You are on page 1of 15

Received: 17 May 2020 Revised: 17 March 2021 Accepted: 19 September 2021

DOI: 10.1002/oca.2808

RESEARCH ARTICLE

Unknown input observer design for vehicle lateral


dynamics described by Takagi–Sugeno fuzzy systems

Naoufal El Youssfi1 Rachid El Bachtiri1 Taha Zoulagh2 Hicham El Aiss2

1
TSI Laboratory, Sidi Mohamed Ben
Abdellah University, Fez, Morocco Abstract
2
Electrical Engineering Department, This article discusses the problem of an unknown input observer for non-
University of Santiago, Santiago, Chile linear lateral dynamic systems. Takagi–Sugeno fuzzy models have been
used to represent the vehicle lateral system with parameters uncertain. The
Correspondence
Naoufal El Youssfi, Industrial observer used is well suited to estimate the vehicle state and actuator/sensor
Technologies and Services Laboratory faults, simultaneously. By using Finsler’s lemma, the conditions obtained are
(Lab-TSI), Higher School of Technology,
Sidi Mohamed Ben Abdellah University,
expressed through linear matrix inequality constraints. Finally, the results of
Fez, Morocco. the simulation clearly show the importance and effectiveness of the proposed
Email: ely.naoufal@gmail.com approaches.
Funding information
KEYWORDS
Fondo Nacional de Desarrollo Científico y
Tecnológico, Grant/Award Numbers: Finsler’s lemma, linear matrix inequality, Lyapunov function, Takagi–Sugeno fuzzy model,
3190378, 3200854 unknown input observer, vehicle lateral dynamics

1 IN T RO DU CT I ON

In recent decades, the automotive industry and the research laboratories have worked hard to solve transport-related
problems and to produce smarter, reliable, and safer vehicles. The vehicles produced have been equipped with pas-
sive safety elements, such as airbags, safety belts, and side protection bars,1 that protect human life during an accident.
In addition, driver-assisted corrective safety devices, such as adaptive cruise control, it is an improved speed control
series, it uses evident proximity sensors (e.g., laser, radar, and camera sensors), it not only maintains a cruising speed
but also detects and ensures a safe distance to the object in front of it in the same trajectory.2 The anti-lock braking
system is also an active safety system, its principle is to adjust the deceleration power of each wheel by adapting the
level of pressure exerted, it prevents the wheels to lock during emergency braking in order to keep the vehicle steering
and avoid an obstacle.3,4 The electronic stability program is also one of the active safety devices, its main role is to con-
trol the direction by affecting the braking system and the engine torque, it operates when the vehicle suddenly turns at
high-speed.4,5
The above processes are normally controlled by both known and unknown inputs. Their functioning and efficiency
require accurate knowledge of the parameters of the vehicle dynamic. The measurements of these parameters sometimes
do not provide complete system details, because of the unavailability of some state measurements. Furthermore, the
number of sensors is reduced due to cost reasons and can sometimes be also unavailable. The idea is to use a software
sensor, which is called observer, able to reconstruct accurately the unmeasurable and measurable states from measurable
ones. Thus, it should be able to build unknown inputs of the system from the model and calculated parameters. Different
estimation techniques have been applied in this context to solve problems of observer design. Among these techniques,
we can cite Luenberger’s classic observer,6,7 which is based on the synthesis of a static gain to ensure the convergence of
observer states to real system states and to stabilize the estimation error. However, disturbances on system often lead to
bad reconstruction, and instability. In References 8 and 9, Kalman filter has been used which is robust to measurement

