12814 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO.

10, OCTOBER 2023

Optimal Intelligent Chassis Layout Design

Framework Based on Particle Swarm Optimization
and Robust Finite-Frequency H∞ Control
Yongkang Zhang , Lei Wang, Hui Zhang , Senior Member, IEEE, and Makoto Iwasaki , Fellow, IEEE

Abstract—Unmanned special vehicles (USVs) that adopt the brings difficulties to design and control. Therefore, it is of great
automatic control method to achieve acceleration, deceleration and importance to properly arrange the position of the equipment on
steering put forward higher requirements for lateral stability. The the chassis [2].
USVs chassis equipped with heavy work equipment and support
equipment will change the center of gravity (c.g.) and significantly As the basic component of a vehicle, the chassis provides
affect lateral stability. In this work, we aim to come up with an support for the engine, gearbox, steering gear, braking
optimal chassis layout design framework to arrange the various mechanism and equipment [3]. Researchers have been paid
equipment for the intelligent chassis from the control point of view. considerable effort in the field of chassis design, including
An algorithm combining the Particle Swarm Optimization (PSO)
chassis frame design and weight reduction. In chassis frame
and enumeration method is developed to derive the candidate
USVs chassis layout schemes. Candidate schemes are then filtered design, Palanivendhan et al. [4] utilized Computer-Aided Design
based on steering characteristics and Lyapunov stability. Finally, (CAD) software to develop a three-wheeled chassis which
the selection method based on robust finite-frequency H∞ control reduces the chance of rollover and improves stability due to the
is proposed to obtain the final scheme with good lateral stability. lower center of gravity. Chen et al. [2] built up a knowledge base
Simulation results validate that the yaw rate response of the final
system for the rapid design of harvesting machinery chassis.
USVs chassis layout scheme is close to the two-degree-of-freedom
(2-DOF) chassis dynamic model compared with other schemes un- According to the safety constraints specified in the rules,
der the front wheel steering angular step signal input and sinusoidal Mohammed et al. [5] designed the chassis of formula car chassis
signal input. by Solidworks under the premise of ensuring driver safety and
Index Terms—Chassis layout design, particle swarm lightweight. Abdullah et al. [6] developed a narrow track
optimization (PSO), linear variable parameter (LPV) system, vehicle chassis for traffic congestion considering the effects
robust finite-frequency H∞ control. of engine and driver loads on the chassis structure. Parlaktas
et al. [7] proposed a new method to design and manufacture
I. INTRODUCTION electric vehicle chassis and its structure at a low cost. These
studies designed the chassis structure and verified the chassis
PECIAL vehicles built for special purposes such as rescue,
S agricultural production and facility maintenance have been
widely used [1]. The unmanned special vehicles with autopilot
performance from different aspects. Although some studies
have considered the effect of load on chassis strength, the effect
on chassis lateral stability has not been examined. Besides, these
system working in specific scenarios have begun being used in design objects are almost for the chassis with a human driver, and
various industries due to their simple working environment and there are few works for the unmanned chassis. In addition, there
have extensive application prospect. In many applications, these is not much research on the chassis layout design for special
USVs chassis are required to carry various equipment with heavy vehicles.
weight and large volume, which lead to the c.g. of these chassis Another aspect of chassis design is weight reduction. Manto-
being relatively high and may change with different equipment vani et al. [8] used the lattice optimization method to optimize
displacement, having a significant effect on lateral stability. the detailed size of the chassis. Cavazzuti et al. [9] adopted
Besides, the speed of the USVs chassis is time-varying, which the topology optimization method to carry out the lightweight
design of chassis of spider and coupe vehicles. Cavazzuti et
Manuscript received 9 August 2022; revised 9 March 2023 and 9 May 2023; al. [10] combined size optimization, topometry and topology
accepted 9 May 2023. Date of publication 15 May 2023; date of current version with finite element analysis and attained an optimum chassis
17 October 2023. This work was supported by Defense Industrial Technology
Development Program. The review of this article was coordinated by Prof. Wei configuration. Nandhakumar et al. optimized the weight of the
Ni. (Corresponding author: Hui Zhang.) existing electric bus chassis frame by replacing its steel compo-
Yongkang Zhang, Lei Wang, and Hui Zhang are with the School of nents with potential aluminum alloys [11]. The lightweight tech-
Transportation Science and Engineering, Beihang University, Beijing 100191,
China (e-mail: [email protected]; [email protected]; huizhang nology of automobile chassis can improve power performance
[email protected]). and fuel economy by reducing chassis mass. Although the center
Makoto Iwasaki is with the Department of Electrical and Mechanical En- of gravity of the chassis can be lowered by weight reduction, this
gineering, Nagoya Institute of Technology, Nayoga 466-8555, Japan (e-mail:
[email protected]). technology has little effect on improving the lateral stability of
Digital Object Identifier 10.1109/TVT.2023.3275959 the vehicle.
0018-9545 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12815

Besides, the safety of the designed chassis is also an important 2) We came up with the PSO algorithm with enumeration
factor that needs to be considered [12]. Chien et al. [13] pro- method to obtain the USVs chassis layout schemes. The
posed a chassis control strategy that integrates feed-forward and fitness function and penalty function are designed to ob-
feedback control schemes to improve the stability and safety of tain the chassis candidate schemes to satisfy the space
the chassis. Soheib et al. [14] proposed a smooth coordination constraints and meet the designer’s requirements.
strategy between braking, steering and suspension to improve 3) The selection method based on robust finite-frequency H∞
vehicle dynamics behavior by LPV/H∞ framework. Guo et control is developed to obtain the final scheme with good
al. [15] used a nonlinear model predictive control method based lateral stability in which the yaw rate response is close to
on active front steering and additional yaw moment to control the 2-DOF chassis dynamic model compared with other
the chassis. Based on the nonlinear vehicle model, Taghavifar et schemes. We have improved the method for calculating
al. [16] combined the optimal robust control and the least squares the H∞ performance for the LPV model, which reduces
neural network to improve the stability of the chassis. Perozzi et the conservatism of the results.
al. [17] and Li et al. [18] designed the controller based on lateral The organization of this article is as follows. Section II illus-
stability, which improved the control performance of the chassis, trates the problem of chassis layout design, and the LPV chassis
respectively. Zhang et al. [19] proposed a gain-scheduling fault dynamic model is established. The acquisition method of candi-
detector and an augmented system based on the uncertain LPV date chassis layout schemes and the final chassis layout scheme
model for the electric ground vehicle. The control strategies filtering method are explained in Section III. In Section IV,
proposed by these studies are designed based on existing chassis, simulations based on front wheel steering angle step signal
ignoring the influence of the control on the chassis layout design input and front wheel sinusoidal signal input response verify
stage. the effectiveness of the proposed chassis design framework.
The USVs achieve different tasks by installing different Conclusions are provided in Section V.
working equipment on a universal chassis. For instance, the
residual rubber on the airport runway needs to be cleaned II. PROBLEM FORMULATION
regularly to restore the friction coefficient to a normal level.
In this section, we first illustrate the problem of chassis layout
The rubber removal trucks need to carry some different equip-
design and explains how different layout schemes affect the
ment such as ultra-high-pressure water systems, high voltage
parameters of the USVs chassis dynamic model. Then, accord-
battery, diesel generators, etc. However, due to the large size
ing to the characteristic of the USVs chassis, the LPV chassis
and heavy weight (even exceeding the mass of the chassis),
dynamic model is established. Finally, three principles of chassis
these equipment will result in a high center of gravity and
layout design are proposed.
uneven load distribution of the chassis, which will greatly affect
the lateral stability and enhance the control difficulty of the
chassis. A. Chassis Layout Illustration
How to rationally arrange some different equipment on the In this section, the problem of chassis layout design is il-
chassis under the premise of satisfying the space constraints and lustrated. Taking the airport maintenance USVs as an example,
design requirements is a problem in the field of vehicle chassis the USVs chassis consists of two parts: the universal chassis
design technology. The traditional chassis design method has and equipment. The universal chassis consists of a drive system,
the problem that the number of candidate schemes is small and steering system, braking system, chassis frame, etc. Since the
the layout only depends on the designer’s experience, which upper plane of the chassis frame is flat, the equipment can be
will make it difficult for the designed scheme to have good arranged and installed on it. The equipment consists of a working
lateral stability. In this article, we come up with a novel op- device, electrical control center, diesel generator, high-voltage
timal chassis layout design framework to address this problem. battery, etc. Note that to save space, the battery is generally
The contribution to vehicular technology is that the framework installed in the battery compartment located at the lower plane
provides an automatic way for the layout design of the USVs of the chassis frame.
chassis without relying on manual work. Specifically, on the The layout of the USVs chassis is shown in Fig. 1. Fig. 1(a)
one hand, the framework can obtain more candidate schemes and (b) are the top and bottom view of the chassis, respectively.
that satisfy the chassis space constraints, which means a higher Taking the geometric center of the chassis as the coordinate
probability of obtaining the optimal layout solution. On the other origin, the longitudinal velocity direction of the chassis as the
hand, the method of filtering the optimal layout scheme based positive direction of the X axis and a Cartesian coordinate system
on the premise of improving lateral stability is proposed. To the (O-X-Y) is established. In Fig. 1, the shaded rectangles 1
authors’ knowledge, there is not a lot of research on this so far. 2 3 4 represent different equipment. Rectangles 1 2
The more detailed contributions of this work are summarized as 3 4 represent the working device, electrical control center,
follows: generator and high-voltage battery, respectively. We refer to the
1) We proposed a novel optimal chassis layout design frame- rectangle i (i = 1, 2, 3, 4) as different equipment in the rest of
work considering the chassis lateral stability. To the this article. Besides, rectangle 6 denotes chassis frame and the
author’s knowledge, there is not a lot of work since rectangles 1 2 3 are arranged above. Rectangle 5 stands
most chassis layout design methods rely mainly on the for the safety officer’s seat and no rectangles can be placed in this
designer’s personal experience. This will not guarantee location. The weight of the rectangle 5 is negligible. On behalf
the design of a chassis with good lateral stability. of battery compartment, rectangle 7 is arranged at the lower
Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12816 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

Fig. 2. 2-DOF chassis dynamic model.

