See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/288364745

A Research on the Body-in-White (BIW) Weight Reduction at the Conceptual

Design Phase

Article in SAE Technical Papers · April 2014

DOI: 10.4271/2014-01-0743

CITATIONS READS
6 3,623

7 authors, including:

Zhenfei Zhan
Chongqing Jiaotong University
101 PUBLICATIONS 656 CITATIONS

SEE PROFILE

All content following this page was uploaded by Zhenfei Zhan on 31 May 2016.

The user has requested enhancement of the downloaded file.

 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

A Research on the Body-in-White (BIW) Weight 2014-01-0743

Reduction at the Conceptual Design Phase Published 04/01/2014

Bo Liu
Changan Auto & Vehicle Mfg. Tech

Zhenfei Zhan
Chongqing University

Xuemei Zhao, Haibo Chen, Bo Lu, Yusheng Li, and Jian Li

Changan Auto & Vehicle Mfg. Tech

CITATION: Liu, B., Zhan, Z., Zhao, X., Chen, H. et al., "A Research on the Body-in-White (BIW) Weight Reduction at the
Conceptual Design Phase," SAE Technical Paper 2014-01-0743, 2014, doi:10.4271/2014-01-0743.
Copyright © 2014 SAE International

Abstract chance of virtual testing and may lead to significant reduction

in prototype building and testing of vehicle designs.
Vehicle weight reduction has become one of the essential
Crashworthiness optimization is a systematic way to achieve a
research areas in the automotive industry. It is important to
lightweight vehicle design, while meeting all safety criteria [7].
perform design optimization of Body-in-White (BIW) at the
However, even with increasing speed of computers, the
concept design phase so that to reduce the development cost
simulation time of full FE models is still too time-consuming to
and shorten the time-to-market in later stages. Finite Element
use it for direct design optimization or robust design. It
(FE) models are commonly used for vehicle design. However,
sometimes takes hours or even days to complete a single FE
even with increasing speed of computers, the simulation of FE
simulation run. Hence, response surface model (RSM) is
models is still too time-consuming due to the increased
commonly used as a surrogate of the FE model to reduce the
complexity of models. This calls for the development of a
development time. One common approach is to use the design
systematic and efficient approach that can effectively perform
of experiment (DOE) and RSM models to construct the
vehicle weight reduction, while satisfying the stringent safety
approximation on the exploratory design space [8, 9, 10, 11,
regulations and constraints of development time and cost.
12, 13]. Several RSM methods have been developed in recent
years including radial basis function (RBF) [14], artificial neutral
In this paper, an efficient BIW weight reduction approach is
network (ANN), support vector regression (SVR), and Kriging
proposed with consideration of complex safety and stiffness
(KRG) [15, 16]. The effectiveness of different RSM techniques
performances. A parametric BIW FE model is first constructed,
varies based on the different modeling criteria, amount of
followed by the building of surrogate models for the responses of
available samples, and the nonlinear nature of the physical
interest. Stochastic design optimization is then performed to
phenomena itself [17, 18]. However, there is always some
reduce the weight of BIW and ensure the robustness and
discrepancy between the results of the true FE model and the
reliability of the optimal design. A BIW vehicle design example is
RSM regardless of the techniques used. This is especially true
employed to demonstrate the proposed methodology in details.
when the RSMs are used to predict new designs in terms of
interpolation and extrapolation. This calls for the development
Introduction of stochastic model interpolation and extrapolation method that
can address the discrepancy between the RSM and FE models
In automotive industry, The Computer aided design, more
and provide bias-corrected prediction intervals of RSM model
specifically the Finite Element (FE) models, are widely used in
responses under uncertainty. Chen et al [19], Li et al [20] and
vehicle design. To ensure the predictive capability of the
Kakkolaras et al [21] brought up the needs for validating
computer models, model validation and extrapolation [1, 2, 3,
models from the perspective of engineering design, and
4, 5, 6] has attracted more and more attention. Model
provide stochastic assessment of the validity of a model.
validation is the process of comparing CAE model outputs with
However, these methods did not fully address the needs for
test measurements in order to assess the validity and
design interpolation and extrapolation under uncertainty. With
predictive capabilities of the CAE model for its intended usage.
these considerations, Zhan et al [22] investigated three
Successful implementation of model validation will increase the
interpolation and extrapolation methods are investigated. They
 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

