Professional Documents
Culture Documents
Research Article
constraint
Yuanzhi Zhang1,2, Mingchun Liu2 , Caizhi Zhang1
1School of Automotive Engineering, The State Key Laboratory of Mechanical Transmissions, Chongqing Automotive Collaborative Innovation
Center, Chongqing University, Chongqing 400044, People's Republic of China
2School of Mechatronics Engineering, Nanchang University, Nanchang 330031, People's Republic of China
E-mail: liumingchun@ncu.edu.cn
Abstract: Currently, majorities of the robust H∞ control methods are designed for active suspensions, and seldom take the
active control of the in-wheel-motor (IWM) into consideration for IWM driven electric vehicles (EVs). In this study, a robust fault-
tolerant H∞ output feedback control strategy with finite-frequency constraint is proposed to synchronously control the active
suspension and dynamic vibration absorber (DVA) for IWM driven EVs. Firstly, a DVA-based electric wheel model is developed,
in which the IWM is designed as DVA. Furthermore, the spring-damper parameters of the DVA are matched by using particle
swarm optimisation (PSO). Then, the robust fault-tolerant H∞ output feedback control strategy is developed based on linear
matrix inequality, in which the finite-frequency constraint is designed in the resonance frequency range of sprung mass. Finally,
simulation results validate that the PSO can effectively optimise the spring-damper parameters of the DVA. The robust fault-
tolerant H∞ output feedback control with finite-frequency constraint can effectively improve the ride comfort and suppress the
vertical vibration caused by IWM compared with entire frequency constraint. Meanwhile, the fault-tolerant effectiveness of the
proposed method is demonstrated under the actuator faults concerning the actuator force noises and losses.
RMS as RMS Fe
Jpara Ke1, Ce1, Ke2, Ce2 = w1 + w2
RMS asCon − EW RMS FeCon − EW
RMS df RMS Ft
+w3 + w4
RMS dfCon − EW RMS FtCon − EW
(1)
Fig. 2 Active DVA-EW model
where asCon − EW, FeCon − EW, dfCon − EW and FtCon − EW are the sprung
In this paper, in order to take the robust H∞ control of the IWM
mass acceleration, dynamic force applied on IWM, suspension
into consideration in the active control design of active suspensions deflection and tire dynamic load of the Con-EW, and as, Fe, df and
for IWM driven EVs, a robust fault-tolerant H∞ output feedback
Ft are those of the DVA-EW, respectively. RMS(⋯) is the root
control strategy is proposed, in which the active suspension and
DVA are synchronously controlled with the finite-frequency mean square (RMS) value. Meanwhile, w1, w2, w3 and w4 are the
constraint. The proposed framework is described as follows. First, weighting factors for the four evaluation indexes, and they are set
a DVA-based electric wheel (DVA-EW) model is employed, and a as 0.3, 0.4, 0.15 and 0.15, respectively.
particle swarm optimisation (PSO) method is designed to match the In addition, the suspension deflection of the DVA-EW df and
spring-damper parameters of the DVA. Then, the robust fault- the vertical displacement between the unsprung mass and IWM
tolerant H∞ output feedback control strategy with finite-frequency (xe − xu) should be limited to ensure safe working spaces of the
constraint is designed to improve the ride comfort and suppress the active suspension and DVA. Meanwhile, tire dynamic load of the
vertical vibration caused by the IWM in the resonance frequency DVA-EW Ft should be maintained to ensure proper road holding.
range of sprung mass (1–2 Hz). Meanwhile, the suspension stroke The relative damping coefficient of the sprung mass ξs and
and road holding are maintained as well. The proposed method is unsprung mass ξu, and natural frequency of the sprung mass ωs and
designed to guarantee the system performance in the actuator faults unsprung mass ωu should be limited within reasonable ranges.
concerning actuator force noises and losses. In detail, the DVA-EW Therefore, the following constrains for spring-damper parameters
and PSO methods are designed in Section 2. The robust fault- of the DVA are defined as [26]
tolerant H∞ output feedback control strategy with finite-frequency
constraint is designed in Section 3. Section 4 presents the RMS df ≤ 5 mm
effectiveness of the proposed strategy by simulation results. The
RMS xe − xu ≤ 4 mm
conclusions are briefly summarised in Section 5.
Notation: P s is the algebraic operation of P + PT. P⊥, P−1 and max xe − xu ≤ 12 mm
T
P are the orthogonal complement, inverse and transpose of matrix RMS Ft ≤ ms + mu + me g/3
(2)
P, respectively. The symbol ∗ is the symmetric part. diag(⋯) is the 0.2 ≤ ξs ≤ 0.5
diagonal matrix.
1 ≤ ωs ≤ 1.5
0.2 ≤ ξu ≤ 0.5
2 System modelling
10 ≤ ωu ≤ 13
2.1 Model of EV, Con-EW and DVA-EW
For an EV, a motor is installed into vehicle chassis, such that the The PSO method is employed to minimise the optimisation
motor is a part of sprung mass, as shown in Fig. 1a. For a objective function (1) with constrains (2), such that the optimal
conventional electric wheel (Con-EW), rotor and stator of the IWM solution of spring-damper parameters of the DVA can be found.
are directly installed to the hub and shaft of vehicle wheels, The optimisation processes of PSO method can be found in [26].
respectively, as shown in Fig. 1b. In order to suppress the vertical
vibration caused by the IWM, the IWM is designed as a DVA, and 2.3 Model of active DVA-EW
attached to unsprung mass and sprung mass through spring-damper
systems, as shown in Fig. 1c. In order to further absorb the vertical vibration of both the IWM
In Figs. 1a–c, Ke1 and Ke2 are the stiffness of DVA springs; Ce1 and active suspension, two active actuator forces of the DVA u1 and
and Ce2 are the damping of DVA dampers; ms, mu and me are the active suspension u2 are added to the active DVA-EW, respectively,
mass of the vehicle chassis, wheel assembly and IWM, as shown in Fig. 2.
respectively; Kt and Ks are the stiffness of the tire and active Focusing on improving the ride comfort and suppressing the
suspension, respectively; Cs is the damping of the active vertical vibration caused by the IWM, the following conditions
should be strictly satisfied:
suspension; xs, xu, xe and xr are the vertical displacements of the
Cs + Ce2 Ce2 Cs K s + K e2 K e2
− − 0
ms ms ms ms ms
Ce2 Ce1 + Ce2 Ce1 K e2 Ke1 + Ke2
− − 0
me me me me me
A= Cs Ce1 Cs + Ce1 Ks K e1 Kt
−
mu mu mu mu mu mu
1 0 −1 0 0 0
0 1 −1 0 0 0
0 0 −1 0 0 0
IET Intell. Transp. Syst. 3
© The Institution of Engineering and Technology 2020
where G jω is the transfer function of the linear system value M̄ fi, respectively. The fault-tolerant matrix includes three
Ā, B̄, C̄, D̄ . different actuator conditions, and are listed as follows:
(ii) There exists symmetric matrices Π, P, Q > 0, and
(i) M fi = M̄ fi = 0, i = 1, 2, mean that the corresponding actuators
T are completely failed.
Γ P, Q, C̄, D̄ C̄, D̄
<0 (7) (ii) 0 < M fi ≤ M̄ fi < 1, i = 1, 2, mean that the corresponding
∗ −I
actuators exist partial faults.
where (iii) M fi = M̄ fi = 1, i = 1, 2, mean that the corresponding actuators
have no faults.
T −Q P + jωcQ Ā
Ā B̄ B̄
Γ P, Q, C̄, D̄ = + In this paper, M fi and M̄ fi are designed by 0.2 and 1, respectively,
I 0 P − jωcQ −ω1ω2Q I 0 which means that the actuators possibly have 0–80% faults.
0 C̄Π12 Therefore, the fault matrix Mf can be obtained by a polytope M
T with four vertices
∗ D̄ Π12 + Π22
s
4 4
with M ≜ Mf Mf = ∑ α jMfj; ∑ α j = 1, α j ≥ 0 (14)
Π11 Π12 T T
j=1 j=1
Π= , D̄ Π12 = D̄ Π12 + ΠT12 D̄ and ωc = ω1 + ω2 /2.
∗ Π22 s
0.2 0 1 0
where the constant matrices Mf1 = , Mf2 = ,
0 0.2 0 0.2
Lemma 2: (Projection lemma): Given real matrices Ω, Λ1 and 0.2 0 1 0
Mf3 = and Mf4 = .
Λ2, there exists a matrix F such that Ω + Λ1FΛ2 + Λ1FΛ2 T < 0, if 0 1 0 1
the two conditions hold [28]: Substituting (12) into the system (3), the system (3) with robust
fault-tolerant H∞ output feedback control is obtained as
T T
Λ1⊥Ω Λ1⊥ < 0, Λ2⊥Ω Λ⊥2 <0 (8) ¯
ẋ = Ax + B̄w
¯
Lemma 3: (Reciprocal projection lemma): Given real matrices z1 = Cx + D̄w (15)
P, S and W, and a positive definite matrix Ψ, the inequality z2 = C2 x
P + S + ST < 0 is equivalent [28]
where Ā = A + B jK f C3, B̄ = G, C̄ = C1 + D1jK f C3, D̄ = 0,
T T
P+Ψ− W s S +W j
B =
j
BM f and D1 = j j
j = 1, ⋯, 4. In (15), the transfer
D1 M f ,
<0 (9)
∗ −Ψ function of the system (15) can be obtained as
−1
Lemma 4: (Schur complement). There exists a symmetric matrix G jω = C̄ sI − Ā B̄ + D̄ (16)
S, the following three inequalities are equivalent [29]:
The gain matrix Kf is calculated by using the robust fault-tolerant
S<0
H∞ output feedback control with finite-frequency constraint. It can
T −1
S11 < 0, S22 − S12 S11 S12 < 0 (10) guarantee that the system (15) is asymptotically stable, and the
T −1 transfer function (16) can satisfy the finite-frequency inequality (4)
S22 < 0, S11 − S S S < 0
12 22 12
under the constrain inequality (5).
Based on Lemmas 1, 2, 3 and 4, the robust fault-tolerant H∞
S11 S12
where S = . output feedback control with finite-frequency constraint is
S21 S22 designed as follows:
3.2 Control strategy design Theorem 1: The fault-tolerant control is designed firstly. If there
The desired actuator force u including two actuator forces of the exist Q = QT > 0 and Msf , such that the inequality (17) is satisfied
DVA u1 and active suspension u2 is considered as as [30]:
T
u = Kf y = Kf C3 x (11) AQ + QAT + B j Msf + B j Msf <0 (17)
−1 −I Λ7
Kf C3 = K̄ f F̄ (25) <0 (33)
∗ −P̄1
Proof 3: H1 = F−1, H2 = diag F−1, F−1, I, I and where (see equation below)
−1
H3 = diag I, F are defined to solve the non-convex problem of
Λ6 = NFNT NTC1T + RFRT RTC1T + RKRT D1jT + NFNT NTKsfT D1jT
inequalities (19) and (20). The three congruence transformations
+ RFRT RTKsfT D1jT
are executed to (19), (20) and (21), respectively, by H1T, H2T and H3T
on the left, and H1, H2 and H3 on the right. Then, inequality (19),
(20) and (21) is equivalent to inequality (22), (23) and (24), Λ7 = ρ C2 i NFN NT + ρ C2 i RFR RT, i = 1, 2.
respectively. □ By solving the numerically tractable LMI problems (31)–(33),
Once the fault-tolerant gain matrix Ksf and finite-frequency H∞ the gain matrix Kf is obtain as
output feedback gain matrix Kf C3 is obtained, the gain matrix Kf C3
is further determined by Theorem 4. Kf = KRFR−1 (34)
Theorem 4: If there exists a general matrix F̄, and symmetric
matrices P̄, P̄1 > 0, Q̄ > 0, which are satisfied as Proof 5: Firstly, the following variables are defined as:
^
Λ3 < 0 (26) K f = KR RT
F̄ = NFN NT + RFR RT (35)
R = C3† + NL1
−0.158 0.352
Ke = 104 × , K f = 104 ×
0.173 −0.216
−0.508 1.616 (36)
0.579 −0.454
Fig. 5 Actuator forces of active suspension and DVA under rand road
(a) Actuator forces of active suspension, (b) Actuator forces of DVA
Hπvx 2πvx L
sin t, 0≤t≤
L L vx
wt = (37)
L
0, 0≤t≤
vx
which indicates that the active control of the active suspension and
DVA can improve the ride comfort and road holding, and reduce
the vertical vibration caused by the IWM and suspension stroke.
Furthermore, the active DVA-EW with control II on the active
suspension and DVA significantly has better performance of the
four evaluation factors compared with the control I.
Meanwhile, the RMS values of as, Fe, df and Ft for the DVA-
EW, active DVA-EW with controls I and II under bumpy road are
listed in Table 4. as of active DVA-EW with controls I and II is
reduced by 5.36 and 46.43%, respectively, compared with DVA-
EW. Meanwhile, Fe of active DVA-EW with controls I and II is
reduced by 9.08 and 21.68%, respectively, compared with DVA-
EW. It is obvious that the active DVA-EW with control II can
achieve better performance of the ride comfort and vertical
vibration caused by the IWM compared with active DVA-EW with
control I. In addition, df and Ft are guaranteed as well.
Figs. 10–12 show the time responses of the DVA-EW, active
DVA-EW with controls I and II under bumpy road with actuator
force losses, which is 20, 50 and 80%, respectively. From these
diagrams, the active DVA-EW with controls I and II can achieve
better performance of the four evaluation factors compared with
Fig. 6 Frequency responses of DVA-EW, active DVA-EW with controls I DVA-EW. It demonstrates the active control effectiveness of the
and II active suspensive and DVA with actuator force losses. Moreover,
(a) Sprung mass acceleration, (b) Dynamic force applied on IWM, (c) Suspension with an increase in the actuator force losses, active DVA-EW with
deflection, (d) Tire dynamic load control II can achieve the better performance of the four evaluation
factors compared with control I.
The RMS values of as, Fe, df and Ft for the DVA-EW, active
DVA-EW with controls I and II under bumpy road with different
actuator force losses are shown in Table 5. as of the active DVA-
EW with control II can be reduced 36.61, 22.32 and 8.93%,
respectively, compared with that of the DVA-EW with different
actuator force losses. Meanwhile, Fe of the active DVA-EW with
control II can be reduced 24.96, 23.15 and 11.67%, respectively.
However, as of the active DVA-EW with control I can be only
reduced 4.46, 2.68 and 0.89%, respectively, compared with that of
the DVA-EW with different actuator force losses. Fe of the active
Fig. 7 Actuator force noises of active suspension and DVA DVA-EW with control I can be reduced 8.04, 5.67 and 2.48%,
respectively. In addition, df and Ft are guaranteed as well. The
shown in Fig. 9. As shown in Fig. 9, the four evaluation factors results show that the active suspension and DVA performance of
including as, Fe, df and Ft of the active DVA-EW with controls I the active DVA-EW with controls I and II are impaired when the
and II are greatly reduced compared with those of the DVA-EW, actuator force losses occur, but they have the better performance
8 IET Intell. Transp. Syst.
© The Institution of Engineering and Technology 2020
Fig. 10 Time responses of DVA-EW, active DVA-EW with controls I and II
under bumpy road with 20% actuator force loss
Fig. 9 Time responses of DVA-EW, active DVA-EW with controls I and II (a) Sprung mass acceleration, (b) Dynamic force applied on IWM, (c) Suspension
under bumpy road with actuator force noises deflection, (d) Tire dynamic load
(a) Sprung mass acceleration, (b) Dynamic force applied on IWM, (c) Suspension
deflection, (d) Tire dynamic load performance of the active suspension and IWM in spite of actuator
force losses.
Table 4 RMS comparison of DVA-EW, active DVA-EW with
controls I and II under bumpy road with actuator force noises 5 Conclusion
Model as, m ⋅ s−2/ Fe, N/ df , mm Ft, N
marked In this paper, the robust fault-tolerant H∞ output feedback control
Decrement, % Decrement, %
with finite-frequency constraint is investigated to optimise the
DVA-EW 1.12/— 37.54/- 14.43 366.23 vertical performance of IWM driven EVs by synchronously
control I 1.06/5.36 34.13/9.08 14.80 345.35 controlling the active suspension and DVA. First, a DVA-EW is
control II 0.60/46.43 29.40/21.68 13.70 207.55 modelled, in which the IWM is designed as the DVA. Then, the
spring-damper parameters of the DVA are optimally matched by
using the PSO method. In addition, a robust fault-tolerant H∞
compared with DVA-EW. Meanwhile, the active DVA-EW with output feedback control with finite-frequency constraint is
control II can achieve better performance compared with control I, proposed to synchronously control the active suspension and DVA
which validates that the control II can achieve the better based on LMI method. The simulation results demonstrate that the
IET Intell. Transp. Syst. 9
© The Institution of Engineering and Technology 2020
Fig. 11 Time responses of DVA-EW, active DVA-EW with controls I and II Fig. 12 Time responses of DVA-EW, active DVA-EW with controls I and II
under bumpy road with 50% actuator force loss under bumpy road with 80% actuator force loss
(a) Sprung mass acceleration, (b) Dynamic force applied on IWM, (c) Suspension (a) Sprung mass acceleration, (b) Dynamic force applied on IWM, (c) Suspension
deflection, (d) Tire dynamic load deflection, (d) Tire dynamic load
Table 5 RMS comparison of DVA-EW, active DVA-EW with controls I and II under bumpy road with different actuator force
losses
Model marked as, m ⋅ s−2 Fe, N df , mm Ft, N
Decrement, % Decrement, %
DVA-EW 1.12/— 37.54/— 14.43 366.23
20% actuator force loss control I/control II 1.07/0.71 34.52/28.17 14.71/14.02 349.63/234.24
4.46/36.61 8.04/24.96
50% actuator force loss control I/control II 1.09/0.87 35.41/28.85 14.59/14.27 355.96/281.36
2.68/22.32 5.67/23.15
80% actuator force loss control I/control II 1.11/1.02 36.61/33.16 14.49/14.39 362.16/331.93
0.89/8.93 2.48/11.67