The total yaw moment of inertia of the USVs chassis Iz is the

sum of the yaw moment of inertia of the chassis frame and the
yaw moment of inertia of all rectangles around the geometric
center of the chassis, which can be calculated by the following
equation:

n
Iz = Ichs + (Ii + mi ρ2i ) (4)
i=1

where Ichs denotes the yaw moment of inertia of the chassis,

which can be obtained by Eq. (3), ρi denotes the distance from
the center of the rectangle i to the geometric center of the USVs
chassis, and n denotes the number of rectangles to arrange.
Fig. 1. Layout of USVs chassis. (a) Top view of chassis. (b) Bottom view of The corning stiffness is an important characteristic of the
chassis. tire which also affects the lateral stability of the chassis. The
vertical load of the tire affects the cornering stiffness, and it
first increases and then decreases as the vertical load increases.
plane of the chassis frame and the rectangle 4 is arranged below The relationship between the cornering stiffness and the vertical
the chassis within the rectangle 7. Due to the large mass and load is illustrated in [20] by experimental analysis. The cornering
volume of these equipment, different layout of these rectangles stiffness of the tire can be obtained by fitting the experimental
will affect the load distribution of the chassis. Different load results in [20], and the result are shown as follows:
distribution will then cause the c.g. of USVs chassis changes 
cf = 2(ε3 G3f + ε2 G2f + ε1 Gf + ε0 )
and affect its lateral stability. (5)
Let li , wi , hi denote the length, width and height of the rect- cr = 2(ε3 G3r + ε2 G2r + ε1 Gr + ε0 )
angle i, respectively. mi denotes the mass of the rectangle i. where cf and cr are the cornering stiffness of the front and rear
xi and yi denote the value of the geometric center of rectangle wheels of the chassis, respectively. The εi , i = 0, 1, 2, 3 are the
i in the (O-X-Y) coordinate at X axis and Y axis, respectively. fit coefficients. The vertical loads Gf and Gr of the front and
The total mass of the USVs chassis m can be calculated by the rear axles can be calculated by:
following equation: 
Gf = mglr /L

n (6)
m = mchs + mi (1) Gr = mglf /L
i=1
where g is the acceleration of gravity.
where n is the number of rectangles to be arranged and mchs is
the mass of universal chassis. The distance from c.g. to the front
B. LPV Chassis Dynamic Model
and rear axles can be calculated as follows:
⎧ 
n This section describes the vehicle’s lateral dynamics and the
⎨ m i xi
parameters used in it. Since the speed of the USVs chassis is
lf = i=1 m , n = 4 (2)
⎩ time-varying, an LPV chassis dynamic model is established to
lr = L − lf improve the accuracy of the model. In order to study the lateral
where lf and lr respectively denote the longitudinal distance dynamic and stability of the USVs chassis, we adopt the 2-DOF
from c.g. to front and rear tires; L denotes the wheelbase of the chassis dynamic model as the ideal chassis dynamic model [21],
USVs chassis. [22], [23].
When calculating the yaw moment of inertia of the USVs Fig. 2 shows the 2-DOF chassis dynamic model of the USVs
chassis, to simplify the calculation, the chassis and all rectangles chassis. The c.g. of the chassis is CG. Taking CG as the coordi-
are simplified as cuboids. The yaw moment of inertia of rectangle nate origin and the longitudinal velocity direction of the chassis
i about its geometric center is: as the positive direction of the X-axis, a Cartesian coordinate
system (CG-X-Y) is established. δ is the front wheel steering
Ii = 12 mi (li
1 2
+ wi2 ) (3) angle. αf and αr represent the slip angle of the front and rear

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12817

The value of f and f can be calculated as follows:

 √ 
f = min − (1+Kv vx
2
)L
, − K
2K x (9)
f =0
The LPV form of (7) is established as follows [24]:

6
(A, B, E, F ) = αi (Ai , Bi , Ei , Fi ) (10)
n=1

where

(A1 , B1 , E1 , F1 )
 −cf −cr lr cr −lf cf   c  
mv x mv 2x
− 1 f

= lr cr −lf cf −lf2 cf −lr2 cr lf cf , E, f

, mv x
,
Fig. 3. Eigenvalue of matrix A in complex plane with the vx increasing.
Iz Iz v x Iz

(A2 , B2 , E2 , F2 )
wheels, respectively. Fyf and Fyr denote the lateral tire force of −cf−cr vx+v   c  
lr cr−lf cf
the front and rear wheels, respectively. v is the velocity of the m × 2v x v x
x
mv x v x −1
f

chassis, vx is the longitudinal velocity of the chassis, and vy is = lr cr−lf cf −lf2 cf−lr2 cr v +v
, lf cfx , E, f ,
mv
× 2vxx v x
the lateral velocity of the chassis. ω is the yaw rate of the chassis, Iz Iz x
Iz

β is the slip angle of the chassis, and z(t) denotes the yaw rate (A3 , B3 , E3 , F3 )
error between the chassis dynamic model and the ideal 2-DOF  −cf −cr   c  
lr cr −lf cf
chassis dynamic model. In order to study the lateral stability of mv x mv 2x − 1 f

the chassis, the slip angle of c.g. and yaw rate are selected as = lr cr −lf cf −lf2 cf −lr2 cr , lf cf , E, f ,
mv x

Iz Iz v x Iz
the state variables, and then the 2-DOF chassis dynamic model
is established as follows: (A4 , B4 , E4 , F4 )
  −cf −cr   c  
ẋ(t) = Ax(t) + Bδ lr cr −lf cf
mv 2x
− 1 f
(7) mv x
, mv
= lf cf , E, f ,
x
z(t) = Ex(t) + F δ lr cr −lf cf −lf2 cf −lr2 cr
Iz Iz v x Iz

where (A5 , B5 , E5 , F5 ) =
   c 
β f  −cf −cr v +v lr cr −lf cf   c  
x(t) = , B = mv × 2vx x v x mv x v x − 1
lf c f , E = 0 1
f
x m
ω lr cr −lf cf
x
−lf2 cf −lr2 cr v +v
, mv
lf cf , E, f
x
,
Iz × 2vx x v x
 −cf −cr lr cr −lf cf
 Iz Iz x
I z

−1
mvx
A = lr cr −lf cf mvx 2
(A6 , B6 , E6 , F6 )
−lf2 cf −lr2 cr
Iz Iz vx  −cf −cr lr cr −lf cf   c  
mv x mv 2x − 1 f

vx m lf lr = lr cr −lf cf −lf2 cf −lr2 cr , lf cf , E, f ,

mv x
F =− ,K = 2 − .
(1 + Kvx2 )L L cr cf Iz Iz v x Iz


6
The longitudinal velocity vx varies during the chassis running, αi = 1, 0 ≤ αi ≤ 1.
and the model is nonlinear. As shown in Fig. 3, when the value i=1
of vx increases, the eigenvalues of the state matrix gradually
move toward the imaginary axis, resulting in the deterioration The polyhedron of the space-model (Ω1 Ω2 Ω3 Ω1 Ω2 Ω3 ) es-
of the stability of the system. However, the USVs only work tablished by (8) and (10) is shown in Fig. 4 [25]. Although the
for particular task such as cleaning rubber, removing snow longitudinal speed vx varies with time, the dynamic model of the
or weeding, etc., which make the chassis work in a certain USVs chassis is always in the space enclosed by the polyhedron.
range: According to the polyhedron, the LPV chassis dynamic model
⎧ with time-varying longitudinal speed can be rewritten as:
⎨v x ∈ v vx ⎧
x ⎪ 
6
(8) ⎪
⎨ẋ(t) = αi (Ai x(t) + Bi δ) = A(α)x(t) + B(α)δ
⎩F ∈ f f i=1
⎪
⎪ 
6
⎩z(t) = αi (Ei x(t) + Fi δ) = E(α)x(t) + F (α)δ
where v x and v x are the lower and upper bounds of vx , respec- i=1
tively. f and f are the lower and upper bounds of F , respectively. (11)

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12818 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

Fig. 5. Optimal chassis layout design framework.