354 © 2021 John Wiley & Sons Ltd. wileyonlinelibrary.com/journal/oca Optim Control Appl Meth. 2022;43:354–368.
EL YOUSSFI et al. 355

noise. In References 10 and 11, sliding mode observer has been implemented to ensure the asymptotic decay of the state
estimation error even in the presence of uncertainties in the system input and nonlinearities.
These observers may not be effective when the dynamics of the system are subject to the influence of unknown inputs.
Habitually, these unknown inputs come from sensor faults, modeling errors, disturbances, or noises. For this reason,
unknown input observer (UIO) is studied in References 12 and 13, it has been shown that it allows to reconstruct the
system states even in the presence of unknown inputs and it is often involved in the diagnosis, for the detection and
estimation of faults affecting the system.
The problem of observing a system affected by unknown inputs has been widely studied. In Reference 14, the authors
studied the observer design for discrete-time Takagi–Sugeno (T-S) fuzzy systems with the existence of uncertainties. While
in Reference 15, the authors have been considered the uncertainty as an unknown input, with a particular derivation.
However, in References 16 and 17, a dynamic observer has been proposed with unknown input so that the augmented
system can simultaneously estimate the sensor fault and the system state without considering the actuator fault. The
design of the UIO for descriptor systems described by the T-S framework and for a class of unknown input switching
descriptor systems are discussed in References 18 and 19, respectively. Previous results have focused only on sensor faults
and sometimes in the presence of limited uncertainties. Observing the actuator/sensor faults does not take more attention
and this motivates this work.
To our best knowledge, the design of the UIO is rarely used to estimate the lateral dynamics states of the automotive
vehicle based T-S fuzzy systems and actuator/sensor faults, simultaneously. For instance, we can refer to these works20-22
and the references therein. Our main contribution is to accurately estimate the vehicle’s lateral dynamics by an UIO with
the presence of uncertainties and sensor/actuator faults. The most commonly used model to explain the lateral motions
of the vehicle is the model of the bicycle, designed in References 23-25, which is illustrated by the T-S multi-model,
widely used in the literature to overcome problems of nonlinearity in nonlinear processes.26-28 It consists of developing
the global model through linear local model interpolation. This method explains accurately the behavior of nonlinear
processes, including the lateral dynamics system of the vehicle. Estimation error stability is mainly studied using the
Lyapunov quadratic function, sufficient asymptotic stability conditions are given in the form of linear matrix inequalities
(LMIs), which can be solved very efficiently using LMI optimization techniques. The results are improved using Finsler’s
lemma.29,30
The structure of this article is organized as follows: Section 2 deals with the model of vehicle lateral dynamics,
as well as its T-S fuzzy representation. Section 3 describes the problem. Section 4 presents the UIO design for an
uncertain T-S fuzzy system. Section 5 is devoted to simulations and results analysis. Finally, the conclusion is given
in Section 6.
Notations: The following notations are taken into account. Rn means the n-dimensional Euclidean space, I is the
identity matrix with appropriate dimension. The notations A < 0 and A ≤ 0 indicate that the matrix A is negatively defined
and negatively semi defined, respectively. The inverse of a matrix A is expressed by A−1 and its transpose indicated by AT .
∗ means symmetrical terms in a symmetrical matrix. A† denotes a generalized inverse of A.

2 VEHICLE MODELING

2.1 Nonlinear vehicle model

Vehicle dynamics are strongly influenced by the interaction with the tires and road conditions, which are difficult
to predict. The complete vehicle dynamics model is documented in several contributions, for example, References 31
and 32, which, due to its complexity and its high freedom degrees, is extremely difficult to be used in control and moni-
toring applications. For this reason, it is required to use a nominal model for the synthesis of observers and controls. The
model used in this application describes the lateral dynamics of the vehicle (see Figure 1),33 based on the bicycle model.
The lateral speed vy and the yaw rate 𝜓̇ of the vehicle are considered as differential variables.
The following differential equations can explain the lateral dynamics of the vehicle

⎧ 1 ( )
⎪v̇ y (t) = mv
2Fyf + 2Fyr − vx 𝜓(t),
̇
⎨ 1 ( ) (1)
̈ =
⎪𝜓(t) Iz
2lf Fyf − 2lr Fyr + Mz (t) ,

356 EL YOUSSFI et al.

FIGURE 1 Single-track model

where vx and vy are longitudinal and lateral velocities, respectively, 𝜓̇ is the yaw rate, mv is the vehicle mass, Mz is the
external yaw moment, Iz is inertia moments around the vertical axis, Fyf and Fyr are lateral tire forces at the front and the
rear wheels, respectively. The system is described in Reference 34 under the form (2).

