Professional Documents
Culture Documents
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
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.
61
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
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:
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
Here, Obj denotes the weighted sum of the objectives. The possibility constraint Gat x is dened as:
Gat x maxCxC lim ,
x 2 xIat
63
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
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).
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
(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
Fig. 10. DOI of the nine parameters with respect to objectives and constraints.
Table 2
Summary of the objectives.
Objectives
Abbreviation Nominal
value
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
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
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
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]