Fig. 4. Polyhedron geometry diagram of the state-space model. rate of the 2-DOF chassis dynamic model compared with
other candidate schemes.
where αi (i = 1, 2, 3, 4, 5, 6) are the weight factors that vary with
the change of the chassis state, which can be calculated according III. DESIGN OF THE USVS CHASSIS LAYOUT
to the current longitudinal velocity of the chassis [24]. The LPV In this framework, the USVs chassis in Fig. 5 is the design
chassis dynamic model (A, B, E, F ) is a weighted sum of six objective we already have. We aim to arrange the equipment
vertices. Ωc is one of the possible choices of the chassis layout in this chassis for better lateral stability as well as satisfy the
scheme, which state is within the polyhedron. Ωc1 and Ωc2 are space constraints. As shown in Fig. 5, the whole framework
the projection of Ωc on plane Ω1 Ω2 Ω3 and Ω1 Ω2 Ω3 , respectively. is composed of three parts: Chassis Model Part, the Candidate
Ωc3 is the projection of Ωc on Ω1 Ω1 . Each αi is a time-varying Schemes Obtain Part and the Schemes Filtering Part. According
parameter related to the longitudinal velocity of the chassis, and to the existing USVs chassis, the space constraints and the
the αi , i = 1, . . . , 6 can be calculated as follows: chassis parameters can be obtained by means of meter stick
DΩ2 Ωc1 DΩcs Ω3 DΩ1 Ωc3 measurement. Based on the chassis parameters, we can derive
α1 = × × the USVs 2-DOF chassis model, and the USVs LPV chassis
DΩ2 Ωcs DΩ1 Ω3 DΩ1 Ω1
model can be derived according to the time-varying parameter
DΩcs Ωc1 DΩ1 Ωc3 vx . The space constraints then send to the Candidate Schemes
α2 = × Obtain Part to get more candidate schemes that satisfy the space
DΩ2 Ωcs DΩ1 Ω1
constraints and designer’s requirements. The LPV chassis model
DΩ2 Ωc1 DΩ1 Ωcs DΩ1 Ωc3 and the candidate schemes are used by the Schemes Filtering
α3 = × ×
DΩ2 Ωcs DΩ1 Ω3 DΩ1 Ω1 Part to get the optimal scheme. In the Schemes Filtering Part,
a three-level is established to get the final optimal scheme with
DΩ2 Ωc1 DΩcs Ω3 DΩ1 Ωc3
α4 = × × good lateral stability.
DΩ2 Ωcs DΩ1 Ω3 DΩ1 Ω1
DΩcs Ωc1 DΩ1 Ωc3 A. Obtaining Candidate USVs Chassis Layout Schemes
α5 = ×
DΩ2 Ωcs DΩ1 Ω1 The main task of this article is to select an appropriate chas-
DΩ2 Ωc1 DΩ1 Ωcs DΩ1 Ωc3 sis layout scheme based on the lateral stability of the USVs
α6 = × × (12) chassis. The first step is obtaining the candidate USVs chassis
DΩ2 Ωcs DΩ1 Ω3 DΩ1 Ω1
layout schemes that satisfy space constraints and other special
According to the chassis space constraints and H∞ perfor- requirements. The traditional method selects the scheme manu-
mance of the LPV chassis dynamic model, the selected chassis ally according to the space constraints, which mostly rely on
layout needs to meet: the individual’s experience [26]. There are some drawbacks
1) The final USVs chassis layout scheme should satisfy space to the traditional method. Firstly, it is difficult to manually
constraints such as equipment does not overlap and exceed list more possible layout schemes, resulting in the omission
chassis boundaries; of better layout schemes. Secondly, it is a challenge for de-
2) The 2-DOF dynamic model of the USVs chassis under the signers to arrange the equipment while satisfying the multiple
final USVs chassis layout scheme is asymptotically stable constraints.
and satisfy understeer characteristics; Similar to the facility layout problem, arranging the equip-
3) The yaw rate of the LPV chassis dynamic model under the ment of different sizes to an irregular USVs chassis is conformed
final USVs chassis layout scheme is close to the ideal yaw to NP-hard combinatorial problems [27]. There are three types

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12819

of optimization approaches to solve the facility layout problem, speed in the PSO at kth iteration. So one particle (pr,k ) repre-
the mathematical programming approach, the neural network sents a USVs chassis layout scheme.
approach and the heuristic approach [28]. Due to the complex 2) Update Particle Velocity and Position: The velocity of the
constraints, and nonlinearity brought about by irregular chassis particles is calculated by the historical optimal position of indi-
and different equipment sizes, it is difficult to describe and vidual particles and the global optimal position of particles. The
solve with the optimization method in optimization theory. The new position of the particles is obtained by the current position
final solutions obtained by the mathematical programming ap- and the current speed of the particles. These two equations show
proaches are sensitive to the initial layouts. The neural network the update mechanism of the particles and shown as follows:
approaches require accurate training data to achieve better re-
sults. The heuristic approaches such as genetic algorithm (GA), v r,k+1 = lr0 ∗ v r,k + lr1 ∗ rand()[pr,k
lb − p
r,k
]
ant colony algorithm (ACA), and PSO algorithm are also used + lr2 ∗ rand()[pr,k
gb − p
r,k
] (14)
to solve the problem of layout optimization. Due to the simple
structure, ease of implementation, speed to get the solutions pr,k+1 = pr,k + lr3 ∗ v r,k (15)
and robustness, the PSO algorithm shows good performance in
solving layout optimization effectively and efficiently compared where v r,k+1 is the updated speed of the single particle at
with other algorithms [29]. Many researchers have used the PSO (k + 1)th iteration, lr0 is inertia weight, lr1 , lr2 are all speed
method to solve the facility layout problem [30]. Inspired by update coefficient, lr3 is position update coefficient. pr,k
lb is the
this, we modified the PSO algorithm to be more suitable for historical optimal position of individual particle at kth itera-
our problem. The methodological contribution mainly includes tion and pr,k
gb is the global optimal position of particles at kth
two aspects. On the one hand, we modified the PSO objective iteration.
function to keep the equipment as far apart from each other as 3) Design of Objective Function and Fitness Function: The
possible under the premise of satisfying the constraints. This objective function of the PSO algorithm in this task consists of
allows for adequate spacing between equipment for servicing the target part and the constraints part. The target part represents
and cooling. On the other hand, through algorithm iteration, the the expectations of designers and requirements for the chassis
PSO method makes the layout scheme represented by particles layout. The objective function and constraints are defined as
move toward the direction that satisfies constraints and satisfy follows [30]:
requirements. This method can get more candidate solutions that

D 
D
 i   D
satisfy the requirements and improves the probability of finding max.  p − pj  + (pix max(wc − pix , 0)
the optimal scheme. i=1 j=i+1 i=1
In this section, we introduce the PSO algorithm with enu-
meration method to obtain the candidate layout schemes of the + piy max(hc − piy , 0)) (16)
chassis. The PSO algorithm is redesigned to derive the locations st. Xij Yij = 0
of rectangle 1 2 3 in Fig. 1(a). The enumeration method
is used to obtain the location of rectangle 4 in Fig. 1(b) and wi /2 − pix ≤ 0
generate the candidate chassis layout schemes combing with
hi /2 − piy ≤ 0
the result of PSO algorithm. The algorithm is described in the
following sections. xi − (wc − wi /2) ≤ 0
1) Chassis Layout Schemes Representation: Let P denote
the number of particles. N indicates the number of iterations. D yi − (hc − hi /2) ≤ 0 (17)
means the number of rectangles. Each particle means a chassis where
layout scheme represent by a 2D vector. In this vector, the
first D elements represent the X-axis value of the rectangle
positions. The last D elements represent the Y-axis value of Xij = max{(wi + wj )/2 − |xi − xj | , 0}
the rectangle positions. The speed of a single particle is also Yij = max{(hi + hj )/2 − |yi − yj | , 0}
denoted by a 2D vector, in which the first D elements represent
the speed component at X-axis and the last D elements repre- pi is the position of rectangle i in the particle. In (16), |pi −
sent the speed component at Y-axis of the rectangles. There- p | represent the distance between rectangle i and rectangle j
j

fore, each particle and its speed are denoted by the following in the particle. pix and piy are the X-coordinate and Y-coordinate
formula [31]: of center of rectangle i in the particle, respectively. wc and hc
 are the width and height of the upper plane of the chassis frame,
pr,k = (x1 , . . . , xD , y1 , . . . , yD ) respectively. wi and hi are the width and height of the rectangle
(13)
v r,k = (vx1 , . . . , vxD , vy1 , . . . , vyD ) i, respectively. Eq. (17) specifies the space constraints of the
rectangles, which the rectangles can not overlap each other and
where the pr,k denotes a particle in the PSO algorithm, which exceed the range of theDplane oni the jchassis frame.
components are the X-coordinates and Y-coordinate of the rect- In (16), the D i=1 j=i+1 |p − p | stands for the sum of the
angles. r and k denotes the rth particle in the PSO algorithm distances between rectangles. The larger this value indicates the
at kth iteration, respectively. The v r,k denotes the rth particle greater the spacing between rectangles, which is beneficial for

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12820 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

both
D the imaintenance and heat dissipation of the rectangles. The
Algorithm 1: PSO Algorithm with Enumeration Method for
i=1 (px max(wc − px , 0) + py max(hc − py , 0)) increases as
i i i
Obtaining Candidate Chassis Layout Schemes.
the rectangles move further away from the USVs chassis bound-
ary. This will keep the rectangles as far away from the chassis
boundary as possible to prevent damage due to collision during
the working process.
The constraints in (17) can be represented by the penalty func-
tion Pen(pi , wi , hi , wc , hc ), so the fitness function is defined as
follows:

D 
D
 i 
Fit(p1 , . . . , pD ) =  p − pj 
i=1 j=i+1


D
+ (pix max(wc − pix , 0) + piy max(hc − piy , 0))
i=1

+ Pen(p1 , . . . , pD ) (18)
where
Pen(p1 , . . . , pD ) = −(c1 ∗ P1 + c2 ∗ P2 + c3 ∗ P3 )
The penalty function Pen(p1 , . . . , pD ) is a weighted sum of distance. The PSO algorithm obtains the locations of rectangle
P1 , P2 and P3 . P1 is a penalty for overlap between equipment. 1 2 3 in Fig. 1(a). The chassis layout scheme is the locations
P2 is the penalty for equipment exceeding chassis boundaries. of rectangle 1 2 3 4 in the USVs chassis, which can be
The chassis is mounted with a driver’s seat, and P3 is the penalty obtained by combining the result of the PSO algorithm with
for overlap of the equipment with the driver’s seat. c1 , c2 , and c3 enumeration method.
are the weight coefficients of the penalty function, respectively. The Algorithm 1 illustrates the acquisition process of the

D 
D candidate chassis layout schemes. The input is Chassis_size
P1 = Xij Yij (19) and Equip_size. The Chassis_size denotes the length and
i=1 j=i+1 width of the chassis 6 in Fig. 1. The Equip_size denotes
the length and width of the rectangles 1–3 in Fig. 1.