[ ] l C −l C [ ]
⎡ − 2 Cf +Cr
v̇ y (t) −vx − 2 f mf vr r ⎤ vy (t) ⎡2 Cf ⎤ ⎡0⎤
m v
= ⎢ l C −l C
v x
2 2
lf Cf +lr Cr
v x ⎥ + ⎢ mv vx ⎥
𝛿f (t) + ⎢ ⎥ Mz (t). (2)
𝜓(t)
̈ ⎢− 2 f f r r −2 ⎥ 𝜓(t)
̇ ⎢ 2 lf Cf ⎥ ⎢1⎥
⎣ Iz v x Iz v x ⎦ ⎣ I z ⎦ ⎣ Iz ⎦
⏟⏟⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏟⏟⏟ ⏟⏟⏟ ⏟⏟⏟
x(t)
̇ A x(t) Bf B

2.2 T-S fuzzy representation for vehicle lateral dynamics

A T-S fuzzy model is a set of linear time-invariant (LTI) models, combined with nonlinear membership functions. There
are different ways to run a nonlinear systems T-S model. A nonlinear sector transformation is an interesting approach
that allows correct T-S representation to be achieved without loss of information on a compact state space set. By using
identification and linearization of the cornering forces on the vehicle which can be approximated as in Reference 35. They
are given by the following expressions

⎧ ∑2
⎪Fyf (t) = 𝜌i (𝜉(t))Cfi 𝛼f (t),
⎪ i=1
⎨ 2 (3)
⎪F (t) = ∑𝜌 (𝜉(t))C 𝛼 (t),
⎪ yr i ri r
⎩ i=1

and

𝜉(t) = |𝛼f (t)|,

where Cfi , Cri , and 𝜉(t) are the front, the rear tire cornering stiffness, and absolute value of slip angle 𝛼f (t), respectively. Cfi
and Cri depend on road adhesion 𝜌 and vehicle mass mv , 𝛼f and 𝛼r are slip angles of the front and the rear tires, respectively,
which are given in Reference 36 as follows

⎧ lf 𝜓̇
⎪𝛼f = 𝛿f − 𝛽 − vx ,
⎨ lr 𝜓̇
(4)
⎪ 𝛼r = −𝛽 + vx ,

where 𝛿f (t) is the road wheel steer angle, commanded by the driver and 𝛽 is the slip angle, with

vy
𝛽≈ . (5)
vx
EL YOUSSFI et al. 357

Membership functions 𝜌i (𝜉(t)) are given as follows

𝜎i (𝜉(t))
𝜌i (𝜉(t)) = ∑2 ; i = 1, 2. (6)
i=1 𝜎i (𝜉(t))

with

1
𝜎i (𝜉(t)) = ( ) . (7)
| 𝜉(t)−ci | 2bi
1+| a |
| i |

The membership functions (6) satisfy the following properties

2
⎧∑
⎪ 𝜌i (𝜉(t)) = 1,
⎨ i=1 (8)
⎪0 ≤ 𝜌 (𝜉(t)) ≤ 1, i = 1, 2.
⎩ i

We obtain the T-S fuzzy systems of the vehicle lateral dynamics model by replacing the lateral forces in the nonlinear
model (1) with their fuzzy expressions (3). Then the lateral motion can be expressed by this model system

2
⎧ ∑ { }
⎪ x(t)
̇ = 𝜌i (𝜉(t)) Ai x(t) + Bfi 𝛿f (t) + BMz (t) ,
⎨ i=1
(9)
⎪y(t) = Cx(t),

where x(t) ∈ Rn is system state vector, y(t) ∈ Rp is system output vector, 𝛿f (t) ∈ Rm is steering angle given by the driver.
Ai ∈ Rn×n , Bfi ∈ Rn×m , B, and C ∈ Rp×n are constant matrices with compatible dimensions.

3 PROBLEM DESCRIPTION

Vehicle lateral dynamics may exhibit unpredictable dangerous behavior in the presence of abnormal external conditions
such as side wind force, road adhesion coefficient variance, and so forth, so that sensor/actuator faults and appropriate
uncertainties have to be implemented as in Equation (10) to solve this problem.

