analyses, the results of the optimisation process obtained with the RSM are veried, which allows assessing the local error
at the optimum between the predictive response model and the simulation analysis.
4. Fuzzy uncertainties and possibility-based design optimisation
In a mechanical product development process, during the different virtual design stages, performance attributes are
optimised and reliability, robustness and safety are evaluated. As computer simulations are given a more central role in the
decision-making process, physical validity of the numerical modelling process becomes crucial. The credibility of the
virtual modelling process can be increased by introducing model parametric non-determinism into the numerical
simulations. Model parametric non-determinism represents different model parameters such as material properties,
geometrical tolerances that carry some physical variation (variabilities) or represent some information deciency
(uncertainties). Example for variability: a thickness that has a certain production tolerance which can be statistically
described with mean and distribution. Example for uncertainty: a geometrical dimension that is around 10 mm; this lack
of information can be due to an open design decision. The fuzzy nite element method (FFEM) is a well-known approach
for uncertainty modelling, which has been applied successfully for NVH and static analysis [12,13]. Furthermore, the FFEM
is a valuable sensitivity and tolerance analysis and robust design optimisation tool [9]. The different numerical engineering
tools can be extended to the concept of fuzzy numbers [10,14] using interval analysis (interval nite element
methodIFEM). The most common implementation of the FFEM approach based on IFEM, is the a-cut strategy, which
is schematically represented in Fig. 5. The fuzzy FE analysis can be directly used in a possibility-based design optimisation
[11] (PBDO). In the performance measure approach applied to PBDO, the optimisation problem is constrained by
possibility constraints, which are the maximum values of the constraints on the at -cut. at represents the target possibility
of failure. In order to propagate the parameter uncertainties during the non-linear performance optimisation, in this paper,
the FFEM is applied in conjunction with PBDO in the context of a crashworthiness application. The aim is to ensure the
bumper system performance in multiple low-speed crash impacts. The multi-objective optimisation problem of the
bumper subsystem is dened as:
minimize Objx subject to Gat x r 0

with xmin r x r xmax

Here, Obj denotes the weighted sum of the objectives. The possibility constraint Gat x is dened as:
Gat x maxCxC lim ,

x 2 xIat

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

63

Fig. 5. Fuzzy procedure.

Fig. 6. CAD-based mesh and assembly of the bumper system.

The failure occurs when the dened constraints C(x) exceed the limit values Clim. The bumper parameters x with fuzzy
uncertainties affect both on the objective and the constraint function. The evaluation of the objective function is a crisp
evaluation. At each evaluation of the constraint Gat x an interval analysis is performed. This interval analysis is based on
the input intervals that corresponds to the at -cut on the fuzzy membership functions of the bumper parameters. The
resulting optimisation guarantees that in presence of fuzzy uncertainties on the bumper parameters, the constraints are
violated only up to a degree of possibility equal to at , while the objectives are minimised.
The automated design process captured in OPTIMUS (see Section 2) can be directly used for the FFEM-based PBDO. The
selected fuzzy analysis approach and the PBDO algorithm implemented in MathWorks MATLAB interfaces with OPTIMUS:
for each design evaluation OPTIMUS is called in batch mode. In this context, the MATLAB fuzzy and PBDO code acts as
external driver for the core OPTIMUS process.
5. Bumper subsystem optimisation study
To illustrate the methodologies described in Sections 24 of this paper, an optimisation study is performed on an
industrial parametric CAD bumper system. This application case has been dened by LMS and PUNCH as a representative
bumper design scenario of industrial complexity, which will be used in this paper to demonstrate the structural
optimisation under fuzzy uncertainties.
5.1. Bumper system
The bumper geometry has been taken from an industrial design practice with a mesh density that is both acceptable for
the predictions of interest and also feasible in terms of computational effort. The geometry consists of a cross-section made
of two chambers. Subsequently, an assembly is made to connect with the bumper, the longitudinal beams through
brackets using seam weld connections and rigid connections (see Fig. 6).
5.2. Load cases: repairability low speed impact
Two load cases are considered for the evaluation of the crashworthiness performance of the vehicle bumper system: the
Allianz crash repair test and the impact to pole test. The Allianz test (AZT) is the most important low speed load case in the
vehicle bumper design. This test aims at evaluating the reparability cost, and is used by insurance companies to determine
the insurance fee of a vehicle. The more damage the vehicle will endure in this impact case, the higher the insurance fee

