Hindawi Publishing Corporation

Mathematical Problems in Engineering

Volume 2014, Article ID 965157, 9 pages
http://dx.doi.org/10.1155/2014/965157

Research Article
An Adaptive Metamodel-Based Optimization Approach for
Vehicle Suspension System Design

Qinwen Yang,1 Jin Huang,2 Gang Wang,1 and Hamid Reza Karimi3
1
College of Mechanical and Vehicle Engineering, Hunan University, Changsha, Hunan 410082, China
2
School of Software, Tsinghua University, Beijing 100084, China
3
Department of Engineering, Faculty of Engineering and Science, University of Agder, 4898 Grimstad, Norway

Correspondence should be addressed to Jin Huang; [email protected]

Received 8 October 2013; Accepted 21 November 2013; Published 4 February 2014

Academic Editor: Hui Zhang

Copyright © 2014 Qinwen Yang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The performance index of a suspension system is a function of the maximum and minimum values over the parameter interval.
Thus metamodel-based techniques can be used for designing suspension system hardpoints locations. In this study, an adaptive
metamodel-based optimization approach is used to find the proper locations of the hardpoints, with the objectives considering the
kinematic performance of the suspension. The adaptive optimization method helps to find the optimum locations of the hardpoints
efficiently as it may be unachievable through manually adjusting. For each iteration in the process of adaptive optimization,
prediction uncertainty is considered and the multiobjective optimization method is applied to optimize all the performance indexes
simultaneously. It is shown that the proposed optimization method is effective while being applied in the kinematic performance
optimization of a McPherson suspension system.

1. Introduction The performance index of a suspension system is a

function of the maximum and minimum values over the
The suspension K&C characteristics have directly effects on parameter interval [6, 7]. Thus, it is impossible to apply
vehicle handling and riding performances and thus gain directly a well-developed optimization algorithm based on
much effort and are of great importance in vehicle develop- gradient information. It can be very difficult to evaluate
ment. With bush uncertainty and mechanical flexibility, it is the analytical design sensitivity of the hardpoints locations
very difficult to predict sensitivity of hardpoints locations in because the deviation is defined by using the maximum and
the kinematic performance of a suspension system as they minimum values over the parameter interval. Metamodeling
are highly nonlinear and coupled [1, 2]. Traditional chassis techniques, which were initially developed as “surrogates” of
developing, which benefits from the development of modern the expensive simulation process for improving the overall
virtual prototyping technology, can now do system design computation efficiency and quality [8], are useful in such
effectively through some techniques, like the DOE (design a field. Metamodel-based methods in vehicle design area
of experiment) technique, as well as other experience-based mainly focused on FEM related problems [9]. Much work
attempts [3–5]. However, the mechanism of the suspension has been done in suspension design area, most of which
system is designed by trial and error based on the designer’s focused on complex structural related area. The authors in
experiences and intuition, which will be time-consuming in [10] studied a mechanical analysis of a suspension optimal
finding a sufficiently good solution since a lot of attempts design for suspension system based on reliability analyses,
may be needed in doing virtual prototyping simulations. taking into consideration tolerances and grafting a reliability
A featured optimization technique may be useful to give analysis that applied the mean-value first order method
guidance through the design process. with tolerance optimization. Choi et al. recently studied
2 Mathematical Problems in Engineering

optimal design for automotive suspension systems based on 7

reliability analyses for enhancing kinematics and compliance
characteristics; they performed reliability optimization with
the single-loop single-variable method by using the results
from a deterministic optimization as initial values. The robust
design problem was solved with 1700 analyses for 15 design
variables and four random constants [11]. Recently, researches
on applying metamodel-based optimization techniques to 4
the K&C performance design of vehicle suspension systems
were tried and gained reasonable good results. Kang et al.
introduced a robust suspension system design approach,
which takes into account the kinematic behaviors influ- 6
enced by bush compliance uncertainty, using a sequential
approximation optimization technique. The robust design 5
problem has 18 design variables and 18 random constants 2
with uncertainty [12]. After then, the team proposed a 1
so-called target cascading method for the robust design
optimization process of suspension system for improving
vehicle dynamic performances [13]. The design target of the 3
system is cascaded from a vehicle level to a suspension system
level. The design problem structure of suspension system
is defined as a hierarchical multilevel design optimization, Figure 1: The kinematic structural model of a McPherson suspen-
and the design problem for each level is solved using the sion system.
robust design optimization technique based on a metamodel.
The researches above opened the way for doing suspension
system optimization by using metamodel-based techniques
with their effectiveness tested, which motivated us to further 2. Optimization Problem in
research along the direction. However, the former research Suspension System
neither considered optimizing several objectives simulta-
neously to make them stable in comparable intervals, nor In multibody dynamics point of view, a suspension sys-
gave the searching guidance for each design parameter in tem can be classified into several groups according to the
accelerating the convergence of objective parameters. mechanical joints. In this study, we take McPherson type
In this paper, we employ a new adaptive metamodel- suspension system in consideration, which is sensitive to
based optimization approach for guiding suspension system the kinematic performance. Figure 1 shows a kinematic
design in determining appropriate hardpoints locations. The structure model of a McPherson suspension system. Major
following characteristics distinguish the approach from other components include the strut, the lower control arm, the
metamodel-based applications in suspension system design. tie rod, and the knuckle. Connections between individual
(1) As we have several vehicle performance related parameters components are spherical, revolute, and universal joints, as
to be optimized, adaptive weighting factors are used for well as compliance elements such as springs, dampers, and
multiobjective optimization to ensure that all the objectives bushings. The design purpose of this study is to determine the
are optimized simultaneously. (2) For each iteration of locations of the hardpoints according to the system kinematic
the adaptive optimization, both predicted mean value and performances, without considering the elastic deformations
prediction standard deviation are considered in case of the of the rigid components except for compliance elements.
system converging to the wrong optimum values. (3) We The commercial software ADAMS, which can easily get the
select kriging method as it is more accurate and efficient than suspension system performance, is employed for modeling
other metamodeling methods in solving highly nonlinear and analyzing the suspension system.
problems. (4) The optimization approach provides the possi- The kinematic characteristics of the system include the
ble trends in selecting hardpoints locations for optimizing the positions of the fixed points of the suspension system. To
system performances. We organize the paper in the following achieve optimal solution for suspension system, designing
manner. Section 2 introduces the engineering requirements a good kinematic performance is the first step and we
for the optimization problem in suspension system design. thus carry out kinematic optimization of the suspension
Section 3 presents the methodology of adaptive metamodel- system in this study. For that purpose, the chosen suspension
based optimization considering modeling uncertainty. The performance indexes are the deviations in the camber angle,
application of adaptive optimization method in suspen- the caster angle, the kingpin incline angle, and the toe angle
sion system is given in Section 4. Section 5 compares the during the wheel stroke. Camber angle is the angle between
results of proposed adaptive metamodel-based optimization the vertical axis of the wheels used for steering and the
considering modeling uncertainty and the regular adaptive vertical axis of the vehicle when viewed from the front or
metamodeling approach. At last, Section 6 summarizes the rear. It is defined as positive when the top of the wheel
contents of this research. moves to the outside. The camber angle alters the handling
Mathematical Problems in Engineering 3

qualities of a suspension system; in particular, a negative ̂ When an input point x is given, the metamodel can then
𝑓(⋅). 0
camber angle improves grip when cornering. Caster angle be used to predict the output 𝑌0 using
is the angular displacement from the vertical axis of the
suspension of a steered wheel in a vehicle, measured in the 𝑌0 = 𝑓̂ (x0 ) . (2)
longitudinal direction. It is the angle between the pivot line
(in a car, an imaginary line that runs through the center of Among various metamodeling schemes, kriging method
the upper ball joint to the center of the lower ball joint) and is often selected due to its high accuracy and efficiency
vertical line. On most modern designs, the kingpin is set at for solving nonlinear problems [14]. Kriging method was
an angle relative to the true vertical line, which is the kingpin originated from the geostatistics community [15] and used by
inclination angle, as viewed from the front or back of the Sacks et al. [16] for modeling computer experiments. Kriging
vehicle. The angle has an important effect on the steering, method is based on the assumption that the true system
making it tend to return to the straight ahead or center response, 𝑌, can be modeled by
position. Toe angle is the symmetric angle that each wheel 𝑚
makes with the longitudinal axis of the vehicle, as a function 𝑌 = ∑ 𝛽𝑖 𝑓𝑖 (x) + 𝑍 (x) , (3)
of static geometry and kinematic and compliant effects. The 𝑖=0
toe angle change plays an important role in determining the
apparent transient oversteer or understeer. Positive toe is the where 𝑓𝑖 (⋅) is a regression function, 𝛽𝑖 is the coefficient for
front of the wheel pointing in towards the centerline of the 𝑓𝑖 (⋅), 𝑚 + 1 is the number of regression functions, and 𝑍(⋅) is
vehicle. Negative toe is the front of the wheel pointing away the stochastic process with zero mean and covariance defined
from the centerline of the vehicle. Large errors will have a by
negative effect on the behavior of the chassis when braking
or accelerating. Cov (Z (x𝑗 ) , Z (x𝑘 )) = 𝜎2 𝑅𝑗𝑘 (𝜃, x𝑗 , x𝑘 ) , (4)
In suspension system design, the locations of hardpoints
are the most important influencing factors that determine where 𝜎2 is the process variance, 𝑅𝑗𝑘 (⋅) is the correlation
the system kinematic performances. Although engineers function, and 𝜃 is a vector with coefficients to be determined.
obtained some experiences on adjusting the locations of the For ordinary kriging, the linear part of (3) is usually
hardpoints, much effort is still needed on trying different assumed to be a constant, whereas the correlation function
trials. Furthermore, all the geometric parameters of suspen- 𝑅𝑗𝑘 (𝜃, x𝑗 , x𝑘 ) is generally formulated as
sion system are coupled, which make finding the influence
𝑝
of the hardpoints locations on the system performances a
more tough work, not mentioning there are several objectives 𝑅𝑗𝑘 (𝜃, x𝑗 , x𝑘 ) = ∏𝑄 (𝜃𝑖 , 𝑥𝑗𝑖 , 𝑥𝑘𝑖 ) , (5)
𝑖=1
to be determined. In this study, while determining the
locations of the key hardpoints, we select the important where 𝑝 is the dimension of x, 𝑥𝑗𝑖 is the 𝑖th component of x𝑗 ,
characteristics, that is, the variations of the camber angle, the 𝑥𝑘𝑖 is the 𝑖th component of x𝑘 , and 𝑄(⋅) is usually assumed to
caster angle, the kingpin incline angle and the toe angle, for be Gaussian as
the objectives since they are related much to the kinematic
performance. An adaptive metamodel-based optimization 𝑄 (𝜃𝑖 , 𝑥𝑗𝑖 , 𝑥𝑘𝑖 ) = exp (−𝜃𝑖 𝑑𝑖2 ) ,
approach will be proposed in the next section considering (6)
󵄨 󵄨
modeling uncertainties. 𝑑𝑖 = 󵄨󵄨󵄨󵄨𝑥𝑗𝑖 − 𝑥𝑘𝑖 󵄨󵄨󵄨󵄨 .

The linear predictor of kriging method can be formulated as

3. Adaptive Metamodel-Based
Optimization Method 𝑔̂ (x) = c𝑇 (x) y, (7)

3.1. Metamodeling Method. In metamodeling, the relation- where c𝑇 (⋅) is the coefficient vector and y is the vector of the
ship between a vector of input parameters, x, and an output observations at the sample sites (x1 , . . . , x𝑛 )
parameter, 𝑌, can be formulated as
𝑇
y = [𝑦 (x1 ) ⋅ ⋅ ⋅ 𝑦 (x𝑛 )] . (8)
𝑌 = 𝑓̂ (x) + 𝜀, (1)
Through minimizing the prediction variance 𝜎𝑡2 :
2
where 𝑌 is a random output variable, 𝑓(⋅) ̂ is the approximated 𝜎𝑡2 = 𝐸 [(𝑔̂ (x) − 𝑌) ] (9)
relationship, and 𝜀 is the error of metamodel due to the
uncertainty introduced by the metamodeling method. Many concerning the coefficient vector c𝑇 (x), the best linear unbi-
different metamodels, such as multivariate polynomial, radial ased predictor (BLUP) is solved as [17]
basis function (RBF), and kriging, can be used to build the 𝑇 −1
approximation relationship 𝑓(⋅).̂ In metamodeling, first 𝑚 𝑔̂ (x) = r𝑇 R−1 y − (F𝑇 R−1 r − f) (F𝑇 R−1 F) (F𝑇 R−1 y) ,
sample data (x𝑖 , 𝑦𝑖 ) (𝑖 = 1, 2, . . . , 𝑚) are collected to build (10)
4 Mathematical Problems in Engineering

where Based on the metamodel relationship 𝑓𝑚 , we can identify

𝑇
the potential input parameters x∗ that lead to the minimum
r = [𝑅 (𝜃, x1 , x) ⋅ ⋅ ⋅ 𝑅 (𝜃, x𝑛 , x)] , output parameter through optimization:
𝑅 (𝜃, x1 , x1 ) ⋅ ⋅ ⋅ 𝑅 (𝜃, x1 , x𝑛 ) min
𝑥
𝑓𝑚 (x) . (16)
R=[ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ],
𝑅 (𝜃, x , x ) ⋅ ⋅ ⋅ 𝑅 (𝜃, x , x ) The optimization result of x∗ is then selected as the vector
[ 𝑛 1 𝑛 𝑛 ]
(11) of input parameters for the (𝑚+1)th sample xm+1 . The output
𝑇
𝑓0 (x1 ) ⋅ ⋅ ⋅ 𝑓0 (xn ) 𝑌𝑚+1 corresponding to the xm+1 is subsequently obtained
F = [ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ] , through experiment or simulation. The new pair of data,
𝑓 (x ) ⋅ ⋅ ⋅ 𝑓𝑚 (xn )]
(xm+1 , 𝑌𝑚+1 ), together with all the previously collected sample
[ 𝑚 1
data are used to update the metamodel into a new relationship
𝑇
f = [𝑓0 (x) ⋅ ⋅ ⋅ 𝑓𝑚 (x)] . 𝑓𝑚+1 :
𝑌 = 𝑓𝑚+1 (x) . (17)
The coefficients in 𝜃 can be obtained by using maximum
likelihood estimation as [17] The process of identifying the potential optimal input
parameters, obtaining the output parameter through exper-
min 𝜓 (𝜃) = |R|1/𝑛 𝜎2 , (12) iment/simulation, and updating the metamodel is continued
𝜃
iteratively until the optimization criteria are satisfied.
where |R| is the determinant of R and 𝜎 is obtained by
generalized least squares fit as [17] 3.3. Adaptive Optimization considering Modeling Uncertainty.
In the process of adaptive optimization, the prediction
1 𝑇
𝜎̂2 = (y − F𝛽∗ ) R−1 (y − F𝛽∗ ) , (13) uncertainty of the developed metamodel would influence
𝑛−𝑚−1 the accuracy of the predicted optimum. Minimizing the 𝑓𝑚
directly to find the optimal input parameters may not lead to
where 𝛽∗ is the vector with coefficients achieved from good convergence for output parameters. The dual response
generalized least squares fit and is calculated by surface methodology is a powerful tool for simultaneously
−1 optimizing the mean and the variance of responses to tackle
𝛽∗ = (F𝑇 R−1 F) F𝑇 R−1 y. (14) the problem of misleading [18]. Lin and Tu [19] gave a
dual response surface method using the mean squared error
When metamodeling is used for solving specific prob- (MSE) approach as follows:
lems, collection of sample data in specific parameter space,
2
rather than the whole parameter space, is then required to min ̂𝜎2 ,
̂𝜇 − 𝑇) + 𝜔
MSE = (𝜔 (18)
𝑥
improve the quality and efficiency. This issue is critical when
expensive or extensive experiments/simulations are required where 𝜔 ̂𝜇 is the predicted response value, 𝑇 is the target
to collect the sample data. Since the relationship is unknown ̂𝜎 is the prediction standard deviation. Following
value, and 𝜔
at the beginning, initial samples are usually collected to build the format of (18), the objective function for our adaptive
the initial metamodel. This developed metamodel is then optimization can be defined as follows:
used to identify the input parameters that have the best 2
potential to lead to the expected output result. Due to the min
𝑥
MSE = (𝑓𝑚 (x) − 𝑇) + 𝜎(x)2 . (19)
errors of the metamodel, the actual output obtained from
experiment/simulation is usually different from the expected The optimization result of x∗ is then selected as the vector of
one. The previously obtained metamodel is subsequently input parameters for the (𝑚 + 1)th sample xm+1 . The output
updated to improve its quality using the new pair of input- 𝑌𝑚+1 corresponding to the xm+1 is subsequently obtained
output data. The method to iteratively modify the metamodel through experiment or simulation. The new pair of data,
through an iterative sampling process is called adaptive (xm+1 , 𝑌𝑚+1 ), together with all the previously collected sample
metamodeling. data are used to update the metamodel into a new relationship
𝑓𝑚+1 , as shown in (17).
We will formulate the specific problem for McPherson
3.2. Adaptive Metamodel-Based Optimization. When adap- suspension system using the adaptive metamodel-based opti-
tive metamodeling is used for optimization, the optimization mization considering prediction uncertainty in the coming
process can be referred to as adaptive metamodel-based opti- section.
mization [8]. The detailed algorithm for adaptive metamodel-
based optimization is introduced here and we call it adaptive
optimization in later parts of the paper. First the 𝑚 initial 4. Adaptive Optimization
samples with input parameters x𝑖 (𝑖 = 1, 2, . . . , 𝑚) and output considering Modeling Uncertainty
parameter 𝑌𝑖 (𝑖 = 1, 2, . . . , 𝑚) are collected to build the in McPherson Suspension
metamodel:
To investigate the performance of the designated suspension
𝑌 = 𝑓𝑚 (x) . (15) system, a classic McPherson suspension system is modeled
Mathematical Problems in Engineering 5

in ADAMS, as shown in Figure 2. The classic parallel wheel

travel suspension system analysis can be employed to do
system analysis. Vertical bound and rebound of 50 mm are
used. By changing the locations of each hardpoint, the
corresponding change of kinematic characteristics can be
obtained.
In this study, we take the locations of the 7 key hardpoints
labeled in Figure 1 as variables, 3 variables at each hardpoint,
that is, the coordinates of the hardpoint along axes 𝑥, 𝑦 and 𝑧
of the vehicle coordinate. Therefore we have 21 variables (v1–
v21) for the designated suspension system. Table 1 listed the
21 variables versus the hardpoint coordinate.
For the kinematic characteristics, as stated in Section 2,
we choose the camber angle, the caster angle, the kingpin
incline angle, and the toe angle as the objective parameters
to be designed and optimized. For McPherson suspension
system, small variations of the four estimate angles surely z
are more acceptable. Thus, we choose the four optimization x
y
objectives to be the minimum deviations of the four angles
versus wheel travel from rebound −50 mm to bound 50 mm. Figure 2: ADAMS model of the McPherson suspension system.
We thus have 21 design variables as input parameters and
4 optimum objectives as output parameters. The optimization
problem is clearly highly nonlinear and coupled on the input Table 1: The key hardpoints and corresponding variables.
variables. We thus employ the metamodeling method for
Number Hardpoint name Coordinates Variables
the optimal design. We clearly have several optimization
objectives; the usual way that people deal with multiobjective 1 LCA front 𝑥, 𝑦, 𝑧 V1, V2, V3
optimization problems is assigning weighting factors to each 2 LCA outer 𝑥, 𝑦, 𝑧 V4, V5, V6
objective and then adding them to build a single objective. 3 LCA rear 𝑥, 𝑦, 𝑧 V7, V8, V9
However, the predefined weighting factors may not be proper 4 Struct lwr 𝑥, 𝑦, 𝑧 V10, V11, V12
for the whole process of optimization. In our work, we only 5 Tierod inner 𝑥, 𝑦, 𝑧 V13, V14, V15
assign the same value to the weighting factors for the output 6 Tierod outer 𝑥, 𝑦, 𝑧 V16, V17, V18
parameters in the initial samples. Adaptive weighting factors 7 Top mount 𝑥, 𝑦, 𝑧 V19, V20, V21
are used in the process of adaptive optimization, which is
carried out as follows:
𝑌𝑖 = 𝑤1 𝑌1𝑖 + 𝑤2 𝑌2𝑖 + ⋅ ⋅ ⋅ + 𝑤𝑟 𝑌𝑟𝑖 𝑌𝑖𝑀 is, the larger weighting factor will be assigned to it in
(20) order to minimize all the output parameters at the same time.
𝑖 = 1, . . . , 𝑚, Consider
where 𝑟 is the number of output parameters and 𝑚 is number 󵄨󵄨 󵄨󵄨
󵄨󵄨𝑌𝑖𝑀󵄨󵄨
of initial samples. The 𝑚 initial samples with input parameters 𝑤𝑖𝑀+1 = 󵄨󵄨 󵄨󵄨 󵄨󵄨 󵄨 󵄨 󵄨
x𝑖 (𝑖 = 1, 2, . . . , 𝑚) and output parameter 𝑌𝑖 (𝑖 = 1, 2, . . . , 𝑚) 󵄨󵄨𝑌1𝑀󵄨󵄨 + 󵄨󵄨𝑌2𝑀󵄨󵄨󵄨 + ⋅ ⋅ ⋅ + 󵄨󵄨󵄨𝑌𝑟𝑀󵄨󵄨󵄨 (21)
are collected to build the metamodel and obtain the new
𝑖 = 1, . . . , 𝑟; 𝑀 = 50, . . . 𝑁 − 1,
group of input parameters using (19).
The initial metamodel is built on the data generated
from Latin hypercube sampling [20], which has the following where 𝑀+1 is the number of samples for each iteration in the
advantages: (1) its sample mean has a relatively smaller process of adaptive optimization, and 𝑁 is the total sample
variance compared with simple random sampling, (2) it can size. Consider
be used for generating design points when the number of
input variables is large and a great many runs are required, 𝑌𝑀+1 = 𝑤1𝑀+1 𝑌1𝑀 + 𝑤2𝑀+1 𝑌1𝑀 + ⋅ ⋅ ⋅ + 𝑤𝑟𝑀+1 𝑌𝑟𝑀. (22)
and (3) it is cheap in computing and easy for implementation
compared with other more complex sampling methods. By The new pair of data, (xM+1 , 𝑌𝑀+1 ), together with all
using ADAMS batch processing tools, the initial sets samples the previously collected sample data are used to update the
can easily be obtained. With the built initial metamodel, the metamodel into a new relationship. The optimization process
adaptive optimization method introduced in Section 3.3 can is stopped when the change of the objective function in
be used. several consecutive iterations is less than a predefined value,
The 4 output parameters corresponding to the new or the maximum number of iterations is reached.
group of input parameters are subsequently obtained through As there are 4 output parameters as optimizing objectives
simulations. The weighting factors 𝑤𝑖 are adjusted using (21) for our suspension system, (23) is used to transform the mul-
and the value of 𝑌𝑀+1 is calculated using (22). The larger the tiobjective optimization into single objective optimization.
6 Mathematical Problems in Engineering

Initial sampling for design parameters ADAMS simulation Multioutput parameters

Adaptive weighting
Single output parameters

Metamodel construction

Optimization

New group of design

parameters ADAMS simulation Multioutput parameters

Terminal
Adaptive weighting No
condition
satisfied

Single output parameter

Yes

New sample End

Figure 3: Flowchart of the proposed adaptive multiobjective optimization approach.

The adaptive weighting factors are obtained from (21). The

Value of objective

8
design objective is to minimize 𝑌 as in 6
function

4
𝑌 = 𝑤1 𝑌camber + 𝑤2 𝑌caster + 𝑤3 𝑌kingpin + 𝑤4 𝑌toe . (23) 2
0
The flowchart for the adaptive optimization process is given 0 10 20 30 40 50 60 70 80 90 100
in Figure 3. Trail number
output parameters

5
5. Comparisons and Analysis of the
Variations of

4
Optimization Results 3
2
1
The number of the initial sample is selected as 50 to build the 0
initial metamodel. 50 initial samples may not be enough for 0 10 20 30 40 50 60 70 80 90 100
a metamodeling problem with 21 variables; however, due to Trail number
the cost consideration in engineering design, we just initially Ycamber Ykingpin
choose 50 samples and gradually add new samples during Ycaster Ytoe
the design process. Based on the initial metamodel, more
trails are sampled sequentially and adaptively to approach Figure 4: Adaptive optimization results.
the optimum value. When the total sample size reaches 100,
the values of the output parameters are generally stable.
We analyze each of the 4 output parameters individually
every time while we update the metamodel. For this high
dimensional (21D input) problem, 100 samples may not distributions from 1 to 50 are obtained from Latin hypercube
be enough to reach the optimum value. However, we can sampling in the whole design space for building the initial
evaluate its effectiveness through its convergent trend. The metamodel. From Figure 4, we can see that corresponding
optimization results are shown in Figure 4. The value of the to the same design interval (−50, 50), the variation interval
objective function is well converged and the variations of the for the camber angle, the caster angle, the kingpin incline
4 output parameters are generally stable. angle, and the toe angle are (0, 2.5), (0, 4.5), (0, 1.5), and
The 4 types of curves with “∙”, “×”, “+”, and “o” indicate the (0, 4.5). While adaptive optimization is applied for the trails
camber angle variation, the caster angle variation, the kingpin from 51 to 100, the variation intervals decrease and the
incline angle variation, and the toe angle variation. The trail curves become smooth. The adaptive weighting factors help
Mathematical Problems in Engineering 7

Table 2: The initial and optimized hardpoints coordinates. 50

Offset from initial

Variables Initial Optimized 0
V1, V2, V3 160, −410, 265 195.7, −381, 234
V4, V5, V6 −80, −710, 165 −95, −710.6, 213 −50
V7, V8, V9 190, −395, 185 190.6, −366, 166 0 10 20 30 40 50 60 70 80 90 100
Trail number
V10, V11, V12 50, −630, 530 64.6, −622, 531.3
V13, V14, V15 155, −440, 355 130, −391.2, 305 Tierod-inner x
Tierod-inner y
V16, V17, V18 210, −700, 240 219.5, −749, 290
Tierod-inner z
V19, V20, V21 7.5, −603.8, 850 35.2, −642, 873.6
50