2
⎧ ∑
⎪x(t)
̇ = 𝜌i (𝜉(t)){(Ai + ΔAi (t))x(t) + (Bfi + ΔBfi (t))𝛿f (t) + BMz (t) + Di f (t)},
⎨ i=1
(10)
⎪y(t) = Cx(t) + Ff (t),

where f (t) ∈ Rs is faults affected to both actuator and sensor. Di ∈ Rn×s and F ∈ Rp×s are constant matri-
ces with compatible dimensions. ΔAi (t) ∈ Rn×n and ΔBfi (t) ∈ Rn×m are matrix functions which represent time
varying parameter uncertainties affecting the state and the input matrices, respectively, which are structures
as follows:

ΔAi (t) = MΔÃi (t), ΔBfi (t) = MΔB̃ fi (t), (11)

where M is a full column rank matrix.


Based on the uncertainties form (11), the T-S model (10) can be rewritten to have the model (12).

2
⎧ ∑
⎪x(t)
̇ = 𝜌i (𝜉(t)){Ai x(t) + Bfi 𝛿f (t) + BMz (t) + Di f (t) + M(Δi (t)x(t) + ΔB̃ fi (t)𝛿f (t))},
⎨ i=1
(12)
⎪y(t) = Cx(t) + Ff (t).

358 EL YOUSSFI et al.

Let’s put the following equalities


}
hi (t) = ΔÃi (t)x(t)
→ vi (t) = ki (t) + hi (t) (13)
ki (t) = ΔB̃ fi (t)u(t)

then, by replacing vi (t) in (12), the system becomes as illustrated in (14).

2
⎧ ∑
⎪x(t)
̇ = 𝜌i (𝜉(t)){Ai x(t) + Bfi 𝛿f (t) + BMz (t) + Mvi (t) + Di f (t)},
⎨ i=1
(14)
⎪y(t) = Cx(t) + Ff (t).

The augmented system formed from the system (14) and the fault f (t) can be expressed as shown in Equation (15).

⎧ ⎧ ⎫
⎪ ⎪ ⎪
⎪[ ] 2 ⎪[ ][ ] [ ][ ][ ][ ]⎪
⎪ x(t)
̇ ∑ ⎪ Ai Di x(t) Bfi B 𝛿f (t) M 0 vi (t) ⎪
⎪ = 𝜌i (𝜉(t)) ⎨ + ⎬,
̇
⎪ f (t) i=1 ⎪ 0 0 f (t) 0 0 Mz (t) 0 I ḟ (t) ⎪
⎪⏟⏟⏟ ⎪⏟⏞⏞⏟⏞⏞⏟⏟⏟⏟ ⏟⏞⏞⏟⏞⏞⏟⏟⏞⏟⏞⏟⏟⏞⏟⏞⏟⏟⏟⏟⎪
⎨ z(t) ⎪ z(t) u(t) wi (t) ⎪
(15)
̇ Ā i Bi M
⎪ [ ]⎩ ⎭
⎪ [ ] x(t)
⎪y(t) = C F .
⎪ f (t)
⎪ ⏟⏟⏟
⎪ ⏟⏟⏟
C
⎩ z(t)

From (15), the augmented system can be written as follows

⎧ ∑2 { }
⎪z(t)
̇ = 𝜌i (𝜉(t)) Ā i z(t) + Bi u(t) + Mwi (t) ,
⎨ i=1
(16)
⎪y(t) = Cz(t).