D 
D
The output is Layout_list which denotes the candidate USVs
P2 = max{wi /2 − pix , 0} + max{hi /2 − piy , 0}
chassis layout schemes. The function Init_Parameters() (Al-
i=1 i=1
gorithm 1, Line 1) sets the weight parameter, the number

D
of iterations, and the size of the particle swarm. In function
+ max{pix − (wc − wi /2), 0} Init_Psoem(Chassis_size, Equip_size) (Algorithm 1, Line
i=1
2), firstly, a large number of particles and their corresponding

D velocities are randomly generated, and each of the particles rep-
+ max{piy − (hc − hi /2), 0} (20) resents a chassis layout scheme. Then, the fitness of the particles
i=1 is calculated. The function Get_InitBest(p_list, f itness_list)

D
  obtains the optimal particle position and its fitness value. In this
P3 = max{(wi + wd )/2 − pix − pdx  , 0} function, gbestp, gbestf itness, pbestp, pbestf itness repre-
i=1 sent the globally optimal particle, the fitness value of the global
 
∗ max{(hi + hd )/2 − piy − pdy  , 0} (21) optimal particle, the optimal value to the particle itself, and the
fitness value corresponding to the optimal value of the particle
where wd and hd are the width and height of the safety itself, respectively. Algorithm 1, Line 4 to Line 8 is the loop
officer’s seat, respectively. pdx and pdy are the X-coordinate part of the program, where the particle speed v_list, particle
and Y-coordinate of the position of the safety officer’s seat, posititon p_list, and particle fitness value f itness_list are
respectively. In Fig. 1, the safety officer seat 5 is fixed on updated here. The function Enumerate(p_list, battery_size)
the chassis and (21) is used to arrange the rectangles avoid combing the p_list with the battery size battery_size to derive
overlapping with it. the USVs chassis layout schemes Layout_list. The candidate
4) The Enumeration Method: The enumeration method is USVs chassis layout schemes are represented by Layout_list,
used to obtain the battery locations in the battery compartment which removes the schemes of rectangular overlap and beyond
and then derive the candidate layout schemes combing with the the chassis boundaries by the function Remove() (Algorithm 1,
result of the PSO algorithm. As shown in Fig. 1(b), the rectangle Line 10).
4 and rectangle 7 represent battery and battery compartment, Remark 1: Although the length and width of the equipment
respectively. The locations of the battery are obtained by evenly have been determined, the placement of the equipment may be
distributing the battery in the battery box according to a certain uncertain in the coordinate system, that is, the long side of the
Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12821

equipment is parallel to the X axis or the short side is parallel stability from candidate chassis layout schemes. In the next
to the X axis, which will also lead to different chassis layout section, the selecting method based on finite-frequency H∞
schemes. Therefore, when executing Algorithm 1, it is necessary control is explained to obtain the final chassis layout scheme.
to consider different equipment placement.
Remark 2: Many candidate schemes can be obtained through C. Selecting Method Based on Robust Finite-Frequency H∞
Algorithm 1. Because the position constraint of the equipment Control
is realized by adding a penalty function to the fitness function,
some particles may still not satisfy the space constraint in The objective of this section is to obtain the chassis layout
the process of particle motion. Therefore, when executing the scheme that the yaw rate response is close to the ideal yaw
Algorithm 1, it is still necessary to remove the scheme that rate response compared with other schemes as the final scheme.
does not satisfy the space constraint (see as function Remove() There are some reasons for using the H∞ control in the selecting
(Algorithm 1, Line 10)). method. First, during the chassis layout design stage, we only
focused on the open-loop lateral stability response of the chassis.
B. Filtering Based on Lyapunov Stability and Steering Second, it is an optimization problem that find the final scheme
Characteristics from the candidate schemes. The H∞ control in the selecting
method can get the most robust final scheme through optimiza-
The last section obtains the candidate USVs chassis layout tion compared with other feedback methods.
schemes with the space constraints. But only some of the can- In (7), the F is the negative ideal yaw rate gain. F δ is the
didate schemes can be applied in practice since some schemes negative ideal yaw rate of the 2-DOF chassis model. Ex(t) is the
may not be stable or satisfy the requirements of steering charac- actual yaw rate of the USVs chassis. Therefore, z(t) is the error
teristics. From the control point of view, this section selects the between the actual yaw rate of the USVs chassis and the ideal
system stability schemes. Then, from the steering perspective, yaw rate of the 2-DOF chassis dynamic model. According to the
choose the schemes that satisfy the understeer characteristics. LPV chassis dynamic model in (11), the robust H∞ performance
Fig. 3 shows that, as the speed increases, the eigenvalues of the dynamic model between front wheel steering angle input
of the state matrix of the 2-DOF chassis dynamics system and the error of yaw rate is:
gradually move toward the imaginary axis, which means that
the unstable trend of the chassis dynamics model increases z 2 < γ(α) δ 2 (23)
gradually. As mentioned in (8), the USVs chassis working at
a fixed speed range vx ∈ v x v x . For the same chassis layout where γ(α) is the H∞ performance parameter of the LPV chassis
scheme, the chassis dynamic model is stable at the speed range dynamic model, z 2 denotes the L2 -norm of the signal z(t) and
δ 2 denotes the L2 -norm of the signal δ.
vx ∈ v x v x if it is stable at v x . According to (11), the state
For each candidate chassis layout scheme, we derive the H∞
matrix A at v x of the 2-DOF chassis dynamic model can be performance index from γ(α) to describe the influence between
rewritten as follow: angle input and error output. The smaller the index value, the
 −cf −cr lr cr −lf cf  smaller the output error between the chassis yaw rate and the
mv x mv 2x
−1
A = l c −l c −l2 cf −lr2 cr (22) ideal yaw rate of the 2-DOF chassis dynamic model when the
r r f f f
Iz Iz v x same angle input is given. While the front wheel steering angle
always works in the low frequency in practice, it is meaningful
According to Lyapunov’s first theorem, the necessary and
to calculate the index by optimization at finite frequency [25].
sufficient condition for all the eigenvalues of the state matrix
The Finsler’s lemma is very helpful to the derivation of the
A of the 2-DOF chassis dynamics model in (22) have negative
following equations, we give the lemma below.
real parts. Therefore, we should select the schemes that all the
Lemma 1: (Finsler’s lemma [33]) Let ξ ∈ Rn , a symmet-
eigenvalues of the matrix A that have negative real parts.
ric matrix Φ = ΦT ∈ Rn×n , a matrix C ∈ Rn×n such that
The steering characteristics can be represented by K in (22),
rank(C) < n. The following statements are equivalent:
which denotes the stability factor for understeer or oversteer
1) ξ T Φξ < 0 holds for all Cξ = 0, ξ = 0.
vehicles [32]. If K = 0, the chassis is neutral steer. If K < 0, the
2) Y⊥T ΦY⊥ < 0.
chassis oversteers. If K > 0, the chassis understeers. When the
3) There exists Y ∈ Rn×m such that Φ + YC + C T Y T < 0.
chassis is oversteered, the chassis turns more than the steering
where Rn denotes the space of real (complex) vectors of
intends while the driver or autopilot system applies a steady
dimension n, Rm×n denotes the space of real (complex) vectors
steering input. Oversteer may lead to tailing and a full spin. The
of dimension m × n and the Y⊥ stands for any matrix whose
stability factor is independent of the longitudinal velocity vx
columns span the right null space of Y.
and determined by the chassis layout. To improve the safety of
Lemma 2 ([19]): For a positive real number γ(α), the system
the chassis, we should choose the chassis layout schemes that
in (11) has the H∞ performance in Eq. (23) if there exist
satisfy the appropriate understeer characteristics.
symmetric matrices Q1 and Q2 such that:
In Section III-A, we obtain the candidate chassis layout
schemes through the PSO algorithm with the enumeration  T   
method. In this section, we remove the chassis layout schemes A(α) B(α) −Q1 Q2 A(α) B(α)
+
with oversteer characteristics or not satisfying the Lyapunov I 0 Q2 π 2 Q1 I 0

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12822 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

 T    According to Lemma 1 (Finsler’s lemma), the condition in (26)

E(α) F (α) I 0 E(α) F (α)
< 0 (24) holds if the following condition satisfied:
0 I 0 −γ(α)2 I 0 I
⎡ ⎤ ⎡ ⎤
Theorem 1: For a positive real number γ(α), the sys- −Q1 Q2 0 0 S1 (α) H1 (α)
tem in (11) has the H∞ performance in (23) if there ex- ⎢ ∗ π 2 Q1 ⎥ ⎢ H2 (α)⎥
⎢ 0 0 ⎥ ⎢S2 (α) ⎥
ist symmetric matrices Q1 and Q2 and random matrices ⎢ ⎥+⎢ ⎥
⎣ ∗ ∗ I 0 ⎦ ⎣S3 (α) H3 (α)⎦
S1 (α), S2 (α), S3 (α), S4 (α), H1 (α), H2 (α), H3 (α) and H4 (α)
∗ ∗ ∗ −γ(α) I
2
S4 (α) H4 (α)
such that the following condition holds:
 
⎡ ⎤ −I A(α) 0 B(α)
−Q1 + Λ1 Q2 + Λ 2 Λ3 Λ4 ×
⎢ 0 E(α) −I F (α)
⎢ ∗ π 2 Q1 + Λ 5 Λ 6 Λ 7 ⎥ ⎥
⎢ ⎥<0 (25) ⎡ ⎤T
⎣ ∗ ∗ Λ8 Λ9 ⎦ S1 (α) H1 (α)
 T
∗ ∗ ∗ Λ10 −I A(α) B(α) ⎢S (α) H2 (α)⎥
0 ⎢ 2 ⎥
+ ×⎢ ⎥ <0
where 0 E(α) −I F (α) ⎣S3 (α) H3 (α)⎦
S4 (α) H4 (α)
Λ1 = − S1 (α) − S1 (α)T , (28)
T The condition (25) can be derived from condition (28). The proof
Λ2 = S1 (α)A(α) − S2 (α) + H1 (α)E(α),
is completed. 
Λ3 = − S3 (α)T − H1 (α), Ref. [19] exploited the H∞ performance at finite frequency of
the augmented system and design the Mixed H_/H∞ observer.
Λ4 = S1 (α)B(α) − S4 (α)T + H1 (α)F (α), The conservatism of the [19] is reflected in two aspects. On the
one hand, the γ obtained in (47) in [19] is fixed. On the other
Λ5 = S2 (α)A(α) + (S2 (α)A(α))T
hand, [19] sets the LPV polygon as quadrilateral according to
+ H2 (α)E(α) + (H2 (α)E(α))T , the variation of the longitudinal velocity. However, the LPV
model is located in a smaller region than the quadrilateral.
Λ6 = (S3 A(α))T − H2 (α) + (H3 (α)E(α))T , To reduce the conservatism of the H∞ performance index, we
Λ7 = S2 (α)B(α) + (S4 (α)A(α))T improve the theorem in [19], and the new theorem is shown
in Theorem 1. The proof of the theorem is proposed above.
+ H2 (α)F (α) + (H4 (α)E(α))T , Then, we define the following to calculate the H∞ performance
index:
Λ8 = I − H3 (α) − H3 (α)T ,
Λ9 = S3 (α)B(α) + H3 (α)F (α) − H4 (α)T , 
6 
6
S1 (α) = αi S1,i , S2 (α) = αi S2,i ,
Λ10 = − γ(α)2 I + S4 (α)B(α) + (S4 (α)B(α))T , i=1 i=1

+ H4 (α)F (α) + (H4 (α)F (α))T 

6 
6
S3 (α) = αi S3,i , S4 (α) = αi S4,i ,
Proof: The (25) can be rewritten as: i=1 i=1

⎡ ⎤T ⎡ ⎤ 
6 
6
A(α) B(α) −Q1 Q2 0 0 H1 (α) = αi H1,i , H2 (α) = αi H2,i ,
⎢ I ⎥ ⎢ ⎥
⎢ 0 ⎥ ⎢ ∗ π 2 Q1 0 0 ⎥ i=1 i=1
⎢ ⎥ ×⎢ ⎥
⎣E(α) F (α)⎦ ⎣ ∗ ∗ I 0 ⎦

6 
6
0 I ∗ ∗ ∗ −γ(α) I
2
H3 (α) = αi H3,i , H4 (α) = αi H4,i ,
⎡ ⎤ i=1 i=1
A(α) B(α)
⎢ I 0 ⎥ 
6
⎢ ⎥ γ(α) = αi γi .
×⎢ ⎥<0 (26) (29)
⎣E(α) F (α)⎦ i=1
0 I
The (25) in Theorem 1 can be rewritten as:
It is note that:
⎡ ⎤ ⎡ ⎤
  A(α) B(α) −Q1 + Λ1 Q2 + Λ2 Λ3 Λ4
−I A(α) B(α) ⎢ I 0 ⎥ ⎢ ∗ π 2 Q1 + Λ 5 Λ7 ⎥
0 ⎢ ⎥ ⎢ Λ6 ⎥
=⎢ ⎥ (27) ⎢ ⎥<0 (30)
0 E(α) −I F (α) ⎣E(α) F (α)⎦ ⎣ ∗ ∗ Λ8 Λ9 ⎦
⊥
0 I ∗ ∗ ∗ Λ10

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12823

where TABLE I
MAIN PARAMETER VALUES OF THE USVS CHASSIS AND EQUIPMENT

T T
Λ1 = − S1,i − S1,j − S1,i − S1,j ,
T T
Λ2 = S1,i Aj + S1,j Ai − S2,i − S2,j + H1,i Ej + H1,j Ei ,
T T
Λ3 = − S3,i − S3,j − H1,i − H1,j ,
T T
Λ4 = S1,i Bj + S1,j Bi − S4,i − S4,j + H1,i Fj + H1,j Fi ,
Λ5 = S2,i Aj + S2,j Ai + (S2,i Aj )T + (S2,j Ai )T TABLE II
T T MAIN PARAMETER VALUES OF THE ALGORITHM 1
+ H2,i Ej + H2,j Ei + (H2,i Ej ) + (H2,j Ei ) ,
Λ6 = (S3,i Aj )T + (S3,j Ai )T − H2,i − H2,j
+ (H3,i Ej )T + (H3,j Ei )T ,
Λ7 = S2,i Bj + S2,j Bi + (S4,i Aj )T + (S4,j Ai )T
+ H2,i Fj + H2,j Fi + (H4,i Ej )T + (H4,j Ei )T ,
T T
Λ8 = I − H3,i − H3,j − H3,i − H3,j ,
T T
Λ9 = S3,i Bj + S3,j Bi + H3,i Fj + H3,j Fi − H4,i − H4,j ,
Λ10 = − γi2 I + S4,i Bj + S4,j Bi + (S4,i Bj )T + (S4,j Bi )T ,
− γj2 I + H4,i Fj + H4,j Fi + (H4,i Fj )T + (H4,j Fi )T , obtained by linear inequalities (30) according to the simulation
effect shown in Figs. 19 and 20.
1 ≤ i ≤ j ≤ 6. Remark 4: Compared with [19], the improvement of theory
1 is mainly reflected in two aspects. One is that our LPV model
Linear matrix inequality (30) is the final derivation of the polygon is a triangle. The area of the LPV polygon in Theorem
H∞ linear matrix inequality (LMI) at finite frequency. For each 1 is smaller than the area of the polygon in [19] and reduced by
scheme, we can obtain {γ1 , . . . , γ6 } by minimize the sum of more than 50%. Another is that we simultaneously solve for the
γi in linear matrix inequality (30). In this article, we select the six {γ1 , . . . , γi } corresponding to the six LPV vertices instead
minimum value of {γ1 , . . . , γ6 } as the H∞ performance index of a fix γ, thereby reducing the conservatism.
γ of the chassis layout scheme:
IV. SIMULATION RESULTS
γ = min{γ1 , . . . , γ6 } (31)
A. Simulation on Obtaining Candidate USVs Chassis Layout
For all candidate chassis layout schemes, we can derive the Schemes
set Γ from (30) and (31): In this section, we implement the Algorithm 1 which com-
bines the PSO algorithm with the enumeration method to obtain
Γ = {γ(1) , γ(2) , . . . , γ(i) , . . . , γ(n) } (32) the candidate USVs chassis layout schemes in a Python environ-
ment. The experimental computer is with i5-8250 U and 16 G
where n is the number of candidate USVs chassis layout RAM. The basic USVs chassis parameters such as mass and size
schemes, i indicates the index of the ith scheme of candidate in Fig. 1 are shown in the Table I. Besides, the acceleration of
chassis layout schemes and γ(i) indicates the H∞ performance gravity g is 9.8 m/s2 . The wheelbase of the chassis is 2.5 m. The
index of the ith scheme. The final scheme we choose is the minimum and maximum chassis speed v x is 1 m/s and 10 m/s,
scheme with the smallest γ(i) in Γ. respectively. The chassis speed in simulation is 9 m/s.
Remark 3: According to (29) and (30), the H∞ performance In terms of Remark 1, the placement of single equipment may
index is a combination of {γ1 , . . . , γ6 } through {α1 , . . . , α6 }. influence the number of the final schemes due to some equipment
However, according to Fig. 4 and (12), {α1 , . . . , α6 } are time- is simplified to rectangles. In this simulation, because the rectan-
varying. So it is difficult to find suitable αi to get the H∞ per- gle 2 and 4 in Fig. 1 are simplified to square, there are 4 types
formance index theoretically. The simulation results in Section of placement. Type 1 is that the long side of rectangle 1 and 3
IV-B show that the errors between the LPV chassis dynamic paralleled to the X axis. Type 2 is that the long side of rectangle
model and 2-DOF chassis dynamic model increase with the 1 and the short side of rectangle 3 paralleled to the X axis.
increase of the γ, which is the minimum value of {γ1 , . . . , γ6 }. Type 3 is that the short side of rectangle 1 and the long side of
Therefore, we select the minimum value of {γ1 , . . . , γ6 } as the rectangle 3 paralleled to the X axis. Type 4 is that the short side
H∞ performance index of ith chassis layout schemes which of rectangle 1 and 3 paralleled to the X axis. Table II shows

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12824 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

TABLE III
RESULT OF ALGORITHM 1

Fig. 7. USVs chassis layout schemes for Type 1 to Type 4.

Fig. 6. Fitness value of during optimization.

the basic simulation parameters used in Algorithm 1. Table III

illustrates the result of Algorithm 1. For different displacements
of rectangles from Type 1 to Type 4, the number of particles is
15000. The more particles, the more schemes will be generated,
which will increase the computational burden of the computer.
Since the battery is separately installed under the chassis, all Fig. 8. Comparison of PSO, GWO and GA algorithms in terms of number of
possible installation positions can be found by enumeration. In candidate schemes and time cost.
this simulation, we set 64 possible battery installation positions.
For different displacements of rectangles from Type 1 to Type
4, the number of candidate schemes is obtained. officer’s seat, and its size is a square with a side length of 0.6 m.
The fitness value during the optimization is shown in Fig. 6. Green rectangular blocks represent rectangle 1 2 3 on the
Different from the traditional PSO method of obtaining the opti- chassis in Fig. 1. Blue rectangle 4 in Fig. 1 represents the
mal solution, the PSO in this article is used to find the candidate battery under the chassis. By running Algorithm 1, the candidate
chassis layout schemes, which is obtaining many solutions, in chassis layout schemes are obtained, and these schemes satisfy
other words. Due to random initialization, most particles do not space constraints.
satisfy the space constraints and design requirements. Using the In order to demonstrate the performance of different heuristic
PSO method, the spatial constraints and design requirements optimization algorithms in obtaining candidate solutions, we
are represented by the objective function and fitness function selected the gray wolf optimization (GWO) algorithm and the
so that the particles move along the direction of maximizing genetic algorithm (GA) for comparison. In these three opti-
the fitness function, and finally, more particles that satisfy the mization algorithms, we used the same fitness function, and
spatial constraints and design requirements are obtained. In this most of the relevant parameters are in Table I. For ease of
experiment, the PSO algorithm took nearly 174 s. As the number calculation, we set the number of populations P to 5000 and the
of particles increases, the time consumed by PSO also increases. number of iterations N to 100. We conducted 10 experiments
This may cause a little latency. But this time delay can be for each method, and the final results were averaged over the 10
allowed, because on the one hand, the time delay is not very experiments. The results of the comparison are shown in Fig. 8
high, and on the other hand, the real-time requirement of the and Fig. 9.
PSO algorithm is not very high in the design stage. As shown in Fig. 8, The blue bar represents the number
Several candidate USVs chassis layout schemes from Type 1 of candidate schemes of each algorithms and the orange bar
to Type 4 are randomly selected to show in Fig. 7. As shown in represents the time cost. The PSO algorithm has advantages in
Table I, the length of the chassis is 3.4 m, and the width is 1.6 m. generating the number of candidate schemes (1822) and time
In Fig. 7, the red dotted line represents the Cartesian coordinate cost (26.84 s). The number of candidate schemes of GWO is
axis, and the coordinate origin is at the geometric center of the 1490 and the time cost is 27.90 s. Although the GA algorithm
chassis. The black dotted line represents the position of the safety obtains 1443 candidate schemes, the time cost reaches 955 s.

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12825

Fig. 9. Comparison of PSO, GWO and GA algorithms in terms of fitness

value.

Fig. 11. Yaw rate error of angular step signal input with the H∞ performance
index of the 200 schemes.

Fig. 10. Front wheel angle input. (a) Angular step signal input.
(b) Sinusoidal signal input.

It is more than 30 times that of the other two methods. As

shown in Fig. 9, it can be seen that the convergence speed of
the PSO algorithm is the fastest. The GA algorithm is second,
and the GWO algorithm is the slowest. However, Due to the
time consumption of the GA algorithm reached 955 s, the GA
algorithm consumes more time to converge compared to the
other two methods. At the same time, the PSO algorithm has a Fig. 12. Angular step signal input yaw rate response of the 5 schemes evenly
select from 200 schemes.
higher fitness value, indicating that the PSO algorithm obtains
better candidate schemes compared to the other two methods.
According to this result, we believe that the PSO algorithm has signal input changes with γ when the input is Fig. 10(a). The yaw
better performance in obtaining chassis candidate schemes. rate error of angular step signal input Estep can be calculated as
follow:
B. Simulation on Selecting Method Based on Robust  
Finite-Frequency H∞ Control V −Vstep 
Estep =  step_ideal
Vstep_ideal  (33)
In this section, we select the final USVs chassis layout scheme
from the candidate schemes based on robust finite-frequency where the Vstep_ideal is the ideal yaw rate obtained from the
H∞ control. Under the front wheel angular step signal input 2-DOF chassis dynamic model and Vstep is the steady-response
and sinusoidal signal input, different chassis layout schemes value under the angular step signal input of the LPV chassis
are simulated and compared with the 2-DOF chassis dynamic dynamic model. From Fig. 11, with the increase of γ, the yaw
model. The simulation parameters of the chassis are shown in rate error of angular step signal input Estep increases gradually.
Table I. Based on these parameters and (11), the LPV chassis We evenly select 5 from the 200 schemes in Fig. 11, and the
dynamic model is established. Fig. 10(a) and (b) are the angular angular step signal input response curve of the 5 schemes is
step signal input and sinusoidal signal input, respectively. shown in Fig. 12. We can see that as γ increases, the response
In the simulation of front wheel angular step signal input, curve gradually moves away from the ideal response curve,
we evenly select 200 schemes from the candidate chassis layout implying that γ has the ability to be used as a performance index
schemes from Type 1 to Type 4. The H∞ performance index of lateral stability.
γ of each USVs chassis layout scheme is obtained by solving In the simulation of front wheel sinusoidal signal input, we
matrix inequalities (30) and (31) using Yalmip in Matlab. All also use the 200 schemes obtained above. Fig. 13 illustrates how
chassis layout schemes are sorted from small to large according the yaw rate error of sinusoidal signal input changes with γ. The
to γ. Fig. 11 illustrates how the yaw rate error of angular step yaw rate error of sinusoidal signal input Esin can be calculated

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12826 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

Fig. 13. Yaw rate error of sinusoidal signal input with the H∞ performance
index of the 200 schemes.

Fig. 15. Scheme of the three methods. (a) is the result of traditional design
method. (b) is the result of the proposed design method in this paper. (c) is the
result with biggest fitness value of PSO.

TABLE IV
RESPONSE ERROR OF THE THREE METHODS

C. Compare With the Traditional Design Method, the

Proposed Design Method in This Paper and the PSO
Fig. 14. Sinusoidal signal input yaw rate response of the 5 schemes evenly
Algorithm With Enumeration
select from 200 schemes. Traditional design methods always depend on the designer’s
experience. That is the method to obtain the scheme in which
the equipment is always located on the longitudinal axis of the
as follow: chassis and as far away from each other as possible. Based
 on the above rules, we get a chassis layout scheme based on

T
the traditional method, which is shown in Fig. 15(a). In this
Esin = (Xsin_ideal,t − Xsin,t )2 (34)
t=1 scheme, the equipment is arranged on the longitudinal axis of
the chassis and far away from each other. Fig. 15(b) is the final
where T is the simulation duration, Xsin_ideal,t is the ideal layout scheme obtained from the selection method based on the
output obtained from the 2-DOF chassis dynamic model of the robust finite frequency H∞ method, which is regarded as the
chassis at time t and the Xsin,t is the sinusoidal signal input proposed design method in this article. As shown in Fig. 15(c),
response of the LPV chassis dynamic model at time t. From we got the scheme represented by the particle with the biggest
Fig. 13, with the increase of γ, the yaw rate error of sinusoidal fitness value of PSO in the candidate scheme generation phase,
signal input Esin increases. We evenly select 5 from the 200 and compared it with the above two schemes.
schemes in Fig. 13, and the sinusoidal signal input response Based on the schemes in Fig. 15(a) and (b), we obtain the
curve of the 5 schemes is shown in Fig. 14. We can find that as γ angular step signal response and sinusoidal signal response. The
increases, the response curve also gradually moves away from ideal response of the two methods is the response of the 2-DOF
the ideal response curve. chassis dynamic model. The input signal is also the input signal
From the simulation results, the H∞ performance index γ in Fig. 10. In Figs. 16 and 17, we find the response curve of the
determines the response error Estep and Esin between the USVs method in this article is closer to its ideal response curve both
chassis and the 2-DOF chassis model to a certain extent. When in angular step signal response and sinusoidal signal response
the γ is larger, the response error is larger. The final layout situation. Table IV illustrates that the γ and the response error
scheme is the USVs chassis layout scheme with the lowest γ. of Estep and Esin of the proposed design method is small than
Compared with other schemes, this scheme makes the lateral the traditional method. It indicates that the designed chassis by
stability performance of the chassis close to the 2-DOF chassis the method in this paper have better lateral stability compared
dynamic model. with the traditional method. Meanwhile, the scheme obtained by

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12827

Fig. 18. Compare of the three methods.

Fig. 16. Response of the angular step signal inputs and its ideal response.
layout scheme index from 1 to 200, and the Y-axis is the H∞
performance index γ calculated by the mentioned three methods
above. The result of the conventional H∞ method without finite
frequency has the biggest γ, which indicates that this method
is the most conservative compared to other methods. The H∞
performance index γ calculated by the improved method in this
article is less than by the method in [19] and the conventional
H∞ method, which indicates that the improved method has less
conservatism and the results obtained by this method have higher
reliability.

E. Verify the Effectiveness of the Selecting Method Based on

Robust Finite-Frequency H∞ Control At Variable Front Wheel
Steering Angle Input and Chassis Speed
In this section, we aim to verify the effectiveness of the
selecting method based on robust finite-frequency H∞ control at
Fig. 17. Response of the sinusoidal signal inputs and its ideal response. variable front wheel steering angle and chassis speed. First, the
front wheel steering angle input of the LPV chassis dynamic
model increases gradually to verify the effectiveness of the
the PSO with the biggest fitness value has poor lateral stability
selecting method at variable front wheel steering angles and
performance, which indicates that even with a better PSO fitness
fix the speed. Then, the effectiveness of H∞ condition in (23) of
value, it’s H∞ performance is not always better. Therefore,
the final scheme is verified at variable chassis speed and fixed
the H∞ performance of the layout scheme needs to be solved
front wheel steering angle.
separately.
1) Verify the Effectiveness of the Selecting Method Based on
Robust Finite-Frequency H∞ Control At Variable Front Wheel
D. Compare With the Calculation of H∞ Performance Index Steering Angle Input and Fix Speed: As shown in (31), we
in [19] and Its Improved Method
define the H∞ performance index γ as the minimum value of
In [19], the calculation method and its proof are proposed {γ1 , . . . , γ6 }. In this part, we verify the effectiveness of H∞
to obtain the H∞ performance index at finite frequency. In performance index in the selection of optimal chassis layout
order to reduce the conservatism of the result of the method, scheme by comparing the relationship between the input and
we proposed an improved calculation method of Theorem 1 output of variable schemes. According to the speed of the
and matrix inequality (30). In this section, we compare the chassis, the αi , i = 1, . . . , 6 can be calculated by (12). The
result of the conventional H∞ method without finite frequency, A(α), B(α), E(α), F (α) of the chassis model is obtained by
the calculation method in [19], and the improved method in (11). Given the angle input signal δ, the output signal z can be
this article. The 200 chassis layout schemes are also obtained obtained according to the chassis model. We evenly select 10
from Section IV-B, which is evenly chosen from the candidate chassis layout schemes obtained from candidate schemes, and
schemes. the result is shown in Fig 19.
Each of the 200 chassis layout schemes is calculated by the Each line in Fig. 19 represents a chassis layout scheme and
method in [19] and the improved method in this paper, respec- has a γ corresponding to it. In each scheme, we select 10 front
tively. The result is illustrated in Fig. 18. The X-axis is the chassis wheel steering angle input signals δ from 0.017 rd to 0.17 rd,

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
12828 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 72, NO. 10, OCTOBER 2023

V. CONCLUSION
In this article, an optimal chassis layout design framework
based on lateral stability is proposed. This framework designs
the layout of the chassis from the perspective of control. First,
the candidate USVs chassis layout schemes are obtained by
the PSO algorithm with the enumeration method to satisfy
the space constraints. Then, the schemes are filtered based
on Lyapunov stability and steering characteristics. Finally, the
final scheme is obtained by selecting method based on robust
finite-frequency H∞ control. In the simulation, the candidate
layout schemes are first obtained by using the PSO algorithm
with the enumeration method, and some schemes are visualized.
Then, the front wheel steering angle step input and front wheel
sinusoidal input response are simulated for the LPV chassis
Fig. 19. Input and output relationship of different chassis layout schemes. dynamic model. The simulation results show that the final
chassis layout scheme obtained by the selecting method based
on robust finite-frequency H∞ control has good lateral stability,
which yaw rate response is close to the ideal yaw rate of 2-DOF
chassis dynamic model compared with other schemes. After that,
we compare the result of the calculation method of the H∞
performance index, which is proposed in [19] and improved in
this article. Finally, the effectiveness of the selecting method
based on robust finite-frequency H∞ control at variable front
wheel steering angle input and chassis speed is verified. Future
studies will focus on designing the layout of the chassis from the
perspective of saving input energy consumption and extending
Fig. 20. γ(α) and z 2 / δ 2 of final chassis layout scheme at variable chassis life.
speed and fix front wheel steering angle. (a) The change in speed over time.
(b) γ(α) and z 2 / δ 2 of final chassis layout scheme.
REFERENCES
[1] H. Kolbe and A. Muetze, “Limitations of established vehicle modelling
approaches for the conceptual design of hybrid special-purpose vehicles,”
and the output signals z are obtained. The fixed speed is 9 m/s. It in Proc. IEEE Veh. Power Propulsion Conf., 2011, pp. 1–5.
shows that with the increasing of H∞ performance index γ, the [2] Y. Chen, Y. Du, Z. Song, and Z. Zhu, “A knowledge base system for rapid
output z, which represents the error between the actual yaw rate design of harvesting machinery chassis,” in Proc. Int. Conf. Intell. Syst.
Res. Mechatronics Eng., 2015, pp. 1459–1464.
of the USVs chassis and the ideal yaw rate of 2-DOF chassis [3] S. Rao and A. Bhattu, “Dynamic analysis and design optimization of
dynamic model increases gradually. Therefore, it is effective to automobile chassis frame using FEM,” in Machines, Mechanism and
select γ as the H∞ performance index of optimal chassis layout Robotics. Berlin, Germany: Springer, 2019, pp. 671–680.
[4] M. Palanivendhan, S. Devanand, J. Chandradass, J. Philip, and S. S. Reddy,
design. “Design and analysis of 3-wheeler chassis,” Mater. Today: Proc., vol. 45,
2) Verify the Effectiveness of the Selecting Method Based on pp. 6958–6968, 2021.
Robust Finite-Frequency H∞ Control At Variable Speed and [5] N. A. Mohammed, N. Nandu, A. Krishnan, A. R. Nair, and P. Sreedharan,
“Design, analysis, fabrication and testing of a formula car chassis,” Mater.
Fix Front Wheel Steering Angle: We select the final scheme Today: Proc., vol. 5, no. 11, pp. 24944–24953, 2018.
obtained by the robust finite-frequency H∞ control to verify the [6] M. Abdullah, M. M. Najmi, M. Harun, F. Ramli, and S. Mat, “Chassis
effectiveness of the H∞ condition in (23) at variable speed and design and analysis of narrow track vehicle,” Proc. Innov. Res. Ind.
Dialogue, vol. 16, pp. 177–178, 2017.
fix front wheel steering angle. The fixed front wheel steering [7] V. Parlaktaş, E. Tanık, N. Babaarslan, and G. B. Çalık, “The design and
angle is 0.017 rd, and the variable speed is shown in Fig. 20(a). facturing process of an electric sport car (EVT S1) chassis,” Iranian J. Sci.
We can derive the αi , i = 1, . . . , 6 at different speed and the Technol., Trans. Mech. Eng., vol. 45, no. 1, pp. 103–113, 2021.
[8] S. Mantovani, G. A. Campo, A. Ferrari, and M. Cavazzuti, “Optimization
γ(α) by the (12) and matrix inequality (30). Meanwhile, the methodology for automotive chassis design by truss frame: A. preliminary
ratio of the L2 -norm of the output signal z to the input signal δ investigation using the lattice approach,” in Transdisciplinary Engineering
of the final scheme is also calculated. Methods for Social Innovation of Industry 4.0. Amsterdam, The Nether-
lands: IOS Press, 2018, pp. 984–992.
In Fig. 20(a), the chassis speed varies sinusoidally, which the [9] M. Cavazzuti, D. Costi, A. Baldini, and P. Moruzzi, “Automotive Chas-
min speed is 3 m/s and max speed is 7 m/s. Fig. 20(b) shows that sis Topology Optimization: A Comparison Between Spider and Coupe
the curve of the z 2 / δ 2 is below the curve γ(α), which indi- Designs,” in Proc. World Congr. Eng., 2011, vol. 3, pp. 6–8.
[10] M. Cavazzuti, A. Baldini, E. Bertocchi, D. Costi, E. Torricelli, and
cates that the final scheme has good H∞ performance according P. Moruzzi, “High performance automotive chassis design: A topology
to the (23). Due that the final scheme is obtained according to optimization based approach,” Struct. Multidisciplinary Optim., vol. 44,
the H∞ performance index γ, which is the minimum value of no. 1, pp. 45–56, 2011.
[11] S. Nandhakumar, S. Seenivasan, A. M. Saalih, and M. Saifudheen, “Weight
{γ1 , . . . , γ6 }, it indicates that it is also effective to select the γ optimization and structural analysis of an electric bus chassis frame,”
as the H∞ performance index of optimal chassis layout design. Mater. Today: Proc., vol. 37, pp. 1824–1827, 2021.

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: OPTIMAL INTELLIGENT CHASSIS LAYOUT DESIGN FRAMEWORK BASED ON PSO 12829

[12] T. Brüdigam, M. Olbrich, D. Wollherr, and M. Leibold, “Stochastic model Yongkang Zhang received the B.Sc. degree form
predictive control with a safety guarantee for automated driving,” IEEE the Shandong University of Science and Technology,
Trans. Intell. Veh., vol. 8, no. 1, pp. 22–36, Jan. 2023. Qingdao, China, in 2018 and the M.Sc. degree in 2021
[13] C. Pai-Chen and C. Chih-Keng, “Integrated chassis control and control from Beihang University, Beijing, China, where he
allocation for all wheel drive electric cars with rear wheel steering,” is currently working toward the Ph.D. degree with
Electronics, vol. 10, no. 22, 2021, Art. no. 2885. the School of Transportation Science and Engineer-
[14] S. Fergani, O. Sename, and L. Dugard, “An LPV/H∞ integrated ve- ing. His research interests include automated vehicle
hicle dynamic controller,” IEEE Trans. Veh. Technol, vol. 65, no. 4, dynamic control and automated chassis design.
pp. 1880–1889, Apr. 2016.
[15] H. Guo, F. Liu, F. Xu, H. Chen, D. Cao, and Y. Ji, “Nonlinear model
predictive lateral stability control of active chassis for intelligent vehicles
and its FPGA implementation,” IEEE Trans. Syst., Man, Cybern. Syst.,
vol. 49, no. 1, pp. 2–13, Jan. 2019.
[16] H. Taghavifar, C. Hu, L. Taghavifar, Y. Qin, J. Na, and C. Wei, “Opti- Lei Wang received the B.Sc. Degree from Shenyang
mal robust control of vehicle lateral stability using damped least-square Aerospace University, Shengyang, China, in 2002
backpropagation training of neural networks,” Neurocomputing, vol. 384, and the M.Sc. degree from Tsinghua University, Bei-
pp. 256–267, 2020. jing, China, in 2006. He is currently working toward
[17] G. Perozzi, J. J. Rath, C. Sentouh, J. Floris, and J. C. Popieul, “Lateral the Ph.D. degree with the School of Transportation
shared sliding mode control for lane-keeping assist system in steer-by-wire Science and Engineering, Beihang University, Bei-
vehicles: Theory and experiments,” IEEE Trans. Intell. Veh., vol. 8, no. 4, jing. His research interests include mobile robot plan-
pp. 3073–3082, Apr. 2023. ning and control, automated driving and vehicle for-
[18] Z. Li, J. Jiang, W.-H. Chen, and L. Sun, “Autonomous lateral maneuvers for mation.
self-driving vehicles in complex traffic environment,” IEEE Trans. Intell.
Veh., vol. 8, no. 2, pp. 1900–1910, Feb. 2023.
[19] H. Zhang and J. Wang, “Active steering actuator fault detection for an
automatically-steered electric ground vehicle,” IEEE Trans. Veh. Technol.,
vol. 66, no. 5, pp. 3685–3702, May 2017.
Hui Zhang (Senior Member, IEEE) received the
[20] K. Guo, P. Chen, N. Xu, C. Yang, and F. Li, “Tire side force characteristics
with the coupling effect of vertical load and inflation pressure,” SAE Int. B.Sc. degree in mechanical design manufacturing and
automation from the Harbin Institute of Technology
J. Veh. Dyn., Stability, NVH, vol. 3, pp. 19–30, 2018.
at Weihai, Weihai, China, in 2006, the M.Sc. de-
[21] N. Ahmadian, A. Khosravi, and P. Sarhadi, “Managing driving distur-
gree in automotive engineering from Jilin University,
bances in lateral vehicle dynamics via adaptive integrated chassis control,”
Proc. Inst. Mech. Engineers, Part K: J. Multi-body Dyn., vol. 235, no. 1, Changchun, China, in 2008, and the Ph.D. degree
in mechanical engineering from the University of
pp. 122–133, 2021.
Victoria, Victoria, BC, Canada, in 2012. Dr. Zhang
[22] Y. Huang, W. Liang, and Y. Chen, “Estimation and analysis of vehicle
was the recipient of 2017 IEEE Transactions on Fuzzy
lateral stability region,” in Proc. Amer. Control Conf., 2017, pp. 4303–
4308. Systems Outstanding Paper Award, 2018 SAE Ralph
R. Teetor Educational Award, IEEE Vehicular Tech-
[23] J. Yoon, S. Yim, W. Cho, B. Koo, and K. Yi, “Design of an unified chassis
nology Society 2019 Best Vehicular Electronics Paper Award, and 2019 SAE
controller for rollover prevention, manoeuvrability and lateral stability,”
International Intelligent and Connected Vehicles Symposium Best Paper Award.
Veh. Syst. Dyn., vol. 48, no. 11, pp. 1247–1268, 2010.
He is a Member of SAE International, Senior Member of IEEE and Member
[24] H. Zhang, X. Zhang, and J. Wang, “Robust gain-scheduling energy-to-peak
of ASME. He is an Associate Editor for IEEE TRANSACTIONS ON VEHICULAR
control of vehicle lateral dynamics stabilisation,” Veh. Syst. Dyn., vol. 52,
no. 3, pp. 309–340, 2014. TECHNOLOGY, Journal of The Franklin Institute, SAE International Journal of
[25] H. Zhang, G. Zhang, and J. Wang, “Sideslip angle estimation of an Vehicle Dynamics, Stability, and NVH, SAE International Journal of Connected
electric ground vehicle via finite-frequency H∞ approach,” IEEE Trans. and Automated Vehicles, and ASME Transactions Journal of Dynamic Systems,
Measurement and Control, and Board Member of International Journal of
Transport. Electrific., vol. 2, no. 2, pp. 200–209, Jun. 2016.
Hybrid and Electric Vehicles, Mechanical Systems and Signal Processing.
[26] Y. Q. Xu, Z. Yang, and Q. X. Meng, “Particle swarm optimization method
for panel layout,” in Key Engineering Materials, vol. 450, Wollerau,
Switzerland: Trans. Tech. Pub., 2011, pp. 308–311.
[27] M. A. Mohammed and R. A. Hasan, “Particle swarm optimization for
facility layout problems FLP − A comprehensive study,” in Proc. 13th
IEEE Int. Conf. Intell. Comput. Commun. Process., 2017, pp. 93–99.
[28] J. Dou, X. Wang, and L. Wang, “Machining fixture layout optimization Makoto Iwasaki (Fellow, IEEE) received the B.S.,
using particle swarm optimization algorithm,” Proc. SPIE, vol. 7997,
M.S., and Dr.Eng. degree in electrical and computer
pp. 193–198, 2011.
engineering from the Nagoya Institute of Technology
[29] X. Liu, Z. Liu, S. Yu, and T. Gong, “Adapted particle swarm optimization (NITech), Nagoya, Japan, in 1986, 1988, and 1991,
algorithm–based layout design optimization of passenger car cockpit for
respectively. Since 1991, he has been with the Nagoya
enhancing ergonomic reliability,” Adv. Mech. Eng., vol. 11, no. 3, pp. 1–14,
Institute of Technology, where he is currently a Pro-
2019.
fessor. His research interests include applications of
[30] S. Ohmori, K. Yoshimoto, and K. Ogawa, “Solving facility layout problem control theories to linear/nonlinear modeling and pre-
via particle swarm optimization,” in Proc. IEEE 3rd Int. Joint Conf.
cision positioning for industrial mechatronic systems,
Comput. Sci. Optim., 2010, vol. 1, pp. 409–413.
through various collaborative research activities with
[31] Y.-Y. Guo, Q. Wang, and F. Liang, “Facility layout design based on
industries. From 2010 to 2014, Dr. Iwasaki was a
particle swarm optimization,” Comput. Integr. Manuf. Syst., vol. 18, no. 11, Technical Editor for IEEE/ASME TMech. He has been an Associate Editor
pp. 2476–2484, 2012.
for the IEEE TIE Since 2014, and the Co-Editors-in-Chief for IEEE TIE since
[32] W. F. Milliken et al., “Race Car Vehicle Dynamics,” vol. 400, Warrendale,
2016. From 2010 to 2018, he was AdCom Member of IES From 2018 to 2021,
PA, USA: SAE, 1995.
he was an IES Vice President for Planning and Development. He is a Fellow of
[33] D. H. Lee, “An improved finite frequency approach to robust filter design the Institute of Electrical Engineers of Japan, and a member of the Japan Society
for LTI systems with polytopic uncertainties,” Int. J. Adaptive Control
for Precision Engineering.
Signal Process., vol. 27, no. 11, pp. 944–956, 2013.

Authorized licensed use limited to: Chang'an University. Downloaded on July 09,2024 at 06:58:53 UTC from IEEE Xplore. Restrictions apply.