are Bayesian inference-based method [23], Gaussian process which adaptive to variable driven. The two characteristics can
modeling (GPM)-based method [24], and Copula-based be achieved by using SFE-concept software. Then the
method [25]. Shi et al. [26] adopted mesh dimension constraint optimization principle of parametric model is shown in figure 2.
for Enhancing the simulation efficiency of vehicle crash [52],
this method not only control of the accuracy of the model but
also improve the simulation efficiency, but still use FE model as
input in the compute term, then it's impossible to apply this
method in the optimization which need hundreds or thousands
of samples. To perform optimization of vehicle crashworthiness,
one approach is the local structural optimization, and the other
way is to modify structure according to the engineering
experience of domain experts. Yi et al. [27] separated the
Fig.2. the optimization principle of parametric model
bumper from a car, used equivalent conditions to modify the
structure with the assistance of Optistruct Software; Xie et al Compared with the traditional CAE model, parametric model is
[28] builded dynamic models to simulate the crash between logically clearer, however, it will bring huge workload when
automobile and guardrail, and optimized the guardrail by describing complex structures in details. As shown in figure 3,
increasing wheel guide board to lifting the stumble resistance only one line and one section are needed to define a straight
between wheel and guardrail post. However, none of the and simple beam, but a complex beam will call for more lines
aforementioned methods could control the optimization in the and sections. So the parametric model requires workers to blur
system level to improve the vehicle crash performance. the details for the unimportant part, which can increase the
work efficiency, reduce the model data, and improve the speed
In this paper, an efficient BIW weight reduction approach is of updating model.
proposed with consideration of complex safety and stiffness
performances. A parametric BIW FE model is first constructed,
followed by the building of surrogate models for the responses of
interest. Stochastic design optimization is then performed to
reduce the weight of BIW and ensure the robustness and
reliability of the optimal design. A BIW vehicle design example is
employed to demonstrate the proposed methodology in details.

Parametric Model of BIW Structure

In this paper, the parametric model only considers the structure
of BIW. the whole BIW is regarded as a composite of several
beam structure, panel structure and a support structure,
various types of structural elements assembled into a whole
model by map, Therefore, the complex body structures are Fig.3. contrast of simple beam and complex beam
described by node, baseline and section which are very
convenient for parametric, as shown below.
Stochastic Interpolation & Extrapolation
Methods
In order to improve the predictive capability of the RSM
models, Zhan et al [23] proposed a Bayesian based model
interpolation and extrapolation method which can utilize the
advantages of Bayesian inference and response surface model
(RSM). For the convenience of formulation, let x represents
Fig.1. Structure parameterize description design variables, Ytest and YCAE stand for test data and CAE
model results, respectively. The relationship between the test
On the other hand, the parametric structure must include the data Ytest (x) and the result generated by a CAE model YCAE (x)
information about part position, material thickness, and the can often be generalized as follows:
connection relationships between the different parts. and the
welding, adhesive, and other relationship are also defined in
the model. In order to drive parameters, definition of (1)
connection process must be able to automatic operate and
update, and overlapping relationship of the structure must be where ε is the random variable representing the test error
subject to change and adapt enough. Therefore, the parametric (relating to both experimental setup and measurement) that
model has two characteristics: one is nodes and lines which may depend on x, and δ is the error of the CAE model. Hence,
describe the structure; the other is the connections of structure it can be drawn that the “reality” or “true value” of output,
 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

namely Ytrue, is equal to the test observation minus the test Note that this method can be also be used for conjugate prior
error, and the CAE model prediction plus the model bias, as of different distributions and non-conjugate prior as well.
shown in Equation 2.
After the posterior distribution of τµ1 at each design point is
calculated, two Kriging response surface models are built for
(2) the mean µ1 and standard deviation of the available design
points within the original design space. Given a new design
Assume τ(x) = δ(x) + ε(x), it is noted that the prediction bias is configuration, xa, where no physical test data are available, to
an essential step for estimating the value of Ytrue. For the estimate the output, a prediction interval which takes into
convenience of formulation and method demonstration, it is account of CAE model result and bias inference needs to be
reasonable to assume the bias τ(x) = δ(x) + ε(x) follows a constructed. According to Equation 2 and 3, the probability p,
normal distribution with known variance as: that Ytest (xa)will fall in a given interval can be expressed as:

(3)

As shown previously, the bias τ(x) = δ(x) + ε(x) follows a (7)

normal distribution with known variance as τ(x) ∼ N(τµ (x),
τσ2(x)). Given the known variance τσ2, the mean value τµ needs where b is the 100((1 + p)/2)th percentile. Therefore, the
to be inferred with available data. In Bayesian probability numbers YCAE (xa) + δµ (xa) ± b·δσ (xa) are the endpoints of a
theory, if the posterior distributions p(θ | x) are in the same 100p% prediction interval for test data Ytest. In case p is
family as the prior probability distribution p(θ), the prior and selected as 0.95, according to normal distribution theory, b
posterior are then called conjugate distributions, and the prior equals to 1.96. Hence the output or the physical test result
is called a conjugate prior for the likelihood. For example, the considering the test error at the design point na is predicted as
Gaussian family is conjugate to itself (or self-conjugate) with
respect to a Gaussian likelihood function. If the likelihood
function is Gaussian, choosing a Gaussian prior over the mean
will ensure that the posterior distribution is also Gaussian. In
this study, the prior distribution of τµ can be assumed as: (8)

BIW Weight Reduction Case Study

(4) The body in white of a car is used as an example to illustrate
the optimization method in the paper, the structure of body in
Consider the general problem of inferring a distribution for a white is shown in figure 3, the red mark structure is engine
parameter θ given some datum or data x. From Bayes' room side beam, 9 groups parameters of the structure are
theorem, the posterior distribution is equal to the product of the selected as optimized parameters, the total number of samples
likelihood function and prior p(θ), normalized (divided) by the is 40, and minimum weight as goal, rocker beam acceleration
probability of the data p(x): value and crash rebound time for constraint.

The proposed method has been applied to a vehicle design. The

design problems for vehicle frontal impact are particularly
complex, in terms of understanding the problem and defining
(5) design requirements and design alternatives. They also involve
a very high degree of uncertainty and variability. A balance must
Given the assumption of normally distributed conjugate prior,
be struck in designing for crashworthiness between designing
let n denote the number of test observations, the posterior
the vehicle structure to absorb/manage the crash energy
distribution of τµ can be expressed as
(through structure deformation) and maintaining passenger
compartment integrity. Figure 3 shows the structure of BIW, the
red mark structure is engine room side beam, select 9 groups
(6) parameters of the structure as optimized parameters, the total
number of samples is 40, and the objective is to minimize the
weight, the rocker beam acceleration value and crash rebound
time are used as constraints.

According to the optimal hypercube Latin method to get the

sample space, after DOE calculation, complete data samples
can be obtained with the assistant of commercial optimization
software ISight, the DOE work flow is shown in figure 4.
Based on the test and CAE data, the bias distribution at each
design point x is obtained as: τ(x) ∼ N(τµ1 (x), τσ2(x)).
 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

Discrepancies exist between RSM model and FE model, hence,

in order to ensure the optimal solutions of approximate model is
also the true optimal solution, five points around the optimal
solution in the RSM model have been simulated in the fine
meshed FE model, and the results are shown in the figure 6, 7,
8, 9. The figures show that the RSM model and the FE follows
the same trend. Therefore the optimal solution is reliable.

Fig.3. the engine room side beam

Table1. Para list

Fig.6. Mass contrast

Fig.7. the maximum acceleration peak value of left B-pillar

Fig.4. DOE work flow

Building a approximate vehicle collision model which is

combine the mathematical model with error model, intended
response values are obtained by modifying parameters, and
the mathematical relationship between parameters and
response is researched. Then put approximate model into
optimization calculation, to seek the optimal solution under
some constraint conditions, that is, achieve the parameters
combination in minimum mass, but do not exceed the
maximum acceleration peak value and not be less than the
maximum rebound time of the based car. Figure 5 shows the Fig.8. the maximum acceleration peak value of right B-pillar
optimization work flow.

Fig.5. Optimization work flow Fig.9. Rebound time contrast

The proposed method can be also used to optimize stiffness

performance, and the stiffness analysis is a linear
mathematical problem, therefore the mathematical model have
 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

a high precision, the following example will prove it. The R2

value of approximate model can be used to measure the fitting
degree approximate model and sample points of degree, R2
closer to 1 indicates the approximate model has a high
credibility. The figure 10 shows that the R2 of the response in
stiff approximate model are greater than 0.9, the model has a
high reliability.

Fig.12. Contrast between approximate bend stiff value and real bend
stiff value

The information can be gotten from figure 12 that the error

between approximate bend stiff value and real bend stiff value
is less than 1.3%, and their trends are similar.
Fig.10. the R2 of the response in stiff approximate model

The parameter model of stiffness analysis and crash analysis

can be shared, wihich eliminates remodeling. 43 parameters of
the rocker, A pillar, B pillar and roof side beam are chose as the
variables for stiffness analysis,. The minimum mass as the
objective, the constraint conditions are not less than the
original model of bending and torsion stiffness, B_stiff ≥ 10968,
T_stiff ≥ 2389 KN.m/rad N/mm. The optimization results as
shown in the following table2.

Table2. Stiff optimization result

Fig.13. Contrast between approximate torsion stiff value and real
torsion stiff value

Also about between approximate torsion stiff value and real

torsion stiff value, the error is less than 1.6%, and their trends
are similar which is showed in figure 13.

To summarize, the mathematic approximate mode has a littler

As showed in table 2, compared with base model, the optimal error and a similar tend compared with the finite elements mode,
solution has a light weight, and the bend stiff and torsion stiff of and it indicates that the optimal solution of RSM mode could be
the new solution also meet the requirements and hence it is considered as a reliable representation of the true FE mode.
acceptable.

Discrepancies exist between RSM model and FE model, Summary

hence, in order to ensure the optimal solutions of approximate With ever increasing environmental requirements, vehicle
model is also the optimal solution of real situation, verify the weight reduction has become one of the essential research
points which are nearby the optimal solution in the real FE areas in the automotive industry. In this paper, an efficient BIW
model, and the results are shown in the figure 11, 12, 13. weight reduction approach is proposed with consideration of
complex safety and stiffness performances. A parametric BIW
FE model is first constructed, followed by the building of RSM
models for the responses of interest. Stochastic design
optimization is then performed to reduce the weight of BIW and
ensure the robustness and reliability of the optimal design. A
BIW vehicle design example is employed to demonstrate the
proposed methodology in details. The proposed method
improves the optimization efficiency, while one of the major
drawbacks of simply applying RSM model to crashworthiness
design optimization is that it may be quite far from the true
results. The presented vehicle design examples shows that the
Fig.11. Contrast between approximate mass value and real mass value biggest problem lies in how to set up an accurate mathematics
model. Further researches on improving the predictive
Figure 11 shows that the error between approximate mass
capability of the proposed method and its application in
value and real mass value is less than 2%, and the trends
reducing or replacing physical experiments will be conducted in
are consistent.
the near future.
 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

Acknowledgments 14. Viana, F. A. C., Haftka, R. T., and Steffen, V., Jr., “Multiple
Surrogates: How Cross-Validation Errors Can Help Us to
The authors would like to acknowledge the support from the
Obtain the Best Predictor,” Structural and Multidisciplinary
“Fundamental Research Funds for the Central Universities of
Optimization, Vol. 39, No. 4, 2009, pp. 439-457.
China” (CDJZR13110070).
doi:10.1007/s00158-008-0338-0.
15. Simpson TW, Peplinski JD, Koch PN, Allen JK (2001)
References Metamodels for computer-based engineering design:
1. Oberkampf, W. L., and Roy C. J., “Verification and survey and recommendations. Eng Comput 17(2):129-150.
Validation in Scientific Computing,” Cambridge University 16. Wang GG, Shan S (2007) Review of Metamodeling
Press, 2010. Techniques in Support of Engineering Design Optimization.
2. Ferson, S., Oberkampf, W. L., and Ginzburg, L., “Model J Mech Design 129(4):370-380.
validation and predictive capability for the thermal challenge 17. Jin R, Chen W, Simpson TW (2001) Comparative studies of
problem,” Computer Methods in Applied Mechanics and metamodeling techniques under multiple modelling criteria.
Engineering, 2008, 197, No. 29-32, pp. 2408-2430. Struct Multidisc Optim 23(1):1-13.
3. Oberkampf, W. L., and Barone, M. F., 2006, “Measures 18. Helton, J. C., Johnson, J. D., and Oberkampf, W. L.,
of agreement between computation and experiment: 2004, “An Exploration of Alternative Approaches to the
Validation. Representation of Uncertainty in Model Predictions,”
4. metrics,” Journal of Computational Physics, 217, No. 1, pp. Reliability Engineering and System Safety, 85, No. 1-3, pp.
5-36. 39-71.
5. Oberkampf, W. L., and Trucano, T. G., 2008, “Verification 19. Chen, W., Baghdasaryan, L., Buranathiti, T., and Cao, J.,
and Validation Benchmarks,” Nuclear engineering and “Model Validation via Uncertainty Propagation and Data
Design, Vol.238, No. 3, pp.716-743. Transformations”, AIAA Journal, 42(7), 1406-1415, 2004.
6. Schwer, L. E., 2007, “Validation metrics for response 20. Li, J., Mourelatos, Z., Kokkolaras, M., Papalambros, P.,
histories: perspectives and case studies,” Engineering with and Gorsich. D., Validating designs through sequential
Computers, 23, No.4, pp. 295-309. simulation-based optimization. In Proceedings of the ASME
International Design Engineering Technical Conference,
7. Fu, Y., Zhan, Z., and Yang, R., “A Study of Model Validation
2010. Paper no. DETC2010-28431.
Method for Dynamic Systems,” SAE Technical Paper 2010-
01-0419, 2010, doi:10.4271/2010-01-0419. 21. Kakkolaras, M., Hulbert, G., Papalambros, P., Mourelatos,
Z., Yang, R. J. Brudnak, M., Gorsich, D., 2011, “Towards a
8. Yang RJ, Akkerman A, Anderson DF, Faruque OM, Gu L,
Comprehensive Framework for Simulation-based Vehicle
Robustness optimization for vehicular crash simulation.
Systems Design Validation”, International Journal of Vehicle
Comput Sci Eng, 2000, 2(6):8-13.
Design, preprint.
9. Yang RJ, Wang N, Tho CH, Bobineau JP, Metamodeling
22. Zhan, Z., Fu, Y., and Yang, R., “On Stochastic Model
development for vehicle frontal impact simulation. ASME J
Interpolation and Extrapolation Methods for Vehicle
Mech Des 2005, 127(5):1014-1020.
Design,” SAE Int. J. Mater. Manf. 6(3):517-531, 2013,
10. Fang H, Rais-Rohani M, Liu Z, Horstemeyer MF, doi:10.4271/2013-01-1386.
A comparative study of metamodeling methods for
23. Zhan, Z., Fu, Y., Yang, R., Xi, Z. et al., “A Bayesian
multiobjective crashworthiness optimization. Comput Struct
Inference based Model Interpolation and Extrapolation,”
2005, 83(25-26): 2121-2136.
SAE Int. J. Mater. Manf. 5(2):357-364, 2012,
11. Stander N, Roux W, Giger M, Redge M, Fedorova N and doi:10.4271/2012-01-0223.
Haarhoff J, A comparison of metamodeling techniques
24. Jiang, Z., Chen, W., Fu, Y., and Yang, R., “Reliability-
forcrashworthiness optimization. Proceedings of the 10th
Based Design Optimization with Model Bias and Data
AIAA/ISSMO multidisciplinary analysis and optimization
Uncertainty,” SAE Int. J. Mater. Manf. 6(3):502-516, 2013,
conference 2004, Paper No.: AIAA-2004-4489.
doi:10.4271/2013-01-1384.
12. Zhu P, Zhang Y, Chen GL, Metamodel-based lightweight
25. Xi, Z., Fu, Y., and Yang, R., “An Ensemble Approach for
design of an automotive front-body structure using robust
Model Bias Prediction,” SAE Int. J. Mater. Manf. 6(3):532-
optimization. Proc Inst Mech Eng. Part D J Automob Eng.
539, 2013, doi:10.4271/2013-01-1387.
2009, 223: 1133-1147.
26. Shi, Y., Zhu, P., Zhang, P., Shen, L., Lin, Z., “Methods of the
13. Mizuno K, Arai Y, Yamazaki K, Kubota H, Yonezawa H,
Mesh Dimension Constraint for Enhancing the Simulation
and Hosokawa N, Effectiveness and evaluation of SEAS
Efficiency of Vehicle Crash” JOURNAL OF SHANGHAI
of SUV in frontal impact. Int J Crashworthiness 2008, 13:
JIAOTONG UNIVERSITY, Vol.42 No.6 Jun. 2008, doi:1006-
533-54113. Shi Lei, Yang Ren-Jye, Zhu Ping. A method for
2467(2008)06-0905-05.
selecting surrogate models in crashworthiness optimization.
Structural and Multidisciplinary Optimization, 2012, 27. Wu Y, Zhu P, Zhang Y., “Research on Structural
doi:10.1007/s00158-012-0805-5. Optimization of the Side Door Beam in SUV Based on
Side Crashworthiness Simulation” Machine Design and
Research, Vo.l 22 No 5 Oct, 2006, doi:1006-2343(2006)05-
108-04.
 Downloaded from SAE International by Chongqing University, Monday, May 30, 2016

28. Xie Q, Zhang W, Zhong Z., “Numerical Simulation Analysis

and Structure Optimization of W-beam Semi-rigid Guardrail
When Crashed with Vehicles” Bus Technology and
Research, No.1 2006, doi:1006-3331(2006)01-0010-03.
29. National Crash Analysis Center (NCAC) (2001) Public
Finite Element Model Archive. www.ncac.gwu.edu/archives/
model/index.html.
30. Forrester IJ, Keane J (2009) Recent advances in surrogate-
based optimization. Progress in Aerospace Sciences 45:
50-79.
31. Forrester IJ, Sobester A, Keane J (2008) Engineering
design via surrogate modeling: a practical guide. A John
Wiley and Sons, Ltd. Publication: 45-49.
32. Apley D, Liu J, Chen W (2006) Understanding the effects
of model uncertainty in robust design with computer
experiments. ASME J Mech Des 128: 945-958.
33. Oberkampf WL, Diegert KV, Alvin KF, Rutherford BM
(1998) Variability, uncertainty, and error in computational
simulation. ASME Proc. 7th AIAA/ASME Joint
thermophysics and heat transfer conference. ASME-HTD-
Vol. 357-2.

The Engineering Meetings Board has approved this paper for publication. It has successfully completed SAE’s peer review process under the supervision of the session
organizer. The process requires a minimum of three (3) reviews by industry experts.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical,
photocopying, recording, or otherwise, without the prior written permission of SAE International.

Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE International. The author is solely responsible for the content of the
paper.

ISSN 0148-7191

http://papers.sae.org/2014-01-0743

View publication stats