In order to estimate the nonmeasurable states and faults in the next section, the following assumptions and lemmas
are required before proceeding to the observer design of the system (16).
Assumption 1. The matrices C and M are full row and full column rank, respectively; to ensure the existence of the
general solution (42).
Assumption 2 (37).

|𝜌̇ q (𝜉(t))| ≤ 𝜆q ,
{

𝜆q ≥ 0.
(17)

Lemma 1 (Finsler’s lemma29 ). Let 𝜁 ∈ Rn , Z ∈ Rn×n , and Σ ∈ Rm×n with rank(Σ) < n and (Σ)⊥ such that ΣΣ⊥ = 0. Then,
the following conditions are equivalent:

𝜁 T Z𝜁 < 0, ∀𝜁 ≠ 0 ∶ Σ𝜁 = 0, (18a)

ΣT⊥ ZΣ⊥ < 0, (18b)

∃𝜇 ∈ R ∶ Z − 𝜇ΣT Σ < 0, (18c)

∃Q ∈ Rn×m ∶ Z + QΣ + ΣT QT < 0. (18d)


EL YOUSSFI et al. 359

FIGURE 2 Observation scheme

Lemma 2 (38). Consider the matrix U ∈ Rnxm , with n ≥ m, and matrix V ∈ Rnxk , the matrix M with the form

M = VU + + N(I − UU + )

is a solution of MU = V when the condition VU + U = V holds. N ∈ Rkxm is an arbitrary matrix and U + is the Moore–Penrose
pseudo-inverse of U which is denoted

U + = (U T U)−1 U T .

As mentioned in Section 1, the next section will be about actuator/sensor fault and state estimation. The block diagram
illustrated in Figure 2 presents the concept of the UIO, where u(t) is the control input, f (t) is the fault signal, f̂ (t) and x̂ (t)
are the fault and state estimation, respectively, and y(t) is the system output.

4 UI O D E S I G N FOR U N C E RTA IN T- S FU ZZY SY ST EM

In this section, our goal is to design an UIO using Finsler’s lemma. This observer can simultaneously estimate the state
of the vehicle and the sensor/actuator faults, despite the presence of uncertainties that simultaneously affects the state
and input matrices of the vehicle lateral dynamics system.
Consider the following observer of the same structure as the observer presented in Reference 33:

⎧ 2
⎪ ∑
⎪p(t)
̇ = 𝜌i (𝜉(t)) {Ni p(t) + Gi u(t) + Li y(t)} ,
⎨ i=1 (19)

⎪ẑ (t) = p(t) − Hy(t).

Ni ∈ Rnxn , Gi ∈ Rnxm , Li ∈ Rnxp , and H ∈ Rnxp are the observer’s parameters to be determined; p(t) and ẑ (t) are the
observer states and augmented system estimation, respectively. The error between faulty system (16) and the observer
(19) is given by

e(t) = ẑ (t) − z(t) (20)

= p(t) − (I + HC)z(t) (21)

= p(t) − Tz(t) (22)

with T = I + HC.
The dynamic of the estimation error (22) is written by the following equation:
360 EL YOUSSFI et al.

e(t)
̇ = p(t)
̇ − T z(t)
̇ (23)
2
∑ { }
= 𝜌i (𝜉(t)) Ni p(t) + Gi u(t) + Li y(t) − T Ā i z(t) − TBi u(t) − TMwi (t) (24)
i=1
∑2 { }
= 𝜌i (𝜉(t)) (Ni T + Li C − T Ā i )z(t) + (Gi − TBi )u(t) − TMwi (t) + Ni e(t) . (25)
i=1

If these conditions are satisfied

Ni T + Li C − T Ā i = 0, (26a)

Gi − TBi = 0, (26b)

TM = 0. (26c)

Then (25) is rewritten as follows

2

e(t)
̇ = 𝜌i (𝜉(t)) {Ni e(t)} . (27)
i=1

The following theorem presents the necessary condition to be satisfied system (27) to be asymptotically stable.
Theorem 1. For a given positive scalar 𝜆q , the states and faults of the system (16) are estimated asymptotically with the
observer (19) if there exist the matrices , , i and positive symmetric matrix i such that the following conditions hold

⎢ 𝜆q q + 𝛼(Πi + Πi )
2
⎡∑ T

Ξij ⎥
⎢ q=1 ⎥ < 0, i, j = 1, 2, (28)
− − T ⎦
⎢ T ⎥
⎣ Ξij

where

Πi = i − i C + i , (29)

Ξij = j − 𝛼 + Ti T − C i + iT  ,


T T T
(30)

i = (I + WC)Ā i , (31)

i = Ut CĀ i , (32)

W = −M(CM)+ , (33)

(CM)+ = ((CM)T (CM))−1 (CM)T , (34)

Ut = I − (CM)(CM)+ . (35)

The existence of , , and i satisfies the correctness of Equations (36) and (37). Furthermore, the linearity of our
matrix inequality formulation is guaranteed.

Y = −1 , (36)

Qi = −1 i . (37)
EL YOUSSFI et al. 361

So, the observer gains can be derived as follows:

H = W + YUt , (38)

Ni = T Ā i − Qi C, (39)

Gi = TBi , (40)

Li = Qi (I + CH) − T Ā i H. (41)

Proof. From (26c), we have

(I + HC)M = 0 ⇔ H(CM) = −M. (42)

By using the same notation given in Lemma 2, where M = H, U = C M, and V = −M, the general solution of (42)
according to Assumption 1 and Lemma 2 is presented by (38); where Y is the solution of conditions in (28).
Let’s consider Qi = Li + Ni H as defined in Reference 13, and by substituting it in (26a), we get

Ni = T Ā i − Qi C (43)

then, we can write

Li = Qi (I + CH) − T Ā i H (44)

by combining the formula expressing T and the one given in (43), we find

Ni = (I + HC)Ā i − Qi C (45)

= (I + WC)Ā i + (YUt C)Ā i − Qi C. (46)

Let us consider the following Lyapunov function candidate

2

V(t) = 𝜌i (𝜉(t))eT (t)i e(t), (47)
i=1

where i = iT > 0. Substituting (27) into (47) and taking the time derivative gives

2 2
∑ ∑ { }
V(t)
̇ = 𝜌̇ i (𝜉(t))eT (t)i e(t) + 𝜌i (𝜉(t)) ė T (t)i e(t) + eT (t)i e(t)
̇ , (48)
i=1 i=1
{ }

2 2
∑ ∑
V(t)
̇ 𝜌i (𝜉(t)) 𝜆q eT (t)q e(t) + ė T (t)i e(t) + eT (t)i e(t)
̇ . (49)
i=1 q=1

By using Finsler’s lemma and from (49), we can write

𝜆q q i ⎥
2
⎡∑ ⎤
T ⎢
𝜁 ⎢ q=1 ⎥𝜁 < 0 (50)
⎣ i
⎢ ⎥
0⎦
⏟⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏟
Z
362 EL YOUSSFI et al.

such that
[ ]
[ ] e(t)
Ni −I = 0; ΣΣ⊥ = 0 (51)
⏟⏞⏞⏟⏞⏞⏟ e(t)
̇
Σ ⏟⏟⏟
𝜁

with Σ⊥ is introduced as the orthogonal complement of Σ.


From (50) and (51), the following condition is satisfied

⎢ 𝜆q q i ⎥  [ ]T 
2
⎡∑ ⎤ [ ] [ ]T
] [
 
⎢ q=1 ⎥+ Ni −I + Ni −I < 0, (52)
⎣ i
⎢ ⎥
0⎦

which is equivalent to this inequality

⎢ 𝜆q q + Ni + NiT T j −  + NiT T ⎥


2
⎡∑ ⎤
⎢ q=1 ⎥ < 0. (53)
− − 
⎢ T ⎥
⎣ ∗ ⎦
It should be noted that Equation (53) is a bilinear matrix inequality (BMI) that can be transformed into a LMI by
considering  = 𝛼 and the variable change N i = Ni . However, the obtained condition cannot be solved since it could
provide a nonunique solution. Furthermore, we cannot provide any simulation test since obtained condition does not
have any constant matrix. So, a new development is needed.
Now, by replacing Ni by its expression in (46), Equation (53) can be rewritten as follow

⎢ 𝜆q q + Πi + Πi
2
⎡∑ T ⎤
Ξij⎥
⎢ q=1 ⎥ < 0, (54)
− −  ⎦
⎢ T
T⎥
⎣ Ξij
Πi = i − Qi C + Y i ,