64

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

will be. The AZT test protocol prescribes a 40% offset impact at 16 km/h against a rigid barrier(see Fig. 7 left). To minimize
the reparability cost, the deformation should be limited within the bumper system and the load transferred to the
longitudinal members should be limited to avoid permanent deformations. Frontal pole-impact test is used to study the
intrusion during a frontal impact with a rigid pole. Similarly to the AZT test, it allows evaluating the repairability cost of
the bumper system in a different typical crash scenario. The larger the intrusion, the higher the risk of damaging costly
parts, such as the engine cooling system. This test consists of a 15 km/h central impact against a rigid pole (see Fig. 7 right).

5.3. Optimisation denition

The goal of the optimisation process is to obtain an optimised bumper prole in terms of mass and Allianz test crash
performance, while satisfying a set of design constraints. Multi-objective optimisation ensures an optimal trade-off
between the two selected objectives. At each iteration of the DOE experiments, two parallel analyses are performed, one
analysis for each load case.

5.3.1. Input parameters

In order to optimise the bumper system, nine parameters are considered (Fig. 8). Parameters L1, H1, H2, G1, G2, D1 and D2
are geometrical parameters that dene the prole of the bumper, while t1 and t2 represent shell thickness values. The
cross-sectional length of the bumper is considered to be xed to L 150 mm. The parameter ranges and the nominal values
are presented in Table 1.

5.3.2. Objectives and constraints

Nowadays, with the increasing awareness of the environmental footprint of the vehicle, mass reduction of the different
vehicle subcomponents is mandatory. Reducing the mass of the bumper is therefore the primary objective. To optimise
energy absorption potential of the bumper for the Allianz test, the deviation with respect to an ideal 85 kN constant curve
is considered. The target curve is the ideal force level to absorb the total kinetic energy of a 1200 kg car that crashes into
the rigid barrier in conformity with the Allianz test, with an initial velocity of 16 km/h. The target force level is equivalent
to 11.9 kJ (total initial kinetic energy), based on a deformation length of 140 mm (total collapse of the bumper section). The
average deviation of the actual force-deection curve from this ideal curve is expressed with the root mean squared error

Fig. 7. The Allianz and impact to pole load cases.

Fig. 8. Bumper parameters.

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

65

Table 1
Design parameters.
Parameter

L1

H1

H2

G1

G2

D1

D2

t1

t2

Min (mm)
Max (mm)
Nom (mm)

60
100
85

70
100
85

55
65
60

5
15
10

0
10
5

 15
15
0

 15
15
0

2
4
3

2
4
3.3

Fig. 9. X-force at Section 1 vs. time.

(RMSE) formula that is based on 10 sample points, sampled over the impact time interval:
v
u 10
uX i
RMSE_F X t
F x 85kN2 =10

i1

Fig. 9 shows the AZT load case sectional force X at Section 1 for the nominal bumper variant. The horizontal line represents
the ideal force curve, while the black dots represent the sampled data for the RMSE calculation. The optimisation is subject
to two constraints: the X force level at Section 1 during the AZT test is limited to 120 kN, and the intrusion for the pole
impact scenario is limited to 100 mm.
5.4. First screening results: DOI
In order to identify the most signicant parameters with respect to the objectives and constraints, a rst output
screening based on the DOE is performed. The objective of this step is to reduce the number of parameters from 9 to 5. This
parameter reduction results in a reduced number of experiments used as basis for the RSM. For a 3-level full factorial (3FF)
design, the full set of nine parameters would result in 19 683 experiments. 3FF design based on the reduced set of
parameters results in a feasible number of 243 experiments. Given 70 min CPU time for 1 experiment, the 3FF design could
be covered within 12 days. The DOE adopted for the large scale sensitivities (DOIs), consists of a set of experiments that
includes the central point and the extreme points, requiring a total number of 19 evaluations.
Based on the DOI results (see Fig. 10), a set containing ve parameters is selected: L1, H1, H2, t1, t2.
5.5. DOE and RSM selection
The ve considered parameters are used for a DOE based on a 3FF design extended by a LHS, to ensure uniform
sampling of the design space. A total number of 543 experiments ensured the optimisation project to t in a time frame

66

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

Fig. 10. DOI of the nine parameters with respect to objectives and constraints.
Table 2
Summary of the objectives.
Objectives

Abbreviation Nominal
value

Total bumper mass

AZT test: RMSE of Xforce w.r.t. 85 kN

Mass
RMSE_F X

5.54 kg
33 kN

Table 3
Summary of the constraints.
Constraints

Abbreviation

Nominal value

Limit value

Max_F X
Max_Int

135 kN
52 mm

120 kN
100 mm

AZT test: highest X force (Section 1)

Pole impact test: largest bumper intrusion

Table 4
The results of the two optimisations.
Results

L1 (mm)

H1 (mm)

H2 (mm)

t1 (mm)

t2 (mm)

Mass (kg)

RMSE_F X (kN)

Max_F X (kN)

Max_Int (mm)

Start
Optimisation 1
Simulation 1
Relative error 1
Optimisation 2
Simulation 2
Relative error 2

80
77.04

85
82.85

60
62.08

3
2.65

3.3
3.23

77.55

84.99

61.02

2.79

3.27

5.54
5.11
5.11
0%
5.31
5.31
0%

33.7
12.0
21.86
45.10%
12.7
24.26
47.65%

135
110
128
14.06%
110
131
16.03%

52.5
65.57
65.6
0.05%
60.45
60.3
0.25%

of 4 weeks. The experimental results of the objectives and constraints are then used to build a meta-model for each
objective and constraint. The Radial Basis Functions-based (RBF) interpolating response models [9] are adopted for this
purpose, and subsequently used in the possibility-based optimisation routine.
5.6. Bumper design optimisation
The most signicant design variables are considered for two optimisation studies, for which the objective is dened as
the weighted sum of the normalized Mass and RMSE_F X (Table 2):
Obj w1 Mass w2 RMSE_F X

with w1 0:5

and

w2 0:5

Two optimisations are performed and compared, both optimisations based on the objective dened in Eq. (5) and the
active constraints dened in Table 3. Both optimisation processes are based on the RSM evaluations of the bumper design
process captured in OPTIMUS. The rst optimisation with the bumper parameters considered as crisp values within the
allowed range implies standard non-linear constrained optimisation. In the second optimisation, the bumper parameters
are subject to fuzzy uncertainties, therefore a dedicated optimisation process (PBDO) is capable of handling the
uncertainties. The different parameters are subject to simple triangular membership functions: the base is spanned by
the active value 71 mm and respectively 70.1 mm for the three geometrical parameters (L1, H1 and H2) and respectively
for the 2 thickness values (t1 and t2). These theoretical uncertainties can represent for example manufacturing tolerances
specic to the adopted production process. Table 4 compares the result of the two different optimisations. The validation
of the optimum shows large difference for the second objective RMSE_F X , which indicates room for improvement of the
RSM for this specic output. New experiments can be computed, improving the predictive accuracy of the RSM model,

0.9

0.9

0.8

0.8

0.8

0.8

0.7
0.6
0.5
0.4
0.3

0.7
0.6
0.5
0.4
0.3

membership level

1
0.9
membership level

0.7
0.6
0.5
0.4
0.3

0.7
0.6
0.5
0.4
0.3

0.2

0.2

0.2

0.2

0.1

0.1

0.1

0.1

5
5.2
Mass [kg]

5.4

1.5
2
RMSE FX [N] x 104

1.1
1.2
1.3
Max FX [kN] x 105

0 50

0.9

0.9

0.9

0.9

0.8

0.8

0.8

0.8

0.7
0.6
0.5
0.4
0.3

0.7
0.6
0.5
0.4
0.3

membership level

membership level

membership level

membership level

67

0.9
membership level

membership level

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

0.7
0.6
0.5
0.4
0.3

0.4
0.3
0.2

0.1

0.1

0.1

0.1

1.5
2
4
RMSE FX [N] x 10

1.1
1.2
1.3
Max FX [kN] x 105

70

0.5

0.2

5.6

60
65
Max Int [mm]

0.6

0.2

5.2
5.4
Mass [kg]

80

0.7

0.2

60
70
Max Int [mm]

