sensors

Article
RBS and ABS Coordinated Control Strategy Based on Explicit
Model Predictive Control
Liang Chu, Jinwei Li, Zhiqi Guo, Zewei Jiang, Shibo Li, Weiming Du, Yilin Wang and Chong Guo *

State Key Laboratory of Automotive Simulation and Control, Jilin University, Changchun 130022, China;
[email protected] (L.C.); [email protected] (J.L.); [email protected] (Z.G.);
[email protected] (Z.J.); [email protected] (S.L.); [email protected] (W.D.);
[email protected] (Y.W.)
* Correspondence: [email protected]

Abstract: During the braking process of electric vehicles, both the regenerative braking system (RBS)
and anti-lock braking system (ABS) modulate the hydraulic braking force, leading to control conflict
that impacts the effectiveness and real-time capability of coordinated control. Aiming to enhance the
coordinated control effectiveness of RBS and ABS within the electro-hydraulic composite braking
system, this paper proposes a coordinated control strategy based on explicit model predictive control
(eMPC-CCS). Initially, a comprehensive braking control framework is established, combining offline
adaptive control law generation, online optimized control law application, and state compensation to
effectively coordinate braking force through the electro-hydraulic system. During offline processing,
eMPC generates a real-time-oriented state feedback control law based on real-world micro trip
segments, improving the adaptiveness of the braking strategy across different driving conditions. In
the online implementation, the developed three-dimensional eMPC control laws, corresponding to
current driving conditions, are invoked, thereby enhancing the potential for real-time braking strategy
implementation. Moreover, the state error compensator is integrated into eMPC-CCS, yielding a
state gain matrix that optimizes the vehicle braking status and ensures robustness across diverse
braking conditions. Lastly, simulation evaluation and hardware-in-the-loop (HIL) testing manifest
that the proposed eMPC-CCS effectively coordinates the regenerative and hydraulic braking systems,
outperforming other CCSs in terms of braking energy recovery and real-time capability.
Citation: Chu, L.; Li, J.; Guo, Z.;
Jiang, Z.; Li, S.; Du, W.; Wang, Y.;
Keywords: four-wheel hub drive electric vehicle; electro-hydraulic composite braking system;
Guo, C. RBS and ABS Coordinated coordinated control strategy (CCS); explicit model predictive control (eMPC); error compensator
Control Strategy Based on Explicit
Model Predictive Control. Sensors
2024, 24, 3076. https://doi.org/
10.3390/s24103076 1. Introduction
Academic Editor: Mohammed Chadli Recently, with advancements in vehicle control systems, there has been a steady
increase in attention to the coordinated control of electro-hydraulic composite braking
Received: 6 April 2024 systems [1,2]. In electric vehicle braking, the real-time monitoring and processing of data
Revised: 4 May 2024
from multiple sensors enable the coordinated control of the regenerative braking system
Accepted: 9 May 2024
(RBS) and anti-lock braking system (ABS), effectively controlling the distribution of braking
Published: 12 May 2024
torque for each wheel based on parameters like battery status and vehicle speed [3–5].
However, the computational burden of data processing and control strategy in the elec-
tronic control unit (ECU) of the vehicle can adversely affect its real-time implementation
Copyright: © 2024 by the authors.
potential. To ensure real-time application capability, robustness, and enhanced braking
Licensee MDPI, Basel, Switzerland. energy recovery efficiency, developing an advanced coordinated control strategy (CCS)
This article is an open access article is crucial.
distributed under the terms and The reported CCSs can be categorized into two groups based on their objectives:
conditions of the Creative Commons optimization-oriented control CCSs [6,7] and real-time capability-oriented CCSs [8,9].
Attribution (CC BY) license (https:// Optimization-oriented control CCSs rely on offline methods to generate reference quantities
creativecommons.org/licenses/by/ and trajectories for achieving optimized control performance, such as dynamic program-
4.0/). ming (DP) [10,11] and particle swarm optimization (PSO) [12,13]. Given prior driving

Sensors 2024, 24, 3076. https://doi.org/10.3390/s24103076 https://www.mdpi.com/journal/sensors

Sensors 2024, 24, 3076 2 of 27

information, optimization-oriented control CCSs can calculate globally optimized control

sequences in short-term driving scenarios, thereby maximizing the braking control effect.
However, the intensive computational demands of these methods limit their practical ap-
plication. Additionally, acquiring the necessary driving information is challenging, which
restricts the adaptability of these methods across various braking conditions. The real-time
capability-oriented CCSs achieve real-time coordinated control by employing various con-
trol strategies, including proportional integral derivative-based CCS (PID-CCS) [14,15],
linear quadratic regulator-based CCS (LQR-CCS) [16,17], and model predictive control-
based CCS (MPC-CCS) [18–20]. In PID-CCS, the coordinated control of the RBS and ABS
is achieved by tuning the parameters of the PID controller based on the vehicle state and
control objectives. However, during the control process, the PID-CCS can only obtain a
suboptimal control law, making it difficult to precisely and effectively adjust the coordi-
nated control relationship between the RBS and ABS. Regarding the LQR-CCS, the state
gain matrix is formulated, and the wheel braking torque is optimized to attain coordinated
control, aligned with the vehicle model and control objectives. Nevertheless, the LQR
method requires a linearization process to simplify the control problem and cannot adjust
the weights assigned to different time steps within the prediction horizon, alleviating the
effect of coordinated braking force distribution. On the contrary, MPC provides flexibility
by allowing the weights assigned to different time steps within the prediction horizon to be
adjusted, taking into account various variables to make coordinated decisions and obtain
the optimal control sequences [21,22]. Additionally, the application of fast MPC techniques,
as demonstrated by Chu et al. [23] and Meng et al. [24], further enhances the adaptability
and real-time capabilities of MPC in handling dynamic and complex control scenarios.
Nonetheless, achieving real-time online solutions to the optimal control problem poses
challenges for practical vehicle controllers due to the online rolling optimization process
in MPC and the multitude of constraints involved. To overcome this MPC limitation, an
alternative approach named explicit model predictive control (eMPC) is proposed [25–27].
By introducing multi-parameter quadratic programming (mp-QP), eMPC acquires the ex-
plicit solution of state variable and control variable in advance, storing them in the memory
inside the controller, thereby transferring online calculation to the offline part to minimize
the computational burden. Furthermore, eMPC can be easily compiled on an embedded
platform, showing promising real-time application scenarios. Despite its appropriate capa-
bility in real-time deployment, eMPC is primarily for generating state feedback control law
in linear time-invariant (LTI) systems, which does not allow for optimal control throughout
the braking process in highly nonlinear vehicle braking systems [28,29]. Moreover, the
initially set constant state variables in the LTI system undergo changes during practical
vehicle braking, resulting in state error and diminishing the adaptability and robustness of
CCS. Therefore, substantially adjusting the vehicle braking state based on various braking
conditions is an intractable task that should be further investigated.
To substantially utilize information collected during vehicle braking, aiming to re-
duce the adverse impact of state error on the control effect, recent research on optimizing
system state variables has introduced adaptive optimization methods [30,31] and state
error compensation techniques [32–34]. Adaptive optimization methods, such as model
reference adaptive control (MRAC) [35], direct adaptive control (DAC) [36], and fuzzy
adaptive control (FAC) [37], allow for the automatic adjustment of controller parameters in
response to changes in system status and parameters, thereby enhancing control perfor-
mance. However, the intricate relationship between control parameters and the current
state necessitates a comprehensive acknowledgment of the optimized control system or
even a burdensome computational modeling process, presenting considerable implementa-
tion difficulty. Regarding state error compensation methods, encompassing feedforward
error compensating [38,39] and feedback error compensating [40,41], these methods are
directed towards diminishing the negative impact of state error generated during the
braking process on the control effect through error compensation. For feedforward error
compensation, although it can offset system error in advance and effectively suppress
Sensors 2024, 24, 3076 3 of 27

external interference, it relies on accurate mathematical models for prediction and faces
obstacles in performing nonlinear compensation, posing challenges for implementation
in electro-hydraulic composite braking systems. In contrast, feedback error compensation
only requires the construction of an error gain matrix from the practical and desired output
of the system, facilitating state variable adjustment based on this matrix to bolster system
robustness. During the braking process, the vehicle information can be divided into micro
trip segments with distinct braking states. Implementing feedback error compensation
for each micro trip segment enhances the capability of eMPC-CCS to counteract external
interference. To sum up, forming a comprehensive solution is crucial for optimizing the
control effectiveness of RBS and ABS while cooperatively ensuring real-time, robust, and
adaptive capability.
In this context, this paper proposes a novel coordinated control strategy based on
eMPC, namely the eMPC-CCS, aiming to enhance the real-time capability, adaptability,
and robustness of the CCS in the solution. As for the eMPC-CCS, it includes offline control
law generation and online control law invocation. In the offline process, a multitude of
micro trip segments corresponding to braking operations are collected to generate real-
time-oriented state feedback control laws, improving the adaptability of the CCS. During
the online implementation, offline-generated state feedback control laws are invoked
accordingly to form a 3D eMPC explicit solution in the basic eMPC controller, enhancing
the real-time coordinated control of RBS and ABS. Furthermore, a state error compensator
is developed to rectify variations in the state variables and integrated into the basic eMPC
controller to enhance its functionality. The improved eMPC controller navigates through the
3D eMPC explicit solution using the adjusted state variables and delivers a corrective torque
that corrects the braking torque for each wheel, thereby further refining the braking process.
Ultimately, simulation evaluation and hardware-in-the-loop (HIL) testing demonstrate the
outstanding real-time responsiveness, robustness, and elevated efficiency in the braking
energy recuperation of the proposed eMPC-CCS across various braking conditions.
The detailed contributions are illustrated in the following:
(1) A coordinated control strategy for the RBS and ABS based on eMPC is proposed,
which integrates the offline-generated state feedback control law into online real-time
braking to fully enhance the real-time performance of coordinated control.
(2) A 3D eMPC law generation method is proposed, which employs the state feedback
control law generated at each micro trip segment to formulate an explicit solution
for the three-dimensional eMPC, thereby enhancing the adaptability of the control
strategy to various braking conditions
(3) A state variable optimization method based on feedback error compensation is pro-
posed. This method can integrate the gain matrix into the eMPC-CCS to compensate
for the state variable under various braking conditions, improving the ability of the
eMPC-CCS to resist external interference.
The remainder of this paper is organized as follows. The general description of the
vehicle model’s construction is provided in Section 2. Section 3 elaborates on the developed
eMPC-CCS. Section 4 discusses the simulation results. and HIL testing, verifying the
superior performance of the raised strategy. The discussions are provided in Section 5. The
conclusions are represented in Section 6.

2. Modeling
In this paper, a four-wheel hub drive electric vehicle is studied. The corresponding
configuration and detailed parameters are, respectively, presented in Figure 1 and Table 1.
The electro-hydraulic composite system of the studied vehicle is composed of the motor
regenerative braking system and the hydraulic braking system. When the driver presses
the brake pedal for deceleration, the vehicle controller calculates the braking intensity and
the necessary braking force based on the pedal position. Subsequently, the established
CCS guides the motor controller and hydraulic controller in response to the current vehicle
states. The motor controller manages the four motors, applying motor braking torque to
 regenerative braking system and the hydraulic braking system. When the driver presses
the brake pedal for deceleration, the vehicle controller calculates the braking intensity and
Sensors 2024, 24, 3076 the necessary braking force based on the pedal position. Subsequently, the established 4 of 27

CCS guides the motor controller and hydraulic controller in response to the current vehi-
cle states. The motor controller manages the four motors, applying motor braking torque
the wheels,
to the andand
wheels, the the
hydraulic controller
hydraulic supervises
controller the hydraulic
supervises braking
the hydraulic unit, unit,
braking facilitating
facili-
the application
tating of hydraulic
the application brakingbraking
of hydraulic torque torque
to the respective wheel. wheel.
to the respective

Figure 1. The configuration of aa four-wheel

Figure 1. four-wheel hub
hub drive
drive electric
electric vehicle.
vehicle.

Table 1.
Table 1. Vehicle
Vehicle parameters.
parameters.

Parameter
Parameter Sign Sign Value
Value
Vehicle
Vehicle massmass m m 2508
2508 [kg]
[kg]
Wheel
Wheel radius radius Rw Rw 0.377 [m][m]
0.377
The horizontal
The horizontal distance
distancefrom the the
from center of gravity
center to the front
of gravity to theaxle lf
front axle lf 1.506
1.506[m][m]
The horizontal distance from the center of gravity to the rear axle lr 1.483 [m]
The horizontal distance from the center of gravity to the rear axlel lr 1.483
Wheelbase 2.992 [m][m]
Center of Wheelbase
gravity height hg l 2.992
0.58 [m][m]
Wheel moment of inertia Jw 42990.58
[kgm 2]
Center of gravity height hg [m]
Wheel moment of inertia Jw 4299 [kgm2]
2.1. Vehicle Model
The five-degree-of-freedom
2.1. Vehicle Model vehicle dynamics model considering the rotational move-
mentTheof the four wheels and the longitudinal
five-degree-of-freedom movement
vehicle dynamics of theconsidering
model vehicle was the
established in
rotational
this study.
movement of the four wheels and the longitudinal movement of the vehicle was estab-
According to the vehicle dynamics equation, the longitudinal movement can be shown
lished in this study.
as follows:
According to the vehicle. dynamics equation, the longitudinal movement can be
mv x = − Fx f l − Fx f r − Fxrl − Fxrr (1)
shown as follows:
where m is the vehicle mass, v x is the longitudinal speed, and Fx f l , Fx f r , Fxrl and Fxrr are
mv = − Fxfl − Fxfr − Fxrl − Fxrr (1)
the ground braking force received byx the wheels, respectively.
where Themrotational motionmass,
is the vehicle of the vwheels canlongitudinal
is the be written asspeed,
follows:
and Fxfl , Fxfr , F and
x xrl

Fxrr are the .

groundJ braking force received by the wheels, respectively.
w ω ij = Fxij Rw − Tmij − Thij , ij ∈ { FL, FR, RL, RR } (2)
The rotational motion of the wheels can be written as follows:
where Jw is the wheel moment of inertia, Rw is the wheel radius, Tmij is the motor brak-
J ij = Fxij Rw − Tmij − Thij , ij   FL, FR, RL, RR (2)
ing torque received by thew wheels, and Thij is the hydraulic braking torque received by
the wheels.
The main parameters of the vehicle model are presented in Table 1.

2.2. Tire Model

The structural parameters and mechanical characteristics of tires connecting the vehicle
and the road determine the dynamic performance of the vehicle. The magic formula tire
 2.2. Tire Model
The structural parameters and mechanical characteristics of tires connecting the ve-
hicle and the road determine the dynamic performance of the vehicle. The magic formula
Sensors 2024, 24, 3076
tire model is selected in this study to reflect the dynamic behavior of the tire, which5 of
can27
be written as follows:

 Fx = D sin{C arctan[ B − E ( B − arctan( B ))]}

model is selected in this  study to reflect the dynamic behavior of the tire, which can be
C = a0
written as follows: 
 D =  (a1 Fz + a2 Fz )
2
(3)
= D sin{Carctan[ Bλ − E( Bλ − arctan( Bλ))]}

F
 Ba= a3 sin(a4 arctan(a5 Fz )) / (CD )

 x
 C=

 0
D =E µ=( aa16 F + aa7 FFz +
2

Fzz2 + a
2 z) 8 (3)
 B = a3 sin( a4 arctan( a5 Fz ))/(CD )

where  is the slip Eratio,  = (vx − wRw ) / vx ,  is the


2
= a6 Fwhich can
z + a7 F z + bea8 calculated from


road adhesion coefficient, Fx is the longitudinal tire force, and Fz is the vertical tire
where λ is the slip ratio, which can be calculated from λ = (v − wR )/v x , µ is the road
force. In addition, B is the stiffness factor, C is the shape factor,x D iswthe crest factor, E
adhesion coefficient, Fx is the longitudinal tire force, and Fz is the vertical tire force. In
is the curvature a0 , a are the model parameters obtained from the ex-
addition, B is thefactor, and
stiffness factor, C is, the8 shape factor, D is the crest factor, E is the curvature
periment.
factor, and a0 , · · · , a8 are the model parameters obtained from the experiment.

Motor Model
2.3. Motor
purpose of
The purpose ofthis
thispaper
paperisistotoinvestigate
investigate
thethe transient
transient response
response of the
of the motor
motor in
in the
the RBS, with the understanding that the dynamic characteristics of the motor
RBS, with the understanding that the dynamic characteristics of the motor will not signif- will not
significantly
icantly impactimpact the coordinated
the coordinated controlcontrol process
process of of
thethe RBSand
RBS andABS.
ABS.The
The relationship
relationship
among torque, speed, and efficiency is described by an efficiency map, which is shown in
Figure 2. The motor efficiency can be expressed by the following:

ηem= η
em = (n(nemem,,TTem
em )) (4)

where nnem Temis the

andTem
em and rotating
is the speed
rotating and
speed motor
and torque,
motor respectively.
torque, respectively.

Figure 2. Motor efficiency map.

The mathematical
The mathematical model
model ofof the
the permanent
permanent magnet
magnet synchronous
synchronous motor,
motor, representing
representing
the first-order
the first-order inertia
inertia link,
link, is
is used
used to
to portray
portray the
the dynamic
dynamic characteristics
characteristics during
duringregenera-
regener-
tive braking:
ative braking:
1
Tm_act = Tm_req (5)
tm s + 1
where Tm_act is the actual motor response torque, Tm_req is the target motor response torque,
and tm is the motor dynamic response constant.

2.4. Battery Model

The effect of temperature on the battery’s operation is ignored, and a simple internal
resistance model is used for simulation, motivated by the complexity of temperature
effects and the importance of Kirchhoff’s law in establishing the battery voltage balance
equation. According to the Kirchhoff law, the battery voltage balance equation can be
shown as follows:
Uoc = U + Ib Rint (6)
where Uoc is the open-circuit voltage, U is the load voltage, Ib is the battery current, and
Rint is the internal resistance of the battery.
Sensors 2024, 24, 3076 6 of 27

During regenerative braking, the charging current of the battery can be calculated by
the following: q
−UOC (t) + 2 ( t ) − 4R p
UOC int m
Ib (t) = (7)
2Rint
where pm is the motor power. Additionally, the state of charge (SOC) of the battery is
calculated through the ampere-hour integration method:
(
Q(t)
SOC = SOC0 + Q0
. (8)
Q(t) = Ib (t)

where SOC0 is the initial value of battery SOC, Q0 is the battery capacity in Ah, and Q(t) is
the variations in battery power.

2.5. Hydraulic Braking System Model

The specific working process of the hydraulic braking system, involving the physical
properties of the fluid and the response characteristics of hydraulic components, were
omitted in this paper as they do not affect electro-hydraulic coordinated control. The
response process is considered equivalent to a first-order inertial hysteresis link:

e−τh s
Th_act = T (9)
th s + 1 h_req

where Th_act and Th_req are the actual braking torque and the target braking torque of the
Sensors 2024, 24, 3076 hydraulic system, respectively, th is the system time constant, and τh is the system7 pure
of 28
lag time.

3. Development of
implementation eMPC-CCS
process of the eMPC-CCS contains offline control law generation and
Tocontrol
online fully excavate the potential
law invocation. of eMPC-based
An illustration control solutions
of the eMPC-CCS is shownin the real-time
in Figures coor-
3 and 4.
dinated control strategy of the RBS and ABS, a comprehensive braking control
As depicted in Figure 3, in offline control law generation, vehicle information with framework,
named
diversethe eMPC-CCS,
braking states, is developed.
collected fromThis
roadapproach strengthensinto
tests, is segmented real-time
microcapability and
trip segments
adaptability
and stored inwhile avoiding
the vehicle control conflict
controller. betweenthe
Subsequently, theeMPC
RBS and ABS. The
algorithm is implementation
applied to these
process
distinct of the eMPC-CCS
micro trip segments,contains offlinethe
generating control law generation
corresponding state and onlinecontrol
feedback controllaws,
law
invocation. An illustration of the eMPC-CCS
which are then ready for online invocation. is shown in Figures 3 and 4.

Figure 3.
Figure 3. The
The illustration
illustration of
of the
the offline
offline control
control law
law generation
generation of
of the
theeMPC-CCS.
eMPC-CCS.
Sensors 2024, 24, 3076 7 of 27

Figure 3. The illustration of the offline control law generation of the eMPC-CCS.

Figure 4. The illustration of the online control law transplantation and state error compensation of
Figure 4. The illustration of the online control law transplantation and state error compensation of
the eMPC-CCS.
the eMPC-CCS.

In
As Figure
depicted4, as for the 3,online
in Figure implementation,
in offline a 3D eMPCvehicle
control law generation, explicit solution iswith
information em-
ployed
diverseto invokestates,
braking the offline control
collected lawroad
from basedtests,
on the
is similarity
segmentedbetween the practical
into micro brak-
trip segments
ing conditions and micro trip segments. The required braking torque for each
and stored in the vehicle controller. Subsequently, the eMPC algorithm is applied to these wheel under
the current
distinct braking
micro condition generating
trip segments, can be allocated according to the
the corresponding ideal
state brakingcontrol
feedback force distri-
laws,
bution strategy, known as the I curve
which are then ready for online invocation. [42]. Ultimately, the eMPC-CCS optimizes the allo-
cation
In of distributed
Figure wheel
4, as for the onlinebraking torque byautilizing
implementation, 3D eMPCthe matched
explicit eMPC
solution explicit
is employed
to invoke the offline control law based on the similarity between the practical braking
conditions and micro trip segments. The required braking torque for each wheel under
the current braking condition can be allocated according to the ideal braking force dis-
tribution strategy, known as the I curve [42]. Ultimately, the eMPC-CCS optimizes the
allocation of distributed wheel braking torque by utilizing the matched eMPC explicit
solution. It employs the state error compensator and basic eMPC controller within the im-
proved eMPC controller to achieve optimal braking torque allocation within the composite
braking system.

3.1. Offline Control Law Generation

In the offline processing of eMPC, the state feedback control law of each braking
segment of the vehicle in the whole braking process is obtained. The offline control law
generation process in eMPC includes several key steps [25]. Firstly, the mathematical
model of the system is established. Secondly, the critical region of the system is determined
by introducing the theory of mp-QP and the Karush–Kuhn–Tucker (KKT) condition. In
each critical region, the control vector can be expressed as an affine function of the state
vector, constituting the control law. This affine function remains constant and allows eMPC
to derive the control law through offline calculation, thereby eliminating the need for
laborious online computation. The calculated optimal control laws are stored in a lookup
table for real-time implementation.

3.1.1. Basic eMPC

Before designing the controller, the fundamentals of eMPC are reviewed and the
explicit law of the control variable is presented in this section. In eMPC, the solution of an
optimization problem includes constraints on the control inputs and system states. The
Sensors 2024, 24, 3076 8 of 27

general nonlinear optimization problem in the finite time domain [ts , tm ] can be defined as
a minimization constraint on the cost function:
Z tm
J ( x, u, p) , L( x (t), u(t), p(t0 ))dt + T ( x (tm ), p(t0 ), tm ) (10)
ts

where x, u, and p are the state vector, control vector, and parameter vector, respectively.
In addition, L is the stage cost, and T is the terminal cost. The problem is subject to
inequality constraints:
xmin < x (t) < xmax (11)

umin < u(t) < umax (12)

G ( x ( t ), u ( t ), p ( t s ), t ) ≤ 0 (13)
The ordinary differential equations (ODEs) describing the system dynamics represent
the equality constraints:
d
x ( t ) = f ( x ( t ), u ( t ), p ( t s ), t ) (14)
dt
In the process of formulating the control problem, the prediction horizon is defined
as tq = Np tr , where Np is the number of prediction steps and tr is the discretization
interval of the internal model. It is assumed that the control input u[ts , tm ] is calculated by
u(t) = ξ (U, t), where U is the control parameter vector. The optimal control problem is
now in its mp-NLP generic form:

J ∗ ( x (ts ), p(ts )) = min J ( x (ts ), U, p(ts ))

U (15)
s.t.G ( x (ts ), U, p(ts )) ≤ 0

where p includes the system and controller parameters, which are considered constant for
the duration of the prediction horizon. One additional vector is defined as x p (ts ) ∈ Rn p ,
where n p = n + d, i.e., n p is the sum of the number of state n and the size of parameter d:

x p (ts ) = [ x (ts ), p(ts )] (16)

Hence, based on Equation (16), the cost function can be reformulated as follows:

J ∗ x p (ts ) = min



J x p ( t s ), U
U∗
(17)
s.t.G x p (ts ), U ≤ 0

In this study, the mp-QP problem is adopted to solve the mp-NLP problem by its
approximation [43,44]. The mp-NLP in Equation (17) is linearized around a predefined
point ( x p,0 , U0 ) by utilizing Taylor expansion. Therefore, the cost function is approximated
with a quadratic function and the constraints assume a linear formulation.
 
J0 ( x p , U ) , 12 (U − U0 ) T H0 (U − U0 ) + C0 + ( x p − x p,0 ) T F0 (U − U0 ) + Y0 ( x p )
(18)
s.t.G0 (U − U0 ) ≤ W0 + S0 ( x p − x p,0 )

By evaluating at the linearization point ( x p,0 , U0 ), the different terms in Equation (18)
can be computed as follows:

2

T
H0 , ∇UU J ( x p,0 , U0 )C0 , ∇U J ( x p,0 , U0 ) (19)

 T  T 
1  2 2
F0 , ∇ x p U J ( x p,0 , U0 ) + ∇Ux p J ( x p,0 , U0 ) (20)
2
Sensors 2024, 24, 3076 9 of 27

Y0 ( x p ) , 12 ( x p − x p,0 ) T ∇2x p x p J ( x p,0 , U0 )( x p − x p,0 )

 T (21)
+ ∇ x p J ( x p,0 , U0 ) ( x p − x p,0 ) + J ( x p,0 , U0 )

T
G0 , ∇U G ( x p,0 , U0 ) (22)

W0 , − G ( x p,0 , U0 ) (23)

 T
S0 , − ∇ x p G ( x p,0 , U0 ) (24)

The mp-QP problem is employed to compute local approximations of the original

mp-NLP problem in the exploration space. This is represented as multiple hyperrectangles,
on which single mp-QP problems are solved. Each hyperrectangle is further partitioned
into polyhedra, i.e., the CRregions of the mp-QP problem. Finally, the mp-QP solution is
represented as a piecewise affine (PWA) function, which is determined by a finite number
of regions dividing state variables. Given that eMPC is designed for LTI systems, achieving
improved control effects in the coordinated control of the RBS and ABS requires both offline
and online optimizations.

3.1.2. Internal Prediction Model

In the basic design of the eMPC controller, only the control process of the left front
(FL) brake component is depicted, with the other wheels following a similar pattern.
In the internal quarter-vehicle model, the slip ratio of the left front wheel, λ FL , is
defined as follows:
v x (t) − ω FL (t) Rw
λ FL (t) = (25)
v x (t)
where v x is the longitudinal speed, ω FL is the wheel velocity, and Rw is the wheel radius.
The time derivative of λ FL can be expressed as follows:
 
d
d d
v x ( t ) − R d
ω
w dt FL ( t ) ( v x ( t ) − ω FL ( t ) R w ) dt v x ( t )
λ FL = dt − (26)
dt v x (t) v2x (t)

where the wheel rotation dynamics equation can be expressed as follows:

d 1
ω FL (t) = ( Fx,FL Rw − TCA,FL + ∆TFL (t)) (27)
dt Jw

where Jw is the wheel moment of inertia, and TCA,FL is the distributed braking torque ac-
cording to the I-curve, which is kept constant over the prediction horizon. The longitudinal
force balance of the quarter-vehicle model associated with the wheel under consideration
can be shown as follows:
d 1
v x (t) = − Fx,FL (28)
dt mF
where Fx is the longitudinal tire force, which can be calculated using the simplified Pacejka
magic formula (MF) model [45], as follows:

Fx,FL = µ x,FL Fz,F
(29)
µ FL = D sin{Carctan[ Bλ FL − E( Bλ FL − arctan( Bλ FL ))]}

where Fz,F is the vertical tire load, considered a constant, and µ x,FL is the longitudinal
tire–road friction coefficient. Additionally, B, C, D and E are the MF parameters, which are
shown in Table 2.
Sensors 2024, 24, 3076 10 of 27

Table 2. Intermodal parameters.

Symbol Parameters Value

mi Apparent front (rear) corner mass (kg) 625 (629)
Fz,i Front (rear) tire vertical load (N) 6125 (6164)
B MF coefficient: stiffness factor 21.52
C MF coefficient: shape factor 1.4
D MF coefficient: peak value 0.45
E MF coefficient: curvature factor −0.28
Np Prediction step 3
tr Sampling time (s) 0.004

Therefore, the time derivative of Equation (25) can be rewritten as follows:

d (λ (t) − 1) Fx,FL Rw
λ FL (t) = FL − ( Fx,FL (t)Rw − TCA,FL (t) + ∆TFL (t)) (30)
dt v x (t) mF Jw v x (t)

Due to the difference in inertia, the longitudinal dynamic of the vehicle is much slower
than the rotational dynamic of the wheels. Therefore, the vehicle speed is considered a
slowly changing parameter.
An integral process is incorporated into the prediction model to tackle the steady-state
error and model uncertainty. Hence, the model includes e FL (t), which is the integral of the
error between the actual slip rate, λ FL , and the reference slip rate, λ FL re f .

d
e FL (t) = λ FL (t) − λ FL re f (31)
dt
For each sampling moment, the state equation of the system after linearization can be
obtained as follows:
.
x (t) = A(t) x (t) + B(t)u(t) (32)
where x (t) ∈ Rn , u(t) ∈ Rm

Rw 2
" #
Fx,FL (t) (λ FL (t)−1) dFx,FL (t) dFx,FL (t)
v x (t)·m F
+ v x (t)·m F
· dλ FL (t)
− Jw ·v x (t)
· dλ FL (t)
0
A(t) = (33)
1 0
" #
− Jw ·Rvwx (t)
B(t) = (34)
0
dF (t)
where dλx,FL(t) denotes the derivative of Fx,FL (t) to λ FL (t).
FL
The continuous-time state space representation, Equation (32), is discretized with a
sampling time.
x (k + 1) = Ad [k ] x (k) + Bd [k ]u(k) (35)
where
A1d [k ] 0
 
Ad [k ] = (36)
ts 1
 2 
A [k]
Bd [k ] = d (37)
0

where Aid [k], i = 1, 2 indicates the varying elements due to v x (t), and k is the sampling
time index.
In the prediction model, the statehvector, input vector,
i and parameter vector are
x FL = [λ FL , e FL ], u = [∆TFL ], and p = v x , TCA,FL , λ FL re f , respectively. The prediction
horizon, t p = Np tr (i.e., Np is prediction steps; tr is sampling time), is selected for the
current implementation. The problem includes five parameters (a 5D-eMPC problem),
Sensors 2024, 24, 3076 11 of 27

h i
i.e., x p,FL = λ FL (ts ), e FL (ts ), v x (ts ), TCA,FL (ts ), λ FL re f , and four decision variables, i.e.,
UFL = [∆TFL (ts ), ∆TFL (ts+1 ), ∆TFL (ts+2 ), ∆TFL (ts+3 )]. The horizon control input that is
applied to the system is u = ∆T (ts ), which will be indicated as u in the remainder. The
5D-eMPC problem will be referred to as eMPC5D in the remainder.
Sensors 2024, 24, 3076 At each moment throughout the vehicle braking process, the state variables at 12 that
of 28
moment are aligned with the previously gathered braking segments, facilitating the genera-
tion of the state feedback control laws depicted in Figure 5 within the basic eMPC controller.
Subsequently, the generated
controller. Subsequently, thecontrol laws control
generated are prepared for online
laws are invocation,
prepared reducing
for online the
invocation,
computational burden of the control problem.
reducing the computational burden of the control problem.

Figure5.5.The
Figure TheGeneration
Generationof
ofstate
statefeedback
feedbackcontrol
controllaws
lawsininthe
thebasic
basicempc
empccontroller.
controller.

3.1.3.
3.1.3.Cost
CostFunction
Functionand andConstraints
Constraints
During
During the offline optimizationprocess,
the offline optimization process, the
the optimal
optimal adjustment
adjustment of the required
of the required braking
brak-
torque for the FL wheel is achieved by minimizing the
ing torque for the FL wheel is achieved by minimizing the cost function. The changes incost function. The changes in braking
torque
braking are applied
torque are to the wheels
applied to theto quickly
wheels to achieve the desired
quickly achieve slip rateslip
the desired of the
rateFLof wheel
the FL
throughout
wheel throughout the ABSthe braking
ABS braking process.process.
Moreover, the costthe
Moreover, function includes
cost function the influence
includes the in-
of braking
fluence of torque
brakingontorque
the wheel on the to wheel
minimize variationsvariations
to minimize in wheel brakingin wheeltorque.
brakingThe cost
torque.
function can be written as follows:
The cost function can be written as follows:

q1  t f  q1 
Z t
" #
2 2q r pp1 1
2 p2
J FL 2= λ FL (t2) (−FL +) +2 e FL ( )
f q r ref 2re f p2

2 2 u 2 2
JFL = (t )re−fFL ref
λ FL 2
(
e t
FL
) ( t+) 2
+ 2
u
∆T T
FL (
FL
t
(t)) 2
 dt ++ 22
 λ
FL (t
FL f
()t −
f )− λ + e (t e) 2 (t f )
+
2 FL 2f FL
(38)
(38)
tk w1 t k
 w1 w2 w2 2
w wu
u
2
 ww1 1 w2 w2
where q1 = 5, q2 = 60, ru = 10, p1 = 5, p2 = 60, w1 = 0.1, w2 = 0.1, and wu = 3000. In
q = 5 q2 = 60 , ru = 10 , p1 = 5 , p2 = 60 , w1 = 0.1 , w2 = 0.1 , and wu = 3000 . In addition,
where 1λ FL re, f is
addition, computed from the longitudinal tire force characteristics as a function of
 ref
theFLslip is ratio, by using
computed fromthethe MFlongitudinal
[46]. tire force characteristics as a function of the slip
For the implementation
ratio, by using the MF [46]. of eMPC, some constraints related to the powertrain perfor-
manceFor and vehicle dynamic should be
the implementation of eMPC, some constraints set, which would ensure relatedthat different
to the components
powertrain perfor-
can operate within limits. Thus, the inequality constraints
mance and vehicle dynamic should be set, which would ensure that different components of Equation (38) can be shown
as
canfollows:
operate within limits. Thus, the inequality constraints of Equation (38) can be shown
as follows: λmin < λ FL (t) < λmax (39)

min  FL (t )  max (39)

Tmin  TFL (t )  Tmax (40)

Tmax = TCA,FL Tmin = 0 min max

where and , while and are used as tuning parameters.
Sensors 2024, 24, 3076 12 of 27

Sensors 2024, 24, 3076 ∆Tmin < ∆TFL (t) < ∆Tmax (40)
13 of 28

where ∆Tmax = TCA,FL and ∆Tmin = 0, while λmin and λmax are used as tuning parameters.

3.2. Online Implementation

Building upon the control laws generated in the preceding offline process, the online
implementation invokes the eMPC solution generated by the three-dimensional three-dimensional eMPC
control law generation method for the adaptability of the CCS through through various
various braking
conditions. To improve the ability of the eMPC-CCS to resist external interference, a state
error compensator is developed, optimizing the control logic of the control problem. The
illustration of the online implementation of the eMPC-CCS is shown in Figure 6. During
the control
control process,
process,the
thestate
stateerror
error compensator
compensator is used
is used to optimize
to optimize the state
the state variables,
variables, and
and
then then the optimized
the optimized state state variables
variables are transmitted
are transmitted to theto3DtheeMPC
3D eMPC solution
solution in theinbasic
the
basic eMPC controller to obtain a correction torque to correct the braking torque
eMPC controller to obtain a correction torque to correct the braking torque of each wheel. of each
wheel. The corrected
The corrected brakingbraking
torquetorque prioritizes
prioritizes the requirements
the requirements of theofmotor
the motor
beforebefore
beingbeing
dis-
distributed to both the motor and the hydraulic braking system, thus ensuring
tributed to both the motor and the hydraulic braking system, thus ensuring the enhanced the enhanced
adaptability
adaptability and
and robustness
robustness of of the
the CCSs.
CCSs.

Figure 6. The illustration of the online implementation of the eMPC-CCS.

Figure 6. The illustration of the online implementation of the eMPC-CCS.

3.2.1. Explicit Solution of RBS and ABS Coordinated Control

Regarding the comment about eMPC5D 5D,, the
the representation
representation of the explicit solution is
presented in Figure 7. Figure 7 illustrates that the three-dimensional eMPC control law
generation
generation method
methodpartitions
partitionsthe state
the input
state space
input intointo
space polyhedral regions,
polyhedral each connected
regions, each con-
via PWA
nected viasystems. The principle
PWA systems. of PWA of
The principle in PWA
eMPCininvolves segmenting
eMPC involves the state the
segmenting space of
state
the system and using linear models and corresponding optimal control
space of the system and using linear models and corresponding optimal control laws laws within each
region
within to
eacheffectively
region tocontrol nonlinear
effectively controlsystems.
nonlinear The offlineThe
systems. control lawcontrol
offline generation stage
law gener-
precomputes and stores optimal
ation stage precomputes control
and stores laws, while
optimal controlthelaws,
onlinewhile
control
thelaw invocation
online controlstage
law
invocation stage dynamically selects and applies the appropriate control law based on the
current system state, achieving real-time control of the system.
Invoking these offline generated control laws across diverse state input domains fa-
cilitates optimal control input determination, amalgamating into an eMPC explicit solu-
tion. To visualize the three-dimensional eMPC explicit solution in Figure 7, three
 Sensors 2024, 24, 3076 14 of 28

Sensors 2024, 24, 3076 13 of 27

x p , FL (2)
parameters are treated as constants: the integral of the slip rate error, , is 0, the
x (3) x (5)
vehicle speed, p , FL , is 100 km/h, and the reference slip ratio, p , FL , is set to 0.15.
dynamically x p , FL and
(4) applies the appropriate control law based on the current system
Furthermore,selects is the demand torque calculated using the I-curve and the red
state,
dashedachieving real-time as
line is indicated control of theslip
a constant system.
ratio reference in the legend.

Figure7.7.The
Figure Theillustration
illustrationofofthe
thethree-dimensional
three-dimensionaleMPC
eMPCcontrol
controllaw
lawgeneration
generationmethod.
method.

Therefore,
Invoking the offline
these solution consists control
generated of threelaws
planes, including
across diverse the following:
state input domains facili-
tates
(1) A plateau of zero-control input for the low slip ratio indicates an explicit
optimal control input determination, amalgamating into an eMPC solution.
input lower con-
To visualize the three-dimensional eMPC explicit solution in Figure
straint in Figure 7. According to Equations (39) and (40), the torque correction must 7, three parameters
are treated as constants: the integral of the slip rate error, x p,FL (2), is 0, the vehicle speed,
be positive.
x p,FL (3), is 100 km/h, and the referencexslip(1) ratio, x p,FL (5), is set to 0.15. Furthermore,
x(2) An inclined plane, torque
parallelcalculated
to the p , using axis, indicates an the
input
redupper constraint
FL
p,FL ( 4 ) is the demand the I-curve and dashed line is
in Figure 7. According to Equations
indicated as a constant slip ratio reference in the legend. (39) and (40), the regulating torque cannot be
larger thanthe
Therefore, thesolution
demand torque.of three planes, including the following:
consists
(3) Another inclined plane is the non-saturated feedback control input.
(1) A plateau of zero-control input for the low slip ratio indicates an input lower constraint
In Figure
in the state 7. prediction
Accordingprocess of eMPC,
to Equations (39)theand
precision
(40), theof torque
the state variable markedly
correction must be
impacts the
positive. efficacy of vehicle control. Due to the eMPC solution being obtained through
linearization,
(2) An inclined it encounters limitations
plane, parallel in the
to the x p,FL (1) mathematical
axis, indicatesstate function
an input upper that represents
constraint in
the change
Figure 7. According to Equations (39) and (40), the regulating torque cannot becharac-
in braking torque, making it challenging to fully reflect the dynamic larger
teristics
thanofthe thedemand
electro-hydraulic
torque. composite braking system. Therefore, a state error com-
pensator
(3) Another that inclined
can address state
plane error
is the within the predictive
non-saturated feedback horizon is crucial for enhancing
control input.
the effectiveness of the eMPC controller, which will be elaborated
In the state prediction process of eMPC, the precision of the state variable upon in the subsequent
markedly
section.
impacts the efficacy of vehicle control. Due to the eMPC solution being obtained through
linearization, it encounters limitations in the mathematical state function that represents the
3.2.2. Online
change Statetorque,
in braking Optimization
makingBased on Feedback
it challenging Error
to fully Compensation
reflect the dynamic characteristics
of theTo precisely anticipate
electro-hydraulic changes
composite in braking
braking torque
system. within athe
Therefore, state predictive horizon, an
error compensator
that can address
algorithm [47] isstate error within
proposed the predictive
to compensate for thehorizon x (t )
is crucial forinenhancing
state variable Equation the
(32).effec-
This
tiveness of the eMPC controller, which will be elaborated upon in the
algorithm enhances the robustness of the eMPC controller against parameter changes.subsequent section.

3.2.2. Due to the

Online vehicle
State speed being
Optimization fixed
Based onduring
Feedback a sampling time, the matrices Ad [k ] and
Error Compensation
Bd [k ] in Equation (35) are regarded as the constant matrices Ad and Bd , and the sys-
To precisely anticipate changes in braking torque within the predictive horizon, an
tem is transformed
algorithm into antoLTI
[47] is proposed system. The
compensate forcorresponding
the state variable statex (space representation
t) in Equation can
(32). This
be shownenhances
algorithm as follows:
the robustness of the eMPC controller against parameter changes.
Due to the vehicle speed being fixed during a sampling time, the matrices Ad [k] and
x(k + 1) = Ad x(k ) + Bd u(k ) (41)
Bd [k] in Equation (35) are regarded as the constant matrices Ad and Bd , and the system
is transformed into an LTI system. The corresponding state space representation can be
However, maintaining a constant longitudinal velocity, vx , in an actual driving sit-
shown as follows:
uation is challenging. In this paper, v as a slowly varying parameter, vx (41) (t ) .
x (k + 1)x =is A regarded
d x ( k ) + Bd u ( k )
The time-varying matrix Ad [k ]
However, maintaining a constant can be determined
longitudinal velocity, as follows:
v x , in an actual driving situa-
tion is challenging. In this paper, v x is regarded Nas a slowly varying parameter, v x (t).
d +   x , i xas
v

The time-varying matrix Ad [k]Acan =A

d [ k ]be determined , i (kfollows:
) (42)
i =1

Nv
Ad [k ] = Ad + ∑ ∆ x,i α x,i (k) (42)
i =1
Sensors 2024, 24, 3076 14 of 27

where ∆ x,i ∈ Rn×n , αi (t) ∈ R, and Nv represent the number of the time-varying elements
in the matrix Ad [k].
In this study, Ad [k] is assumed to be a time-varying matrix with sufficiently slow
variation, but the mechanisms of its evolution are not clear. Additionally, the effect of the
time-varying matrix Bd [k] is neglected, and it is treated as a constant matrix. Thus, Ad , Bd
and the matrix ∆ x,i are known. However, the function α x,i (k ), describing the time-varying
part of Ad [k], remains unclear.
Equation (35) can be rewritten as follows:

Nv
x ( k ) − A d x ( k − 1) = ∑ [∆x,i αx,i (k − 1)x(k − 1)] (43)
i =1

Assuming that Dx (k − 1) = [∆ x,1 x (k − 1), . . . , ∆ x,Nv x (k − 1)], ∆ x,i x (t − 1)(1 ≤ i ≤ Nv )

is a column vector in Rn , Equation (43) can be rewritten as follows:

x (k) − Ad x (k − 1) = Dx (k − 1)[α x,1 (k − 1), . . . , α Nv (k − 1)] (44)

In Equation (44), the previous states, x (k − 1), and input, u(k − 1), are known via mea-
surement, so the current state, x (k), can be calculated. To simplify the study in this section,
the ranks of Dx (k − 1) and [ Dx (k − 1), x (k ) − Ad x (k − 1) − Bd x (k − 1)] are considered to
be equal for all k ≥ 1. Therefore, based on the Rouché–Capelli theorem [48], Equation (44)
has a unique solution, α x,i (k − 1)(1 ≤ i ≤ Nv ), for all k ≥ 1.
As a result, Ad [k − 1] can be obtained from the most recent state, x (k − 1), control
history, u(k − 1), and current state, x (k). To obtain the state gain matrix, L(k), for compen-
sating the state vector, the equation is expressed as follows:

x̂ (k + 1) = Ad x̂ (k ) + Bd u(k ) (45)

where x̂ (k) denotes a compensated state vector at the time instant k. Ad is considered to be
invertible, and x̂ (k) in Equation (45) is replaced by L(k) x (k), where L(k) can be shown as
follows:
L(k ) = A− 1
d Ad (k ) (46)
Then, x̂ (k + 1) = x (k + 1) in Equations (35) and (45), meaning the compensated state
vector can be identical to the state vector obtained by the time-varying system matrix,
Ad [k], in Equation (35). By substituting Ad [k − 1] from Equation (44) for Ad [k] in Equation
(46), L(k − 1) can be calculated. With sufficiently slow time-varying parameters, it can be
assumed that L(k ) ∼ = L(k − 1). Consequently, the compensated state vector can be shown
as follows:
x̂ (k) = L(k) x (k ) ∼
= L ( k − 1) x ( k ) (47)
The online implementation structure of the state error compensation is represented in
Table 3.
Moreover, considering the research target in the study is to ensure braking stability
while improving the braking energy recovery rate of electric vehicles, the corresponding
system parameter vector, x p,FL , is rewritten as follows:
 
x̂ p,FL = λ̂ FL (tk ), e FL (tk ), v x (tk ), TCA,FL (tk ) (48)

where λ̂ FL (tk ) = λ̂ FL (tk ) + A− 1

d Ad ( tk ), and other system parameters are the same as
in x p,FL .
 Step 1: Measure the state vector x ( k ) of the system Equation (35).
Step 2: If k = 0 then Ad (−1) = Ad . Otherwise, Ad (−1) is calculated by using Equation (35) with
x (k ) , x(k − 1) , u (k − 1) if ranks of Dx (k − 1) and [Dx (k − 1), x(k ) − Ad x(k − 1) − Bd x(k − 1)] are equal
Sensors 2024, 24, 3076 for all k  1 . 15 of 27
Step 3: Obtain L(k − 1) by substituting Ad [k − 1] from Equation (44) for Ad [k ] in Equation(46),
and estimate xˆ( k ) by Equation (47) with x ( k ) and L(k − 1) . Note that, by Equation (46),
Table 3. Implementation structure of the error compensation.
L(−1) = Ad−1 Ad (−1) = Ad−1 Ad = I .
Initialization:
Step 4:From the system
Evaluate the functionEquationu (k )(35),
by set
eMPCAd [solution
0] = Ad , in [0] = B7d .based
BdFigure For the xˆ(k ) and
onsystem Equation
apply (41)
u (k )
obtain the optimal look-up map, i.e., the eMPC solution in Figure 7 by solving the mp-QP
to the
problem system (10).
Equation Equation
Set the(35).initial condition x (0) of the system Equation (35) and k = 0.
Step 5: Set k = k + 1
Step 1: Measure the state vector x1.
, and go to Step (k) of the system Equation (35).
Step 2: If k = 0 then Ad (−1) = Ad . Otherwise, Ad (−1) is calculated by using Equation (35) with
Moreover, − 1), u(k − 1the
x (k), x (kconsidering ) if ranks
research x ( k − 1in
of Dtarget ) and
the study is to ensure braking stability
[ Dx (k − 1), x (k) − Ad x (k − 1) − Bd x (k − 1)] are equal for all k ≥ 1.
while improving the braking energy recovery rate of electric vehicles, the corresponding
Step 3: Obtain L(k − 1) by substituting Ad [k − 1] from Equation (44) for Ad [k] in Equation (46),
and estimate
system parameter x̂ (k) byx pEquation
vector, , FL (47) withas
, is rewritten k) and L(k − 1). Note that, by Equation (46),
x (follows:
L(−1) = Ad Ad (−1) = A−
−1
d
1
A d = I.
Step 4: Evaluate the functionxˆup ,(FLk) =byeMPC ˆ (t ),solution (tk )  on x̂ (k) and apply u(k(48)
 FL k eFL (tk ), vin x (tFigure
k ), TCA, FL
7 based ) to
the system Equation (35).
Step 5:ˆSet (t k) =
= kˆ +(t1,)and
+ A−go
1
A to(t Step
) 1. x p , FL
where FL k FL k d d k
, and other system parameters are the same as in .
The architecture of the improved eMPC controller, which consists of the basic eMPC
The and
controller architecture
the stateof the compensator,
error improved eMPC controller,
is shown which8.consists
in Figure The basic of eMPCthe basic eMPC
control-
controller and the state error compensator, is shown in Figure 8. The basic
x p , FL (1) x p , FLeMPC
(4) controller
ler can generate critical regions in Figure 8 when only and vary and
can generate critical regions in Figure 8 when only x p,FL (1) and x p,FL (4) vary and the
the other parameters are fixed. Each region owns its unique sequence of optimal vectors,
other parameters are fixed. Each region owns its unique sequence of optimal vectors, and
and the controller explicitly chooses a region based on the parameter values. The state
the controller explicitly chooses a region based on the parameter values. The state error
error compensator
compensator employsemploys feedback
feedback errorerror compensation
compensation according
according to the to the variation
variation in the in the
system
ˆ p , FL
x
system i.e., x̂ p,FL
matrix,matrix, i.e.,in Equation (48), and
in Equation enhances
(48), the robustness
and enhances the robustnessof the controller against
of the controller
parameter
against variation.
parameter variation.

Figure 8. The architecture of the improved eMPC.

Equation (48) dictates that compensating the state variable in Figure 8 involves straightfor-
ward matrix multiplication considering the previous state and input. This compensator avoids
the necessity for modifying the critical regions of eMPC concerning parameter variation.
In summary, a comprehensive braking control framework is established by combin-
ing offline adaptive control law generation, online optimal control law application, and
state compensation.

4. Comparison of Simulation Results

In this section, based on the studied vehicle and the proposed structure of the electro-
hydraulic composite braking system, a MATLAB/Simulink simulation platform is estab-
lished for a four-wheel hub drive electric vehicle. The coordinated control of the electro-
hydraulic composite braking system emphasizes regulating the electric and hydraulic
Sensors 2024, 24, 3076 16 of 27

braking force to ensure braking safety in both regular and emergency braking conditions.
Additionally, the regenerative braking system should provide a certain amount of electric
braking force during the test, and the test conditions need to include regular braking and
emergency braking conditions. Therefore, the performance of the proposed eMPC-CCS
is evaluated in comparison with that of conventional CCSs, such as the PID-CCS, LQR-
CCS, and MPC-CCS. Evaluations include assessing the performance of the driver in an
integrated braking operation, starting with low-intensity braking and then transitioning
to high-intensity braking, to verify the effectiveness of the proposed eMPC-CCS. Another
assessment involves emergency braking on joint and bisectional roads to verify the braking
stability and robustness of the proposed eMPC-CCS. The initial battery SOC is set to 0.8.
The key parameters of the vehicle model are shown in Tables 1 and 2. According to the
above analysis, the simulated braking conditions selected in this section are as follows:
• The vehicle performs integrated braking on the road with a high-adhesion coefficient
of 0.8 and the initial braking speed is 100 km/h;
• The vehicle performs integrated braking on the road with a low-adhesion coefficient
of 0.3 and the initial braking speed is 50 km/h;
• The vehicle performs emergency braking on the joint road with an adhesion coefficient
that changes from 0.2 to 0.8, and the initial braking speed is 60 km/h;
• The vehicle performs emergency braking on the bisectional road with an adhesion
coefficient of 0.3 on the left and 0.8 on the right, and the initial braking speed is
60 km/h.

4.1. Simulation Analysis of Integrated Braking Conditions

As shown in Figures 9 and 10, the integrated braking conditions are simulated and
analyzed for vehicle braking on the high- and low-adhesion coefficient road, respectively.
In addition, the driver is set to perform an integrated braking operation, i.e., small-intensity
braking with a value of 0.1 at 0–1 s first, which changes to large-intensity braking with a
value of 1 after 1 s. When the vehicle brakes on this road, the front-left wheel is selected to
be analyzed, i.e., the simulation results of the four coordinated control strategies controlling
whole braking are demonstrated by the front-left wheel motion.
The vehicle braking simulation results on the dry asphalt road with an adhesion
coefficient of 0.8 are shown in Figure 9. The initial braking speed is 100 km/h and the
battery SOC is 0.8. In Figure 9d–f, within 1 s from braking, according to the current
vehicle status, the braking mode analysis module judges the braking system working
mode to be conventional regenerative braking. At this time, the regenerative braking
system can completely provide the whole vehicle’s demanded braking torque. After 1 s, the
braking intensity is more than 0.2, and the eMPC-CCS distributes the front and rear axle
braking force according to the I-curve and prioritizes the motor to provide the demanded
braking force. When the demanded braking force is larger than the maximum braking
force that can be provided by the motor, the motor outputs the current maximum braking
force, and the remaining demanded braking force is provided by the hydraulic braking
system. In Figure 9a–c, with the increase in braking force applied to the wheels, the braking
system enters into the RBS and ABS coordinated control mode. Then, the RBS and ABS
coordinated control strategy is triggered to control the wheels for antilock control, at which
point the wheels engage in emergency braking. When the wheel reaches the ABS-triggering
condition, the eMPC-CCS calculates the wheel demand braking torque according to the
vehicle state parameters, to maintain the slip ratio at the optimal road slip ratio. It can
be seen from Figure 9a that during the emergency braking, compared with the other
three control strategies, the eMPC-CCS can quickly stabilize the slip rate around 0.15 with
minimal fluctuation. This means that the adhesion between the wheels and the road is
more stable, and makes the brake system control the wheels more accurately. When the
vehicle speed drops to 10 km/h, the braking torque of the motor decreases to 0 Nm and the
motor exits the braking process. After that, the vehicle is not suitable for anti-lock braking,
Sensors 2024, 24, 3076 17 of 27

Sensors 2024, 24, 3076 so the coordinated control strategy controls the hydraulic braking torque to increase 18
rapidly
of 28
to complete the final parking brake.

Sensors 2024, 24, 3076 19 of 28

Figure
Figure 9. Simulation
Simulation results
resultsofoffour
four coordinated
coordinated control
control strategies
strategies for high-adhesion-coefficient
for high-adhesion-coefficient road.
road.

The vehicle braking simulation results on the dry asphalt road with an adhesion co-
efficient of 0.8 are shown in Figure 9. The initial braking speed is 100 km/h and the battery
SOC is 0.8. In Figure 9d–f, within 1 s from braking, according to the current vehicle status,
the braking mode analysis module judges the braking system working mode to be con-
ventional regenerative braking. At this time, the regenerative braking system can com-
pletely provide the whole vehicle’s demanded braking torque. After 1 s, the braking in-
tensity is more than 0.2, and the eMPC-CCS distributes the front and rear axle braking
force according to the I-curve and prioritizes the motor to provide the demanded braking
force. When the demanded braking force is larger than the maximum braking force that
can be provided by the motor, the motor outputs the current maximum braking force, and
the remaining demanded braking force is provided by the hydraulic braking system. In
Figure 9a–c, with the increase in braking force applied to the wheels, the braking system
enters into the RBS and ABS coordinated control mode. Then, the RBS and ABS coordi-
nated control strategy is triggered to control the wheels for antilock control, at which point
the wheels engage in emergency braking. When the wheel reaches the ABS-triggering con-
dition, the eMPC-CCS calculates the wheel demand braking torque according to the vehi-
cle state parameters, to maintain the slip ratio at the optimal road slip ratio. It can be seen
Figure
Figure
from 10. Simulation
Figure Simulation results
resultsof
9a that during offour
the fourcoordinated
coordinated
emergency control strategies
control
braking, forwith
strategies
compared low-adhesion-coefficient
for the
low-adhesion-coefficient
other three controlroad.
road.
strategies, the eMPC-CCS can quickly stabilize the slip rate around 0.15 with minimal
The vehicle braking simulation results on a snowy road with an adhesion coefficient
fluctuation. This means that the adhesion between the wheels and the road is more stable,
of 0.3
The arevehicle
shownbraking
in Figure 10. The initialon
simulation braking speed
roadiswith 50 km/h, and the battery
and makes the brake system control results
the wheelsamore snowy accurately. an adhesion
When coefficient
the vehicle speed
SOC is
of 0.3 are 0.8.
shownIn Figure 10d–f, within 1 s from braking, according to the current vehicle
drops to 10 km/h,inthe
Figure 10. The
braking initial
torque braking
of the motorspeed is 50 km/h,
decreases to 0 Nm andand thethe
battery
motor SOCexitsis
status,
0.8. In the braking
Figure 10d–f, mode
within analysis
1 s from module
braking, judges the braking
according to the systemvehicle
current working modethe
status, to
the braking process. After that, the vehicle is not suitable for anti-lock braking, so the co-
be conventional
braking mode braking.
analysis At this
module time,the
judges thebraking
regenerative
system braking
working system can completely
ordinated control strategy controls the hydraulic braking torque tomode to be
increase conven-
rapidly to
provide
tional the demanded
braking. At this braking
time, the torque. After 1braking
regenerative s, the braking
system intensity
can is more than
completely 0.2, and
provide the
complete the final parking brake.
the eMPC-CCS
demanded brakingdistributes the front
torque. After andbraking
1 s, the rear axle braking
intensity is force
more according
than 0.2, and to the I-curve
the eMPC-
and prioritizes the motor to provide the demanded braking force.
CCS distributes the front and rear axle braking force according to the I-curve and priori- At this point, large
fluctuations in the slip ratio occur due to emergency braking
tizes the motor to provide the demanded braking force. At this point, large fluctuations taking place, and the four
in
coordinated control strategies perform anti-lock braking by adjusting
the slip ratio occur due to emergency braking taking place, and the four coordinated con- the braking torque on
the strategies
trol wheels. Inperform
Figure 10a, compared
anti-lock brakingwithbythe other three
adjusting the control
brakingstrategies,
torque onthe theeMPC-CCS
wheels. In
can maintain the slip rate around 0.15 faster and more stably
Figure 10a, compared with the other three control strategies, the eMPC-CCS can maintain with minimal fluctuation.
When
the slipthe
ratevehicle
aroundspeed is lessand
0.15 faster thanmore10 km/h, the wheel
stably with minimal speed drops directly
fluctuation. When theto 0, the
vehi-
cle speed is less than 10 km/h, the wheel speed drops directly to 0, the regenerative brak-
ing system quits braking, and the required braking force is completely provided by the
hydraulic braking system, until complete parking.
By comparing the braking performance indicators among the four coordinated con-
trol strategies, shown in Figure 11, the eMPC-CCS has the maximum energy recovery ef-
Sensors 2024, 24, 3076 18 of 27

regenerative braking system quits braking, and the required braking force is completely
provided by the hydraulic braking system, until complete parking.
By comparing the braking performance indicators among the four coordinated control
strategies, shown in Figure 11, the eMPC-CCS has the maximum energy recovery efficiency
and adhesion factor utilization rate on high- and low-adhesion-coefficient roads. The
MPC-CCS performs a bit worse than the eMPC-CCS but better than the LQR-CCS and
PID-CCS; the PID-CCS has the worst performance. The differences in performance among
these strategies are due to the optimal control logic invocation and the method of achieving
the invoking process. The improved eMPC controller can reasonably distribute the braking
force of each wheel and solidly underpin the control logic update, ensuring the motor is in
a high-efficiency field in different driving conditions. The MPC-CCS with driving condition
Sensors 2024, 24, 3076 identification in the whole braking process can also invoke the optimized control logic but 20 of
struggles to obtain a flexible, timely update in contrast to the eMPC-CCS.

Figure 11. Comparison

Figure 11. of four coordinated
Comparison control
of four strategies
coordinated on high-
control and low-adhesion-coefficient
strategies roads.
on high- and low-adhesion-coeffici
roads.
Tables 4 and 5 present the simulation results of various CCSs, emphasizing energy
recovery efficiency as the
Tables main
4 and evaluation
5 present the criterion forresults
simulation assessing the coordinated
of various control
CCSs, emphasizing ener
effectiveness of the RBS and ABS. As shown in Equation (49), the vehicle speed changes
recovery efficiency as the main evaluation criterion for assessing the coordinated cont
from v0 to veffectiveness
1 during the oftime
theinterval
RBS and[0,ABS.
tb ]. As shown in Equation (49), the vehicle speed chang

from
v0 to v1 Rduring
t0

4
the time interval
[0, tb ] . 
0 ∑
1000( T · ω · η ( T , ω ))/9550 dt
i i i i i
i =1
η=  4
 (49)

t0
m(v200−1000(
v21 )/2 Ti  i i (Ti , i )) / 9550  dt
 (4
= i =1

m(v02 − v12 ) / 2
where Ti and ωi are the braking torque and speed output by the motors, respectively.
ηi (i = 1, 2, 3, 4) is the  of motors when outputting torque, Ti , and speed, ωi . m is
T efficiency
where i and i are the braking torque and speed output by the motors, respective
the vehicle mass, and η is the braking energy recovery efficiency.
i (i = 1,2,3,4) is the efficiency of motors when outputting torque, Ti , and speed, i .
Table 4. Comparison of fourmass,
is the vehicle
 is the strategies
coordinated
and control on high-adhesion-coefficient
braking energy recovery efficiency.road.
Control Energy Recovery Braking Adhesion Factor Final SOC of the Braking
Table 4. Comparison of four coordinated control strategies on high-adhesion-coefficient road.
Strategy Efficiency (%) Distance (m) Utilization Rate (%) Battery (%) Stability (%)
eMPC-CCSControl Energy Recovery
30.38 75.94 Adhesion Factor80.37 Final SOC of the
100 Braking Stabil
Braking Distance(m)98
MPC-CCSStrategy Efficiency (%)
22.36 76.88 96.6Utilization Rate (%)
80.27 Battery (%) 100 (%)
LQR-CCS
eMPC-CCS 20.99 30.38 77.79 75.94 95 98 80.25 80.37 100 100
PID-CCS 18.63 78.79 93 80.22 100
MPC-CCS 22.36 76.88 96.6 80.27 100
LQR-CCS 20.99 77.79 95 80.25 100
PID-CCS 18.63 78.79 93 80.22 100

Table 5. Comparison of four coordinated control strategies on low-adhesion coefficient road.

Control Energy Recovery Adhesion Factor Final SOC of the Braking Stabili
Braking Distance(m)
Strategy Efficiency (%) Utilization Rate (%) Battery (%) (%)
eMPC-CCS 37.64 45.83 96 80.14 100
Sensors 2024, 24, 3076 19 of 27

Table 5. Comparison of four coordinated control strategies on low-adhesion coefficient road.

Control Energy Recovery Braking Adhesion Factor Final SOC of the Braking
Strategy Efficiency (%) Distance (m) Utilization Rate (%) Battery (%) Stability (%)
eMPC-CCS 37.64 45.83 96 80.14 100
MPC-CCS 33.59 46.29 94.8 80.11 100
LQR-CCS 20.02 48.98 94.1 80.06 100
PID-CCS 25.21 47.32 94.5 80.08 100

According to the numerical results in Figure 11a and Table 4, the eMPC-CCS con-
tributes to an increase of 8.02% in the energy recovery efficiency compared with the
MPC-CCS, an increase of 9.39% compared with the LQR-CCS, and an increase of 11.75%
compared with the MPC-CCS on a high-adhesion-coefficient road. Similarly, numerical
results in Figure 11b and Table 5 reveal that the eMPC-CCS contributes to an increase of
4.05% in the energy recovery efficiency compared with the MPC-CCS, an increase of 17.62%
compared with the LQR-CCS, and an increase of 12.43% compared with the MPC-CCS on
a low-adhesion-coefficient road. From the perspective of energy recovery efficiency, the
eMPC-CCS can more effectively invoke the optimized control thresholds in real time by
referring to the wheel braking torque corresponding to the optimal road slip ratio, so it has
better adaptability to different braking modes. In terms of the adhesion factor utilization
rate in Tables 4 and 5, the eMPC-CCS can improve by 1.4% and 1.2% compared with the
MPC-CCS, which means that the vehicle has more braking force on the road, can transfer
power more efficiently, and provides better handling performance and stability.

4.2. Robustness Verification for eMPC-CCS

As shown in Figures 12 and 13, braking simulations are performed on the joint
road with varying adhesion coefficients, and a bisectional road with different adhesion
coefficients on the left and right sides, respectively.
The vehicle braking simulation results of four coordinated control strategies on the joint
road are shown in Figure 12. At the initial braking stage, emergency braking on a snowy road
with an adhesion coefficient of 0.3 is simulated; the initial braking speed is 60 km/h and the
initial SOC is 0.8. In Figure 12d–i, With the increase in braking torque applied to the wheels,
the front and rear wheels reach the condition of triggering the ABS one after another. After
the ABS is triggered, the proposed eMPC-CSS calculates the demanded braking torque for the
wheels to maintain the optimal slip rate, and then controls the braking system to apply the
corresponding braking torque on the wheels to quickly stabilize the slip rate near the optimal
slip rate. After the vehicle has traveled 15 m, i.e., after 1 s, the road suddenly becomes a dry
asphalt road with a value of 0.8. The proposed eMPC-CSS can still quickly make the wheel
slip ratio track the best slip ratio on the current road. In Figure 12b,c, it is clear that the slip rate
under the eMPC-CSS fluctuates more gently than that under the other three control strategies,
and can quickly keep up with the desired slip rate and stay close to it so that the wheels do
not lock up during ABS braking. The smaller fluctuations in the slip rate allow the braking
system to control the wheels more accurately and can improve vehicle stability and reduce
unnecessary vehicle sway and drift due to braking. Thus, the eMPC-CSS has better stability
and robustness on the joint road.
 the MPC-CCS, which means that the vehicle has more braking force on the road, can trans-
fer power more efficiently, and provides better handling performance and stability.

4.2. Robustness Verification for eMPC-CCS

As shown in Figures 12 and 13, braking simulations are performed on the joint road
Sensors 2024, 24, 3076 20 of 27
with varying adhesion coefficients, and a bisectional road with different adhesion coeffi-
cients on the left and right sides, respectively.

Sensors 2024, 24, 3076 22 of 28

and drift due to braking. Thus, the eMPC-CSS has better stability and robustness on the
Figure
joint 12. Simulation results of four coordinated control strategies for the joint road.
road.

The vehicle braking simulation results of four coordinated control strategies on the
joint road are shown in Figure 12. At the initial braking stage, emergency braking on a
snowy road with an adhesion coefficient of 0.3 is simulated; the initial braking speed is 60
km/h and the initial SOC is 0.8. In Figure 12d–i, With the increase in braking torque ap-
plied to the wheels, the front and rear wheels reach the condition of triggering the ABS
one after another. After the ABS is triggered, the proposed eMPC-CSS calculates the de-
manded braking torque for the wheels to maintain the optimal slip rate, and then controls
the braking system to apply the corresponding braking torque on the wheels to quickly
stabilize the slip rate near the optimal slip rate. After the vehicle has traveled 15 m, i.e.,
after 1 s, the road suddenly becomes a dry asphalt road with a value of 0.8. The proposed
eMPC-CSS can still quickly make the wheel slip ratio track the best slip ratio on the current
road. In Figure 12b,c, it is clear that the slip rate under the eMPC-CSS fluctuates more
gently than that under the other three control strategies, and can quickly keep up with the
desired slip rate and stay close to it so that the wheels do not lock up during ABS braking.
The smaller fluctuations in the slip rate allow the braking system to control the wheels
more accurately and can improve vehicle stability and reduce unnecessary vehicle sway

Figure 13.
Figure Simulation results
13. Simulation results of
of four
four coordinated
coordinated control
control strategies for the bisectional road.

As shown in Figure 13, the vehicle performs emergency braking on the bisectional
road with an adhesion coefficient on the left side of 0.3 and an adhesion coefficient on the
right side of 0.8. The initial braking speed is 60 km/h and the initial SOC is 0.8. At the
beginning of braking, as the braking torque increases, the front-left wheel on the low-
adhesion-coefficient road triggers the ABS earlier than the front-right wheel on the high-
Sensors 2024, 24, 3076 21 of 27

As shown in Figure 13, the vehicle performs emergency braking on the bisectional
road with an adhesion coefficient on the left side of 0.3 and an adhesion coefficient on
the right side of 0.8. The initial braking speed is 60 km/h and the initial SOC is 0.8. At
the beginning of braking, as the braking torque increases, the front-left wheel on the
low-adhesion-coefficient road triggers the ABS earlier than the front-right wheel on the
high-adhesion-coefficient road. At this time, there is a large gap between the slip rate of
the front-left wheel and the front-right wheel. The eMPC-CSS controls the slip rate of
the front-left wheel to be near the optimal road slip rate while continuing to increase the
braking torque on the front-right wheel. Subsequently, the rear-left wheel, which is on the
low-adhesion-coefficient road, reaches the ABS-triggering condition, and the eMPC-CSS
controls the slip rate of the rear-left wheel near the optimal road slip rate while continuing
to increase the braking torque on the rear-right wheel. Finally, the front-right wheel, which
is on a high-adhesion-coefficient road, also reaches the ABS-triggering condition. Aiming
at providing good braking performance, the eMPC-CSS controls the wheel slip rate near
the optimal road slip rate, which utilizes the road adhesion conditions to the fullest. When
the vehicle speed is more than 10 km/h, the left and right wheels are not locked to ensure
stability and safety during braking.
As shown in Figure 14, compared with the braking performance indicators among
Sensors 2024, 24, 3076 the four coordinated control strategies, the eMPC-CCS has the maximum energy recovery 23 of 28
efficiency and braking deceleration on the joint road and the bisectional road. In addition,
four coordinated control strategies can stabilize the vehicle during braking, but the eMPC-
CCS has the shortest
eMPC-CCS braking
but better distance.
than the LQR-CCSThe MPC-CCS
and PID-CCS; performs a bit worse
LQR-CCS than
has the the perfor-
worst eMPC-
CCS
mance. The eMPC-CCS can reasonably distribute the braking force of each wheel, The
but better than the LQR-CCS and PID-CCS; LQR-CCS has the worst performance. and
eMPC-CCS can reasonably
make corrections to ensuredistribute the braking
that the motor force
is in the of each
field wheel,
of high and make
efficiency corrections
under different
to ensureconditions.
driving that the motor is in the field of high efficiency under different driving conditions.

Figure 14. Comparison

Figure 14. Comparisonofoffour
fourcoordinated
coordinatedcontrol strategies
control on on
strategies thethe
joint road
joint andand
road the bisectional road.
the bisectional
road.
According to the numerical results in Figure 14a and Table 6, the eMPC-CCS con-
tributes to an increase
According of 5.63%results
to the numerical in the in
energy
Figurerecovery efficiency
14a and Table 6, thecompared
eMPC-CCSwith the
contrib-
MPC-CCS, an increase of 12.86% compared with the LQR-CCS, and an increase
utes to an increase of 5.63% in the energy recovery efficiency compared with the MPC- of 8.33%
compared with theofMPC-CCS
CCS, an increase on the joint
12.86% compared with road. Similarly, numerical
the LQR-CCS, resultsof
and an increase in8.33%
Figurecom-
14b
and Table 7 reveal that the eMPC-CCS contributes to an increase of 3.51% in the energy
pared with the MPC-CCS on the joint road. Similarly, numerical results in Figure 14b and
recovery efficiency compared with the MPC-CCS, an increase of 9.82% compared with the
Table 7 reveal that the eMPC-CCS contributes to an increase of 3.51% in the energy recov-
LQR-CCS, and an increase of 5.38% compared with the MPC-CCS on the bisectional road.
ery efficiency compared with the MPC-CCS, an increase of 9.82% compared with the LQR-
From the perspective of energy recovery efficiency, the eMPC-CCS can more effectively
CCS, and an increase of 5.38% compared with the MPC-CCS on the bisectional road. From
invoke the optimized control thresholds in real time by referring to the wheel braking
the perspective of energy recovery efficiency, the eMPC-CCS can more effectively invoke
torque corresponding to the optimal road slip ratio, so it has better adaptability to different
the optimized control thresholds in real time by referring to the wheel braking torque
braking modes. In Tables 6 and 7, in terms of braking deceleration, the eMPC-CCS has the
corresponding to the optimal road slip ratio, so it has better adaptability to different brak-
best performance, which means that the vehicle can slow down to the target speed or stop
ing modes. In Tables 6 and 7, in terms of braking deceleration, the eMPC-CCS has the best
faster, and allow the driver to control braking effort more accurately.
performance, which means that the vehicle can slow down to the target speed or stop
faster, and allow the driver to control braking effort more accurately.

Table 6. Comparison of four coordinated control strategies on the joint road.

Energy Recovery Effi- Braking Dis- Braking Deceleration Final SOC of the Braking Stabil-
Sensors 2024, 24, 3076 22 of 27

Table 6. Comparison of four coordinated control strategies on the joint road.

Energy Recovery Braking Braking Final SOC of the Braking

Control Strategy
Efficiency (%) Distance (m) Deceleration (m/s2 ) Battery (%) Stability (%)
eMPC-CCS 23.68 24.08 5.89 80.12 100
MPC-CCS 18.05 25.68 5.5 80.08 100
LQR-CCS 10.82 26.7 5.2 80.05 100
PID-CCS 15.35 26.2 5.34 80.07 100

Table 7. Comparison of four coordinated control strategies on the bisectional road.

Energy Recovery Braking Braking Final SOC of the Braking

Control Strategy
Efficiency (%) Distance (m) Deceleration (m/s2 ) Battery (%) Stability (%)
eMPC-CCS 26.07 27.18 5.11 80.13 100
MPC-CCS 22.56 28.9 4.81 80.10 100
LQR-CCS
Sensors 2024, 24, 3076 16.25 29.31 4.74 80.06 100 24 of 28
PID-CCS 20.69 29.16 4.76 80.09 100

Through the
Through the simulation
simulation analysis,
analysis, the proposed eMPC-CCS
eMPC-CCS can control
control vehicles
vehicles to
to
performanti-lock
perform anti-lock braking
braking with
with excellent
excellent stability and robustness when the road
road adhesion
adhesion
coefficient
coefficientchanges
changes suddenly
suddenly and
and the
the road adhesion coefficients on the left and right
right sides
sides
are
aredifferent.
different.

4.3.
4.3. Hardware-in-the-Loop
Hardware-in-the-Loop Test Test
To
Tovalidate
validate the
the performance
performance of of the
the eMPC-CC
eMPC-CC and and its
its functionality
functionality in in lowering
loweringcom-com-
putation
putation time
time and improving real-time
and improving real-timeperformance,
performance,a ahardware-in-the-loop
hardware-in-the-loop (HIL)
(HIL) testtest
is
isconducted.
conducted. Hardware test planning is depicted in Figure 15, primarily
Hardware test planning is depicted in Figure 15, primarily composed of host composed of
host
PC1 PC1 (the controller),
(the controller), hostnamed
host PC2 PC2 named Speedgoat,
Speedgoat, and theand themachine
target target machine interface.
interface. In host
In host PC1, the eMPC-CCS and vehicle model, including
PC1, the eMPC-CCS and vehicle model, including component sub-models, are con-component sub-models, are
constructed in MATLAB/Simulink. The Simulink model is then
structed in MATLAB/Simulink. The Simulink model is then compiled into C code compo- compiled into C code
components. Subsequently,
nents. Subsequently, the compiled
the compiled Simulink Simulink model
model from from
host host
PC1 PC1 is downloaded
is downloaded to PC2
to
andPC2 and displayed
displayed throughthrough themachine
the target target machine interface.
interface. The communication
The communication between between
the con-
the controller
troller and host and
PCshost PCs is attained
is attained via CANvia busCAN bus communication.
communication. To better To better the
illustrate illustrate
brak-
the braking energy recovery performance and real-time performance
ing energy recovery performance and real-time performance of the eMPC-CCS, the test of the eMPC-CCS,
the test scenario
scenario involvesinvolves
emergency emergency
braking,braking,
startingstarting
at a speedat aofspeed of 100
100 km/h km/h
and and
using using a
a braking
braking intensity
intensity of 0.8. of 0.8.

Figure15.
Figure 15. HIL
HIL test
test platform.
platform.

Figure 16 represents the slip ratio changes of different CCSs in the test scenario. Table
8 lists the numerical results of the energy recovery efficiency of different CCSs in the test
scenario, and the step time costs are shown in Table 9.
Sensors 2024, 24, 3076 23 of 27

Figure 15. HIL test platform.

Figure16
Figure 16represents
represents the
the slip ratio changes of different
different CCSs
CCSsininthe
thetest
testscenario.
scenario.Table
Table8
8lists
liststhe
thenumerical
numerical results
results of
of the
the energy
energy recovery efficiency of
recovery efficiency of different
different CCSs
CCSsininthe
thetest
test
scenario, and the step time costs are shown in Table
scenario, and the step time costs are shown in Table 9. 9.

Figure16.
Figure Slipratio
16.Slip ratiocurves
curvesby
bydifferent
differentCCSs
CCSs.
.

Brakingenergy
Table8.8.Braking
Table energyrecovery
recoveryefficiency
efficiencyof
ofdifferent
differentCCSs.
CCSs.

Control Energy Recovery Braking Final SOC of Braking

Strategy Efficiency (%) Distance (m) the Battery (%) Stability (%)
eMPC-CCS 25.58 47.18 80.24 100
MPC-CCS 22.51 48.9 80.20 100
LQR-CCS 20.61 49.31 80.18 100
PID-CCS 16.44 52.16 80.15 100

Table 9. Real-time applications of different CCSs.

Minimum Step Time Maximum Step Time Average Step Time

Control Strategy
Cost (s) Cost (s) Cost (s)
eMPC-CCS 7.26 × 10−6 2.54 × 10−5 7.76 × 10−6
MPC-CCS 7.5 × 10−6 2.73 × 10−5 8.27 × 10−6
LQR-CCS 7.25 × 10−6 3.02 × 10−6 7.78 × 10−6
PID-CCS 7.33 × 10−6 2.71 × 10−5 7.91 × 10−6

Table 8 lists the maximum step time cost, minimum step time cost, and average time
cost in the HIL test for different control strategies. Compared with other CCSs, the eMPC-
CCS has the smallest step time cost and meets real-time applications in road driving with a
maximum computational frequency of more than 50 Hz [49].
In the control strategies, the MPC-CCS, LQR-CCS, and PID-CCS tend to activate the
ABS frequently, but the slip ratio curve of the eMPC-CCS is closer to the reference slip ratio.
To be specific, from Figure 16, it can be seen that the slip ratio gradually increases during
the initial 0.5 s of braking, then vehicle braking torque is adjusted in real-time according
to the reference slip ratio. After 3.3 s, the vehicle speed decreases to 10 km/h, at which
point the ABS disengages from the braking process, resulting in a sudden surge in the slip
ratio to 1. Compared with the slip ratio curve of the other traditional CCSs in the partially
enlarged figure, the slip ratio of the eMPC-CCS changes smoothly, so the eMPC-CCS shows
a strong adjustment ability. According to the results in Table 9, the eMPC-CCS can increase
braking energy recovery efficiency by nearly 3.07% to 9.14% compared with other CCSs.
Through the simulation analyses, the developed eMPC-CCS, formed by offline control
laws generation and online control law invocation, can better coordinate the control RBS
and ABS, which could follow the braking energy recovering trend of the reference slip
ratio. To be specific, in the HIL test, the real-time calculation ability of the eMPC-CCS can
Sensors 2024, 24, 3076 24 of 27

also meet the requirement of the communicating frequency for the CAN bus. In short, the
eMPC-CCS shows an advantage in coordinated control between RBS and ABS.

5. Discussion
In this study, a novel eMPC-based coordinated control strategy, namely the eMPC-
CCS, is proposed for the electro-hydraulic composite braking system. The aim is to ensure
the real-time performance and stability of the braking process and maximize the braking
energy recovery of the four-wheel-drive hub electric vehicle. Comparative studies are
conducted through simulations to verify the feasibility and validity of the eMPC-CCS.
From the analysis in Section 3, we can draw the following conclusions:
(1) The proposed CCS based on eMPC, named the eMPC-CCS, greatly improves the
online calculation speed of coordinated control strategy allocation through offline
processing and online implementation, and can provide more accurate and intuitive
control performance analysis.
(2) A three-dimensional eMPC law generation method based on multiple braking condi-
tions generates a 3D eMPC explicit solution by invoking multiple sets of micro trip
segments to generate state feedback control laws, achieving the adaptability of the
control strategy.
(3) The eMPC-CCS includes an improved eMPC controller with a basic eMPC controller
and state error compensator, which improves real-time capability, adaptability, and
robustness under various braking conditions. Compared with the other CCSs, namely
the PID-CCS, LQR-CCS, and MPC-CCS, the braking energy recovery efficiency of the
eMPC-CCS is increased by at least nearly 4%.

6. Conclusions
In this paper, a novel eMPC-based coordinated control strategy named the eMPC-CSS
is proposed for electro-hydraulic composite braking systems. By combining offline control
law generation with online control law invocation, this strategy augments real-time capa-
bility and robustness between the RBS and ABS. Offline control law generation, including
real-time-oriented state feedback control laws under micro braking segments, supporting
the eMPC-CCS to have a properly coordinated control tendency. The online implemen-
tation, containing 3D eMPC control law generation and state error compensation, can
facilitate control law application in practice while also allowing for the rational distribution
of motor and hydraulic braking torque. Compared to other CCSs such as the MPC-CCS,
LQR-CCS, and PID-CCS, the proposed eMPC-CSS demonstrates a significant improvement
in braking energy recovery efficiency, with gains ranging from approximately 4% to 17%.
The simulation-based test and HIL validation verify that the proposed eMPC-CCS effec-
tively ensures the real-time capability, adaptability, and robustness of the CCS, showcasing
its anticipated superior performance.
However, it is crucial to acknowledge the significant discoveries and limitations
presented in these studies. Firstly, one key limitation identified is the considerable braking
torque fluctuations due to the differing dynamic response characteristics of the regenerative
and hydraulic braking systems, particularly during mode-switching sequences. This
issue warrants further investigation into mode transition-smoothing techniques. Secondly,
the current study solely considers the influence of longitudinal force on vehicle braking,
disregarding the impact of lateral force throughout the time course. Consequently, it is
imperative for future studies to comprehensively examine both the implications of braking
torque fluctuations and the influence of lateral force during vehicle braking for a more
complete understanding and enhancement of system performance.
Sensors 2024, 24, 3076 25 of 27

Author Contributions: L.C.: Conceptualization, Methodology, and Writing—Original Draft Prepa-

ration. J.L.: Data Collection, Investigation, and Writing—Reviewing and Editing. Z.G., Z.J., S.L.,
W.D. and Y.W.: Formal Analysis, Software, and Visualization. C.G.: Supervision, Project Admin-
istration, and Funding Acquisition. All authors have read and agreed to the published version of
the manuscript.
Funding: This work was supported in part by the National Natural Science Foundation of China
(Grant No. 52272365).
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: No new data were created or analyzed in this study. Data sharing is
not applicable to this article.
Acknowledgments: We are thankful to State Key Laboratory of Automotive Simulation and Control,
Jilin University, for providing access to facilities and equipment essential for conducting experiments
and gathering data. We would like to express our gratitude to all participants and volunteers who
generously contributed their time and effort to this study.
Conflicts of Interest: The authors declare that they have no known competing financial interests or
personal relationships that could have appeared to influence the work reported in this paper.

References
1. Li, N.; Jiang, J.; Sun, F.; Ye, M.; Ning, X.; Chen, P. A cooperative control strategy for a hydraulic regenerative braking system
based on chassis domain control. Electronics 2022, 11, 4212. [CrossRef]
2. Deemyad, T.; Moeller, R.; Sebastian, A. Chassis design and analysis of an autonomous ground vehicle (AGV) using genetic
algorithm. In Proceedings of the 2020 IEEE Intermountain Engineering, Technology and Computing (IETC), Orem, UT, USA,
2–3 October 2020; pp. 1–6.
3. Yang, Z.; Wang, B.; Jiao, K. Life cycle assessment of fuel cell, electric and internal combustion engine vehicles under different fuel
scenarios and driving mileages in China. Energy 2020, 198, 117365. [CrossRef]
4. Tang, Q.; Yang, Y.; Luo, C.; Yang, Z.; Fu, C. A novel electro-hydraulic compound braking system coordinated control strategy for
a four-wheel-drive pure electric vehicle driven by dual motors. Energy 2022, 241, 122750. [CrossRef]
5. Yao, Y.; Zhao, Y.; Yamazaki, M. Integrated regenerative braking system and anti-lock braking system for hybrid electric vehicles &
battery electric vehicles. SAE Int. J. Adv. Curr. Pract. Mobil. 2020, 2, 1592–1601.
6. Jamadar, N.M.; Jadhav, H.T. A review on braking control and optimization techniques for electric vehicle. Proc. Inst. Mech. Eng.
Part D J. Autom. Eng. 2021, 235, 2371–2382. [CrossRef]
7. Zhu, Y.; Wu, H.; Zhang, J. Regenerative braking control strategy for electric vehicles based on optimization of switched reluctance
generator drive system. IEEE Access 2020, 8, 76671–76682. [CrossRef]
8. Cheon, S.Y.; Choi, S.W.; Yang, S.Y. Development of Real-time Simulator for Vehicle Electric Brake System. J. Drive Control 2019,
16, 22–28.
9. Pei, X.; Pan, H.; Chen, Z.; Guo, X.; Yang, B. Coordinated control strategy of electro-hydraulic braking for energy regeneration.
Control Eng. Pract. 2020, 96, 104324. [CrossRef]
10. Sardarmehni, T.; Heydari, A. Optimal switching in anti-lock brake systems of ground vehicles based on approximate dynamic
programming. In Proceedings of the Dynamic Systems and Control Conference, Columbus, OH, USA, 28–30 October 2015;
Volume 57267, p. V003T50A010.
11. Kim, D.; Eo, J.S.; Kim, K.K.K. Parameterized energy-optimal regenerative braking strategy for connected and autonomous
electrified vehicles: A real-time dynamic programming approach. IEEE Access 2021, 9, 103167–103183. [CrossRef]
12. Zhang, Y.; Wang, W.; Yang, C.; Han, L.; Zhang, Z.; Liu, J. An effective regenerative braking strategy based on the combination
algorithm of particle swarm optimization and ant colony optimization for electrical vehicle. In Proceedings of the 2019 IEEE 28th
International Symposium on Industrial Electronics (ISIE), Vancouver, BC, Canada, 12–14 June 2019; pp. 1905–1910.
13. Gad, A.G. Particle swarm optimization algorithm and its applications: A systematic review. Arch. Comput. Methods Eng. 2022,
29, 2531–2561. [CrossRef]
14. Aparow, V.R.; Ahmad, F.; Hudha, K.; Jamaluddin, H. Modelling and PID control of antilock braking system with wheel slip
reduction to improve braking performance. Int. J. Veh. Saf. 2013, 6, 265–296. [CrossRef]
15. Zhang, L.; Pang, Z.; Wang, S.; Zhang, S.; Yuan, X. Electromechanical composite brake control for two in-wheel motors drive
electric vehicle with single motor failure. Proc. Inst. Mech. Eng. Part D J. Autom. Eng. 2020, 234, 1057–1074. [CrossRef]
16. Karaşahin, A.T.; Karali, M. Optimization with Genetic Algorithm of Linear Quadratic Regulator Controller for Active Trailer
Braking System. Eur. J. Res. Dev. 2022, 2, 1–12. [CrossRef]
17. Jin, H.; Xu, H.; Wang, S. The electromechanical brake control strategy based on linear quadratic regulator. In Proceedings of the
Tenth International Symposium on Precision Mechanical Measurements, Qingdao, China, 15–17 October 2021; Volume 12059,
pp. 11–16.
Sensors 2024, 24, 3076 26 of 27

18. Dou, J.; Cui, G.; Li, S.; Zheng, S.; Zhu, X.; Yu, Z. Research on the Composite Braking Control of Electric Vehicle. In Proceedings of
the 2017 2nd International Conference on Automation, Mechanical and Electrical Engineering (AMEE 2017), Shenzhen, China,
17–18 September 2017; pp. 4–8.
19. Yu, D.; Wang, W.; Zhang, H.; Xu, D. Research on anti-lock braking control strategy of distributed-driven electric vehicle. IEEE
Access 2020, 8, 162467–162478. [CrossRef]
20. Zhang, L.; Wang, Q.; Chen, J.; Wang, Z.P.; Li, S.H. Brake-by-wire system for passenger cars: A review of structure, control, key
technologies, and application in X-by-wire chassis. eTransportation 2023, 18, 100292. [CrossRef]
21. Mönnigmann, M.; Pannocchia, G. Reducing the computational effort of MPC with closed-loop optimal sequences of affine laws.
IFAC-PapersOnLine 2020, 53, 11344–11349. [CrossRef]
22. Yuan, L.; Zhao, H.; Chen, H.; Ren, B. Nonlinear MPC-based slip control for electric vehicles with vehicle safety constraints.
Mechatronics 2016, 38, 1–15. [CrossRef]
23. Chu, H.; Meng, D.; Huang, S.; Tian, M.; Zhang, J.; Gao, B.; Chen, H. Autonomous high-speed overtaking of intelligent chassis
using fast iterative model predictive control. IEEE Trans. Transp. Electrif. 2024, 10, 1244–1256. [CrossRef]
24. Meng, D.; Chu, H.; Tian, M.; Gao, B. Real-Time High-Precision Nonlinear Tracking Control of Autonomous Vehicles Using Fast
Iterative Model Predictive Control. IEEE Trans. Intell. Veh. 2024, 9, 3644–3657. [CrossRef]
25. Bemporad, A.; Morari, M.; Dua, V.; Pistikopoulos, E.N. The explicit linear quadratic regulator for constrained systems. Automatica
2002, 38, 3–20. [CrossRef]
26. Pistikopoulos, E. Perspectives in multiparametric programming and explicit model predictive control. AIChE J. 2009, 55, 1918–1925.
[CrossRef]
27. Bemporad, A.; Morari, M.; Dua, V.; Pistikopoulos, E.N. The explicit solution of model predictive control via multiparametric
quadratic programming. In Proceedings of the 2000 IEEE American Control Conference, ACC (IEEE Cat. No.00CH36334),
Chicago, IL, USA, 28–30 June 2000; pp. 872–876.
28. Gupta, A.; Falcone, P. Low-Complexity Explicit MPC Controller for Vehicle Lateral Motion Control. In Proceedings of the
2018 21st IEEE International Conference on Intelligent Transportation Systems (ITSC), Maui, HI, USA, 4–7 November 2018;
pp. 2839–2844.
29. Aksjonov, A.; Ricciardi, V.; Augsburg, K.; Vodovozov, V.; Petlenkov, E. Hardware-in-the-loop test of an open-loop fuzzy control
method for decoupled electrohydraulic antilock braking system. IEEE Trans. Fuzzy Syst. 2020, 29, 965–975. [CrossRef]
30. Modares, H.; Lewis, F.L.; Naghibi-Sistani, M.B. Adaptive optimal control of unknown constrained-input systems using policy
iteration and neural networks. IEEE Trans. Neural Netw. Learn. Syst. 2013, 24, 1513–1525. [CrossRef]
31. Li, Y.; Liu, Y.; Tong, S. Observer-based neuro-adaptive optimized control of strict-feedback nonlinear systems with state constraints.
IEEE Trans. Neural Netw. Learn. Syst. 2021, 33, 3131–3145. [CrossRef]
32. Weber, D.; Schenke, M.; Wallscheid, O. Steady-State Error Compensation for Reinforcement Learning-Based Control of Power
Electronic Systems. IEEE Access 2023, 11, 76524–76536. [CrossRef]
33. Han, Y.; Gong, C.; Yan, L.; Wen, H.; Wang, Y.; Shen, K. Multiobjective finite control set model predictive control using novel delay
compensation technique for PMSM. IEEE Trans. Power Electron. 2020, 35, 11193–11204. [CrossRef]
34. Kazemi, N.; Abdolrazzaghi, M.; Musilek, P. Comparative analysis of machine learning techniques for temperature compensation
in microwave sensors. IEEE Trans. Microw. Theory Tech. 2021, 69, 4223–4236. [CrossRef]
35. Barkana, I. Simple adaptive control—A stable direct model reference adaptive control methodology–brief survey. Int. J. Adapt.
Control Signal Process. 2014, 28, 567–603. [CrossRef]
36. Yao, J.; Jiao, Z.; Ma, D. A practical nonlinear adaptive control of hydraulic servomechanisms with periodic-like disturbances.
IEEE/AsmE Trans. Mechatron. 2015, 20, 2752–2760. [CrossRef]
37. Zhao, X.; Shi, P.; Zheng, X. Fuzzy adaptive control design and discretization for a class of nonlinear uncertain systems. IEEE
Trans. Cybern. 2015, 46, 1476–1483. [CrossRef]
38. Yan, L.; Dou, M.; Hua, Z.; Zhang, H.; Yang, J. Robustness improvement of FCS-MPTC for induction machine drives using
disturbance feedforward compensation technique. IEEE Trans. Power Electron. 2018, 34, 2874–2886. [CrossRef]
39. Zeng, T.; Xiao, L.; Chen, J.; Li, Y.; Yang, Y.; Huang, S.; Deng, C.; Zhang, C. Feedforward-based decoupling control of air supply for
vehicular fuel cell system: Methodology and experimental validation. Appl. Energy 2023, 335, 120756. [CrossRef]
40. Yu, J.; Zhao, L.; Yu, H.; Lin, C. Barrier Lyapunov functions-based command filtered output feedback control for full-state
constrained nonlinear systems. Automatica 2019, 105, 71–79. [CrossRef]
41. Deng, W.; Yao, J.; Wang, Y.; Yang, X.; Chen, J. Output feedback backstepping control of hydraulic actuators with valve dynamics
compensation. Mech. Syst. Signal Process. 2021, 158, 107769. [CrossRef]
42. Li, S.; Yu, B.; Feng, X. Research on braking energy recovery strategy of electric vehicle based on ECE regulation and I curve. Sci.
Prog. 2020, 103, 0036850419877762. [CrossRef] [PubMed]
43. Domínguez, L.F.; Pistikopoulos, E.N. Recent advances in explicit multiparametric nonlinear model predictive control. Ind. Eng.
Chem. Res. 2011, 50, 609–619. [CrossRef]
44. Grancharova, J.A.; Johansen, T.A. Explicit Nonlinear Model Predictive Control: Theory and Applications; Springer: Berlin, Germany,
2012; Volume 429.
45. Pacejka, H.B. Tire and Vehicle Dynamics; Butterworth Heinemann: London, UK, 2012.
Sensors 2024, 24, 3076 27 of 27

46. Bardawil, C.; Daher, N.; Shammas, E. Applying the Similarity Method on Pacejka’s Magic Formula to Estimate the Maximum
Longitudinal Tire-Road Friction Coefficient. In Proceedings of the 2020 American Control Conference (ACC), Denver, CO, USA,
1–3 July 2020; pp. 218–223.
47. Chang, H.; Pistikopoulos, E.N.; Astolfi, A. Robust multi-parametric model predictive control for discrete-time LPV systems. In
Proceedings of the American Control Conference, Washington, DC, USA, 17–19 June 2013.
48. Ventre, A.G. Determinants and Systems of Linear Equations. In Calculus and Linear Algebra: Fundamentals and Applications;
Springer: Cham, Switzerland, 2023; pp. 209–241.
49. Olumuyiwa, O.; Chen, Y. Virtual CANBUS and Ethernet Switching in Future Smart Cars Using Hybrid Architecture. Electronics
2022, 11, 3428. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.