Ξij = j −  + Ti T − C QTi T + iT Y T T


T

by substituting  = 𝛼,  = Y and i = Qi , the inequality of the linear matrix (28) can be determined.
This completes the proof. ▪
Remark 1. In this article, we have tackled free slack variables as a method to improve the result. Consequently, the matrix
 has been involved as slack in order to obtain less conservative results and to get good states/fault estimation for vehicle
lateral dynamics
Remark 2. The given 𝜆k and 𝛼 are determined based on Assumption 2 and Finsler’s lemma, respectively. The choice of
𝜆k and 𝛼 is made arbitrarily in such a way that the conditions of Theorem 1 are feasible and with taken into account the
quality of simulation results.
The standard method of solving the constraints of Theorem 1 is given by the following algorithm:

Algorithm

• Step 1: If assumption 1 is satisfied, Equation (33) is used to determine the matrix W. Subsequently, the matrix i is
deduced using Equation (31).
• Step 2: Solving the matrix inequalities (28) where the matrices Πi and Ξij are substituted by their expressions (29) and
(30), respectively, yields the matrices , , and i . Thereafter, the matrices Qi and Y are deduced from Equations (36)
and (37).
EL YOUSSFI et al. 363

• Step 3: Then, the observer gains Ni , H, Gi , and Li are calculated using Equations (37)–(40).
Remark 3. Using the algorithm, Assumption 1 should be satisfied, that is, M and C must be full column rank. Thus,
this constraint is satisfied from the fact that our system is formulated by taking in consideration that M is full col-
umn rank as given in (11) and (15). From the other hand, C is taken a full column rank matrices in the numerical
example.

5 S IM U L AT IO N R E S U LT S

We conducted some simulations using the vehicle model (1) and MATLAB software to illustrate the observer’s efficiency
in estimating vehicle states and faults. We take the longitudinal velocity as it is constant vx = 25 m/s, we have considered
the steering angle shown in Figure 3 and the values of other parameters are listed in Table 1.35
The simulation results are obtained without control, which means the yaw moment is equal to zero (Mz (t) = 0). Note
that each (Ai , C) is observable.
Parameters of membership functions and stiffness coefficients are given in Table 2.

0.06

0.04

0.02

-0.02

-0.04

-0.06

0 2 4 6 8 10 12 14 16 18 20

FIGURE 3 Steering angle given by driver 𝛿f (t) [rad]

T A B L E 1 Simulation vehicle parameters


Constants Explanation Value Unit

mv Vehicle total mass 1653 + 2 × 90 kg


Iz Yaw inertia moment at the CG 2765 kg m2
lf Distance from the CG to the front axle 1.402 m
lr Distance from the CG to the rear axle 1.646 m

T A B L E 2 Nominal stiffness and membership function coefficients


Nominal stiffness coefficients Cf 1 Cf 2 Cr1 Cr2
Values 55,234 15,544 49,200 13,543
Membership functions coefficients
a1 = 3.1893, b1 = 0.5077, c1 = 0.9496
a2 = 0.5633, b2 = 5.3907, c2 = 0.8712
364 EL YOUSSFI et al.

The parameter matrices of system (14) are:


[ ] [ ]
1 0 [ ] 0.2 0 [ ]
C= ; D1 = D2 = 0.5 0.5 ; M= ; F = 0 0.5 .
0 1 0 0.2

Assuming that the fault f (t) has the following form

⎧0.46t − 2 0 ≤ t ≤ 5,
5 ≤ t ≤ 10,

f (t) = ⎨1 (55)
t ≥ 10.

⎩sin(3t)

Selecting (𝜆1 = 0.8, 𝜆2 = 0.8) and solving LMIs of the previous theorem, by using LMI toolbox,39 the following observer
gains are recovered

⎡− 6.2877 0.0000 0.0000 ⎤ ⎡− 6.2877 0.0000 0.0000 ⎤


⎢ ⎥ ⎢ ⎥
N1 = ⎢− 0.0000 −6.2928 0.5155 ⎥ ; N2 = ⎢− 0.0000 −6.2921 0.4475 ⎥ ,
⎢ ⎥ ⎢ ