Professional Documents
Culture Documents
quasi-NLPV Observer FaultDetection
quasi-NLPV Observer FaultDetection
Abstract: This paper proposes an H∞ Non Linear Parameter Varying (NLPV) observer for
fault estimation in semi-active Electro-Rheological (ER) suspensions. The damper fault (a loss-
of-efficiency factor) is modeled as a lost force of unknown/free dynamics to be estimated.
Thanks to the parameter-dependent descriptor-form system modeling, there is no assumption
made on the fault dynamics, thus making this method applicable to all considered types of
damper faults. The nonlinearity in the damper model is bounded by its Lipschitz property,
while the road disturbance and the measurement noise are handled using the H∞ condition.
The observer is parameterized and then designed by solving Linear Matrix Inequalities (LMIs)
and is implemented in a polytopic gain scheduling approach. Synthesis results including Bode
plots and simulations illustrate the method in both the frequency and the time domains.
Keywords: Descriptor form, NLPV observer, fault estimation, semi-active suspension, Lipchitz
condition.
3.1 Parameterization of the Observer Matrices Remark: If the matrices N (ρ), J(ρ), H(ρ), R, and S can
be chosen according to (21), (22), (11), (23), and (24),
First note that from (11), we get respectively, then all conditions (10)-(12) are fulfilled.
H(ρ) = T B(ρ). From the results of above parameterization, for brevity,
Therefore, H(ρ) will be given when T is chosen. Now, the matrices of system (13) can be rewritten as
in order to determine T and the observer matrices N (ρ), A(ρ) = N (ρ) = A11 + Z1 (ρ)A12 , (25)
J(ρ), R, and S of the proposed observer satisfying all the
conditions equalities (10)-(12), parameterization is made B(ρ) = J(ρ)D2 − T D1 = B11 + Z1 (ρ)B12 , (26)
by using the general solution of (10) and (12). W(ρ) = −T B(ρ), (27)
C = R = C11 + Z2 C12 , (28)
First, from (10) and (12), one obtains
D = SD2 = D11 + Z2 D12 , (29)
N (ρ) J(ρ) TE TA
= . (14) where A11 = T Aα1 , A12 = β1 , B11 = T Aα2 D2 − T D1 ,
R S C I
B12 = β2 D2 , C11 = α1 , C12 = β1 , D11 = α2 D2 , and
The equation (14) is solvable if and only if D12 = β2 D2 . Notice that all the matrices A11 , A12 , B11 ,
TE B12 , C11 , C12 , D11 , and D12 are known and the matrix
C TE W(ρ) is known at each vertex of ρ. Therefore, the observer
rank = rank = nx . (15) design problem is reduced to determining Z1 (ρ) and Z2 ,
T A C
I which is discussed in the following part.
3.2 Polytopic H∞ Observer Design From (6), the following condition is obtained
(Φ(Ex) − Φ(E x̂))⊤ (Φ(Ex) − Φ(E x̂)) ⩽ e⊤ Γ⊤
e Γe e
Using (25)-(29), we rewrite the estimation error dynamics
(13) as ⇐⇒ (∆Φ)⊤ ∆Φ ⩽ (Cϵ + Dω)⊤ Γ⊤
e Γe (Cϵ + Dω)
⇐⇒ η ⊤ Q2 η ⩽ 0,
ϵ̇ = A(ρ)ϵ + W(ρ)∆Φ + B(ρ)ω (36)
(30)
e = Cϵ + Dω. −C⊤ Γ⊤ ⊤ ⊤
e Γe C 0 −C Γe Γe D
where Q2 = 0 I 0 .
After the parameterization step, while ∆Φ is bounded by ⊤ ⊤ ⊤ ⊤
−D Γe Γe C 0 −D Γe Γe D
the Lipschitz condition (6), the observer design problem is
now to determine the matrices Z1 (ρ) and Z2 such that By applying the S-procedure (Boyd et al., 1994) to the
• The system (30) is asymptotically stable for ω(t) = 0; inequalities (35) and (36), we have that V̇ +e⊤ e−γ 2 ω ⊤ ω <
• ∥e(t)∥L2 < γ∥ω(t)∥L2 for ω(t) ̸= 0; γ is minimized. 0 if there exists a scalar ϵl > 0 such that
V̇ + e⊤ e − γ 2 ω ⊤ ω − ϵl (η ⊤ Qη) < 0
In this paper, the design of the observer will be carried
out using the polytopic method. It means that we assume ⇐⇒ η ⊤ (Q1 (ρ) − ϵl Q2 )η < 0, (37)
that Z1 (ρ) depends in an affine way on the parameter ρ. The condition (37) is equivalent to
In such a case, the design method is restricted to solving Q1 (ρ) − ϵl Q2 < 0
the given problem only to get the vertices Z1,i of Z1 (ρ)
formed when ρ varies within its bounds (Apkarian et al., Ωa (ρ) XW(ρ) Ωb (ρ)
1995). Theorem 1 then solves the observer design problem ⇐⇒ W⊤ (ρ)X −ϵl I 0 < 0, (38)
⊤
in an LMI framework. Ωb (ρ) 0 Ωc
Theorem 1. Consider the system model (5) and the ob- where Ωa (ρ) = A(ρ)⊤ X + XA(ρ) + C⊤ C + ϵl C⊤ Γ⊤
e Γe C,
server (7). The observer design problem is solved if there Ωb = XB(ρ) + C⊤ D + ϵl C⊤ Γ⊤ ⊤
e Γe D, and Ωc = D D +
exist matrices X = X ⊤ > 0, Yi , Z2 , and a scalar ϵl > 0 ϵD⊤ Γ⊤ Γ
e e D − γ 2
I.
minimizing γ such that
Applying Schur’s complement to (38), one obtains
Ω11,i XW(ρi ) Ω13,i Ω14 Ω15
Ω1 (ρ) XW(ρ) XB(ρ) C⊤ C⊤ Γ⊤
⊤
W (ρi )X −ϵl I 0 0 0 e
Ω⊤
0 −γ 2
I Ω Ω
W⊤ (ρ)X −ϵl I 0 0 0
13,i 34 < 0,
35 (31) ⊤
B X 0 2 ⊤ ⊤ ⊤
−γ I D D Γe < 0, (39)
Ω⊤ Ω⊤
14 0 34 −I 0
0 −I 0
C D
Ω⊤15 0 Ω⊤
35 0 −ϵl I
Γe C 0 Γe D 0 −ϵl I
for ρi , i = 1, 2, . . . , 2 at the 2nρ vertices of ρ, where
nρ
-150
• The control u is obtained from a Skyhook controller.
It is important to note that in such a case, u varies
-150
e2
-150
e3
-200
0.4
-150
e4
0.01 0.3
-200 0.005
0.2
-150 0
e5
0.1
-0.005
-200
-50 -0.01 0
e6
1.5 1.5
Fig. 2. Transfer ∥e/żr ∥—Bode diagrams of NLPV observer 1 1
w.r.t the road profile derivative with ρ1 = 0 (red
Lost Force (N)
0.5 0.5
Error (N)
dashed line) and ρ2 = 1 (blue line). 0 0
-0.5 -0.5
-1 -1
(c) (d)
-100
-60
e2
-80
e3
-90
-100
0.8
-80
e4
-90 0.01
0.6
-100
0.4
0
-100
e5
0.2
-120 -0.01
0
-20 0 5 10 15 0 5 10 15
e6
-30
-40
(a) (b)
100 101 102 103 104
Frequency (Hz) 10
10 1
0.5
Lost Force (N)
0 5
Fig. 3. Transfer ∥e/ωn ∥—Bode diagrams of NLPV ob- 5
Error (N)
-0.5
0.5 1 1.5
0
server w.r.t the measurement noise with ρ1 = 0 (red 0
-10 -10
0 5 10 15 0 5 10 15
Time(s) Time(s)
5. SIMULATION RESULTS
(c) (d)
To emphasize the effectiveness of the proposed approach,
simulations are now performed considering the nonlinear Fig. 5. Simulation 2: (a) Road profile, (b) Loss-of-efficiency
quarter-car model (5). The initial conditions are x(0) = factor, (c) Fault estimation, and (d) Estimation error.
The simulation results are shown in Fig. 4 and Fig. 5. It Guzman, J., López-Estrada, F.R., Estrada-Manzo, V., and
can be seen that the asymptotic estimation is achieved Valencia-Palomo, G. (2021). Actuator Fault Estimation
with small errors (see Table 1) and for various kinds of based on a Proportional-integral Observer with Non-
variations in the fault, which highlights the advantage that quadratic Lyapunov Functions. International Journal
our method is appropriate for all fault dynamics. of Systems Science, 1–14.
Ha, Q.P. and Trinh, H. (2004). State and Input Simul-
Table 1. Normalized Root-Mean-Square Errors taneous Estimation for a Class of Nonlinear Systems.
(NRMSE). Automatica, 40(10), 1779–1785.
Hernández-Alcántara, D., Tudón-Martı́nez, J.C.,
Simulation NRMSE (-) Amézquita-Brooks, L., Vivas-López, C.A., and
Scenario 1 0.0315
Morales-Menéndez, R. (2016). Modeling, Diagnosis and
Scenario 2 0.0091
Estimation of Actuator Faults in Vehicle Suspensions.
Control Engineering Practice, 49, 173–186. doi:
6. CONCLUSION https://doi.org/10.1016/j.conengprac.2015.12.002.
Koenig, D. (2006). Observers Design for Unknown Input
Nonlinear Descriptor Systems via Convex Optimization.
This paper presents an NLPV observer to estimate the IEEE Transactions on Automatic control, (06), 1047–
damper fault (modeled as the lost damper force) in SA 1052.
ER automotive suspensions. The descriptor NLPV mod- Morato, M.M., Sename, O., Dugard, L., and Nguyen,
eling avoids any assumption made on the fault dynamics, M.Q. (2019). Fault Estimation for Automotive
making the approach suitable for any faults. While the Electro-Rheological Dampers: LPV-based Observer Ap-
Lipschitz condition is used to bound the system nonlinear- proach. Control Engineering Practice, 85, 11–22. doi:
ity, the combined effects of unknown inputs (road profile https://doi.org/10.1016/j.conengprac.2019.01.005.
derivative and measurement noise) on the estimation error Morato, M., Pham, T.P., Sename, O., and Dugard, L.
are minimized using the H∞ condition. Both frequency- (2020). Development of a Simple ER Damper Model for
domain analysis and time-domain simulations assess the Fault-tolerant Control Design. Journal of the Brazilian
performance of the method. Society of Mechanical Sciences and Engineering, 42(10),
502 (2020). doi:10.1007/s40430-020-02585-y.
REFERENCES Nguyen, M.Q., Sename, O., and Dugard, L. (2016). Com-
parison of Observer Approaches for Actuator Fault Es-
Apkarian, P., Gahinet, P., and Becker, G. (1995). Self- timation in Semi-active Suspension Systems. In 2016
scheduled H∞ Control of Linear Parameter-varying Sys- 3rd Conference on Control and Fault-Tolerant Systems
tems: A Design Example. Automatica, 31(9), 1251–1261. (SysTol), 227–232. IEEE.
Boyd, S., El Ghaoui, L., Feron, E., and Balakrishnan, Osorio-Gordillo, G.L., Darouach, M., Astorga-Zaragoza,
V. (1994). Linear Matrix Inequalities in System and C.M., and Boutat-Baddas, L. (2019). Generalised Dy-
Control Theory, volume 15. Siam. namic Observer Design for Lipschitz Non-linear De-
Darouach, M. and Boutat-Baddas, L. (2008). Observers scriptor Systems. IET Control Theory & Applications,
for a Class of Nonlinear Singular Systems. IEEE 13(14), 2270–2280.
Transactions on Automatic Control, 53(11), 2627–2633. Pham, T.P., Sename, O., and Dugard, L. (2019). Unified
doi:10.1109/TAC.2008.2007868. H∞ Observer for a Class of Nonlinear Lipschitz Sys-
Darouach, M., Amato, F., and Alma, M. (2017). tems: Application to a Real ER Automotive Suspen-
Functional Observers Design for Descriptor sion. IEEE Control Systems Letters, 3(4), 817–822. doi:
Systems via LMI: Continuous and Discrete- 10.1109/LCSYS.2019.2919813.
time Cases. Automatica, 86, 216–219. doi: Poussot-Vassal, C., Spelta, C., Sename, O., Savaresi, S.M.,
https://doi.org/10.1016/j.automatica.2017.08.016. and Dugard, L. (2012). Survey and Performance Evalu-
Darouach, M., Boutat-Baddas, L., and Zerrougui, M. ation on some Automotive Semi-active Suspension Con-
(2011). H∞ Observers Design for a Class of Nonlinear trol Methods: A Comparative Study on a Single-corner
Singular Systems. Automatica, 47(11), 2517–2525. Model. Annual Reviews in Control, 36(1), 148–160.
Delshad, S.S., Johansson, A., Darouach, M., and Gustafs- Savaresi, S.M., Poussot-Vassal, C., Spelta, C., Sename, O.,
son, T. (2016). Robust State Estimation and Unknown and Dugard, L. (2010). Semi-active Suspension Control
Inputs Reconstruction for a Class of Nonlinear Systems: Design for Vehicles. Elsevier.
Multiobjective Approach. Automatica, 64, 1–7. Sename, O. (2021). Review on LPV Approaches for
Do, M.H., Koenig, D., and Theilliol, D. (2018). Suspension Systems. Electronics, 10(17), 2120.
Robust H∞ Proportional-Integral Observer for Tran, G.Q.B., Pham, T.P., and Sename, O. (2022).
Fault Diagnosis: Application to Vehicle Suspension. Multi-objective Grid-based Lipschitz NLPV PI
IFAC-PapersOnLine, 51(24), 536–543. doi: Observer for Damper Fault Estimation. In
https://doi.org/10.1016/j.ifacol.2018.09.628. 10th 11th IFAC Symposium on Fault Detection,
IFAC Symposium on Fault Detection, Supervision and Supervision and Safety for Technical Processes
Safety for Technical Processes SAFEPROCESS 2018. - SAFEPROCESS 2022. Paphos, Cyprus. URL
Do, M.H., Koenig, D., and Theilliol, D. (2020). H∞ https://hal.archives-ouvertes.fr/hal-03611561.
Observer Design for Singular Nonlinear Parameter-
varying System. In 2020 59th IEEE Conference
on Decision and Control (CDC), 3927–3932. doi:
10.1109/CDC42340.2020.9303844.