Offset from initial

0
the output parameters with different variation intervals con-
verge to the same variation interval (0, 0.33). It can be seen −50
0 10 20 30 40 50 60 70 80 90 100
from Figure 4 that our method succeeds in optimizing the
Trail number
4 output parameters simultaneously while the value of the
objective function is well converged. Table 2 listed the initial Tierod-outer x
and the optimized hardpoints locations for the suspension Tierod-outer y
system. We have chosen the 100th sample as the optimized Tierod-outer z
result. Figure 5: The evolution trend of the hardpoint tierod inner and
As to the regular adaptive optimization, prediction stan- tierod outer.
dard deviation is not considered in the process of the
optimization, and the objective function is usually defined
as in (16). We used the same 50 initial samples to build the
initial metamodel and then apply the objective function given the optimization procedure, and other hardpoints perform
in (16) to search the subsequent 50 samples. We compare similarly. From this figure, it can be seen that the 3 coordinate
the mean output values of the last 5 samples for each of the values for each input parameter tend to change in a smaller
4 output parameters from optimization objective functions interval rather than change in the original design interval
defined by (19) and (16), as shown in Table 3. The mean output (−50, 50). This result from our research can help to reduce
values are compared by a ratio, 𝛼, representing how much the design space in the process of adaptive optimization,
the adaptive optimization method considering uncertainty is in order to greatly improve the optimization efficiency,
better than the regular adaptive optimization method. Here, which would be especially significant for higher dimensional
problems.
Varmean − Varmean(un)
𝛼= , (24) The apparent result that the optimization procedure can
Varmean(un) achieve is reducing the variation of the design objectives.
Figure 6 gives the variations of the camber angle, the caster
where Varmean indicates the mean output value in the last angle, the kingpin incline angle, and the toe angle, versus the
5 iterations from regular adaptive optimization method, wheel travel for the optimized trail number 100 and the initial
and Varmean(un) gives the mean output value in the last 5 trail during the parallel suspension travel analysis. We can see
iterations from adaptive optimization method considering that the optimized data significantly reduced the variation of
the modeling uncertainty. the camber angle, the kingpin incline angle, and the toe angle,
From Table 3, we could see that the adaptive optimiza- while also slightly reducing the caster angle. This shows the
tion method with consideration of modeling uncertainty effectiveness of the proposed approach on guiding to search
performs better. The adaptive optimization performs better for the optimized solution for kinematic performance design
up to 81.11% for all the four output parameters. The reason of vehicle suspension systems. Other related parameters can
should be that the number of sample points is far from be considered in a similar way.
sufficient to build an accurate metamodel, especially at the
very beginning. Thus the optimization result based on the
built metamodel only considering predicted mean value may 6. Conclusion
not converge to the right direction. With the prediction
standard deviation being considered at the same time, the This study introduced adaptive metamodel-based optimiza-
inaccuracy of the metamodel can be compensated to some tion considering modeling uncertainty to optimize the kine-
extent to approach the right direction. matic performance of a McPherson suspension system. The
What is interesting is that the approach can illustrate optimization design problem is of 21 input parameters and
a suggested trend that the hardpoints should be moved 4 output parameters. The multioutput optimization in the
over to. Figure 5 shows the approximate trends that the McPherson suspension system is transformed to a single out-
hardpoints tierod inner and tierod outer stay around during put optimization problem using adaptive weighting factors.
8 Mathematical Problems in Engineering

Table 3: Comparison between adaptive optimization considering modeling uncertainty and regular adaptive optimization.

Camber angle Caster angle Kingpin angle Toe angle

Optimization objective function
variation variation variation variation
Adaptive optimization considering modeling uncertainty 0.1056 0.0916 0.3044 0.0577
Regular adaptive optimization 0.1616 0.1585 0.4717 0.1045
𝛼 53.03% 73.03% 54.96% 81.11%

2.5 11.5

1.5
Camber angle (deg)

9.8333

Caster angle (deg)

1

0.5

0
8.1667
−0.5

−1

−1.5 6.5
−50 −40 −30 −20 −10 0 10 20 30 40 50 −50 −40 −30 −20 −10 0 10 20 30 40 50
Wheel travel (mm) Wheel travel (mm)
(a) (b)

13.5 6

12.625
4.3333
11.75
Kingpin incline angle (deg)

2.6667
Toe angle (deg)

10.875

10 1

0
9.125
−0.6667
8.25
−2.3333
7.375

6.5 −4
−50 −40 −30 −20 −10 0 10 20 30 40 50 −50 −40 −30 −20 −10 0 10 20 30 40 50
Wheel travel (mm) Wheel travel (mm)
Initial Initial
Metamodel oriented Metamodel oriented
(c) (d)

Figure 6: The variation of camber angle, caster angle, kingpin incline angle, and toe angle versus wheel travel before and after the optimization.

50 initial samples and 50 sequential trails are generated to design trends of the hardpoints are given as the trail goes
analyze the convergent trend of the objective parameters. It on. Comparisons showed that the adaptive metamodel-
shows that the proposed optimization method provided a set based optimization method considering modeling uncer-
of relatively good results of the 4 output objective parameters tainty worked better than general adaptive metamodel-based
simultaneously for the 50 sequential trails. Possible optimal optimization for the suspension design problem.
Mathematical Problems in Engineering 9

Conflict of Interests target cascading method,” International Journal of Automotive

Technology, vol. 13, no. 1, pp. 109–122, 2012.
The authors declare that there is no conflict of interests [14] D. Zhao and D. Xue, “A comparative study of metamodeling
regarding the publication of this paper. methods considering sample quality merits,” Structural and
Multidisciplinary Optimization, vol. 42, no. 6, pp. 923–938, 2010.
Acknowledgment [15] G. Matheron, “Principals of geostatistics,” Economic Geology,
vol. 58, no. 8, pp. 1246–1266, 1963.
This work is supported by NSFC Grant no. 11302074. [16] J. Sacks, W. J. Welch, T. J. Mitchell, and H. P. Wynn, “Design and
analysis of computer experiments,” Statistical Science, vol. 4, no.
4, pp. 409–435, 1989.
References [17] S. N. Lophaven, H. B. Nielsen, and J. Søndergaard, “DACE: A
[1] M. Blundell and D. Harty, The Multibody Systems Approach to Matlab Kriging Toolbox, Version 2.0,” Tech. Rep. IMMREP-
Vehicle Dynamics, Maple-Vail, Kirkwood, NY, USA, 2004. 2002-12, Technical University of Denmark, Kongens Lyngby,
Denmark, 2002.
[2] M. Zapateiro, F. Pozo, H. R. Karimi, and N. Luo, “Semiactive
control methodologies for suspension control with magne- [18] R. Ding, D. K. J. Lin, and D. Wei, “Dual-response surface
torheological dampers,” IEEE/ASME Transactions on Mecha- optimization: a weighted MSE approach,” Quality Engineering,
tronics, vol. 17, no. 2, pp. 370–380, 2012. vol. 16, no. 3, pp. 377–385, 2004.
[19] D. K. J. Lin and W. Tu, “Dual response surface optimization,”
[3] H. Li, X. Jing, and H. R. Karimi, “Output-feedback-based 𝐻∞
Journal of Quality Technology, vol. 27, no. 1, pp. 34–39, 1995.
control for active suspension systems with control delay,” IEEE
Transactions on Industrial Electronics, vol. 61, no. 1, pp. 436–446, [20] M. D. McKay, R. J. Beckman, and W. J. Conover, “A comparison
2013. of three methods for selecting values of input variables in the
analysis of output from a computer code,” Technometrics, vol.
[4] H. Zhang, Y. Shi, and A. S. Mehr, “Robust 𝐻∞ PID con-
21, no. 2, pp. 239–245, 1979.
trol for multivariable networked control systems with distur-
bance/noise attenuation,” International Journal of Robust and
Nonlinear Control, vol. 22, no. 2, pp. 183–204, 2012.
[5] R. Wang, H. Zhang, and J. Wang, “Linear parameter-varying
controller design for four-wheel independently actuated electric
ground vehicles with active steering systems,” IEEE Transactions
on Control Systems Technology, 2013.
[6] Z. Shuai, H. Zhang, J. Wang, and J. Li, “Combined AFS and DYC
control of four-wheel-independent-drive electric vehicles over
CAN network with time-varying delays,” IEEE Transactions on
Vehicular Technology, 2013.
[7] H. Zhang, J. Wang, and Y. Shi, “Robust 𝐻∞ sliding-mode
control for Markovian jump systems subject to intermittent
observations and partially known transition probabilities,” Sys-
tems & Control Letters, vol. 62, no. 12, pp. 1114–1124, 2013.
[8] G. G. Wang and S. Shan, “Review of metamodeling techniques
in support of engineering design optimization,” Journal of
Mechanical Design, Transactions of the ASME, vol. 129, no. 4, pp.
370–380, 2007.
[9] P. Zhu, Y. Zhang, and G.-L. Chen, “Metamodel-based
lightweight design of an automotive front-body structure
using robust optimization,” Proceedings of the Institution of
Mechanical Engineers D, vol. 223, no. 9, pp. 1133–1147, 2009.
[10] H. H. Chun, S. J. Kwon, and T. Tak, “Reliability based design
optimization of automotive suspension systems,” International
Journal of Automotive Technology, vol. 8, no. 6, pp. 713–722,
2007.
[11] B.-L. Choi, J.-H. Choi, and D.-H. Choi, “Reliability-based
design optimization of an automotive suspension system for
enhancing kinematic and compliance characteristics,” Interna-
tional Journal of Automotive Technology, vol. 6, no. 3, pp. 235–
242, 2005.
[12] D. O. Kang, S. J. Heo, and M. S. Kim, “Robust design optimiza-
tion of the McPherson suspension system with consideration of
a bush compliance uncertainty,” Proceedings of the Institution of
Mechanical Engineers D, vol. 224, no. 6, pp. 705–716, 2010.
[13] D. O. Kang, S. J. Heo, M. S. Kim, W. C. Choi, and I. H. Kim,
“Robust design optimization of suspension system by using