0
55

Fig. 11. The fuzzy optimums for the two optimisation cases. (a) Crisp optimum, (b) PBDO optimum.

however considering the demonstrative scope of this project this potential improvement was not addressed. Both
optimisation runs result in an improved bumper performance: both the mass and RMSE of the normal sectional force as
compared to the ideal force prole are signicantly reduced. The prescribed constraints are however violated in the
presence of parametric uncertainties for the crisp optimisation case. Fig. 11 presents the fuzzy results of the optimum
design for the two different optimisation cases. Notice in Fig. 11(a) the constraint Max_F X is violated for the crisp
optimisation case and it is satised up to a degree of possibility at 0:05 for the BPDO case (Fig. 11(b)).
6. Conclusion
This paper presents the application of the Fuzzy Finite Element Method (FFEM) in conjunction with Possibility-Based Design
optimisation (PBDO) in the context of crashworthiness. A generic methodology for automated crash performance optimisation
is the basis for the vehicle bumper subsystem optimisation case. LMS Virtual.Lab offers an integrated solution for CAD-based
simulations with the benets of decreasing analysis time by means of quick model updates, by offering an integrated platform
for multi-attribute system modelling for design engineers. The crash design process from parametric model modication and
preprocessing in LMS Virtual.Lab and the solution of the crash problems with LS-DYNA is captured with the use of OPTIMUS.
The OPTIMUS software package is a dedicated platform for process automation that enables multi-disciplinary design
optimisation. The automated design process captured in OPTIMUS is used for the FFEM-based PBDO. For this, the uncertainty
descriptions are carried through the optimisation process by the application of the possibility-based design optimisation
methodology in such way that the optimised design performance is guaranteed in the presence of uncertainties.

Acknowledgements
The work presented in this paper has been performed in the frame of the ongoing research project IWT-070401 SimulationBased Development of an Innovative Bumper Beam Concept with Integrated Crashbox Functionality (I-CRASH), which is carried

68

L. Farkas et al. / Mechanical Systems and Signal Processing 32 (2012) 5968

out by PUNCH Metals N.V. and LMS International, and which is supported by IWT Vlaanderen. Furthermore the support of IWT
Vlaanderen with the SBO project called Fuzzy Finite Element Method is gratefully acknowledged.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]

LMS International, LMS Virtual. Lab, Leuven, 2009.

Livermore Software Technology Corporation, LS-DYNA Keyword Userss Manual, Version 971, vol. III, 2009.
Noesis Solutions N.V., OPTIMUS, Leuven, 2009.
Noesis Solutions N.V., OPTIMUS Theoretical Background, Leuven, 2008.
D.C. Montgomery, Design and Analysis of Experiments, John Wiley, New York, 1984.
J.F.M. Barthelemy, R.T. Haftka, Approximation concepts for optimum designa review, Struct. Optim. 5 (1993) 129144.
L. Farkas, D. Moens, D. Vandepitte, W. Desmet, Application of fuzzy numerical techniques for product performance analysis in the conceptual and
preliminary design stage, Comput. Struct. 86/10 (2007) 10611079.
F. Hongbing, F.H. Mark, Metamodeling with radial basis functions, 46th AIAA/ASME/ASCE/ ASC Structures, Structural Dynamics and Materials
Conference, AIAA-2005-2059, 2005.
D. Moens, D. Vandepitte, A survey of non-probabilistic uncertainty treatment in nite element analysis, Comput. Methods Appl. Mech. Eng. 194
(2005) 15271555.
L. Zadeh, Fuzzy sets, Inf. Control 8 (1965) 338353.
L. Du, K.K. Choi, B.D. Youn, Inverse possibility analysis method for possibility-based design optimization, AIAA J. 44 (11) (2006) 26822690.
D. Moens, D. Vandepitte, An interval nite element approach for the calculation of envelope frequency response functions, Int. J. Numer. Methods
Eng. 61 (2004) 24802507.
L. Farkas, D. Moens, S. Donders, D. Vandepitte, W. Desmet, The Fuzzy FE approach to assess the uncertain static response of an industrial vehicle,
in: Proceedings of ISMA 2006, Leuven, Belgium, September 2006.
M. Hanss, The transformation method for the simulation and analysis of systems with uncertain parameters, Fuzzy Sets Syst. 13 (2002) 277289.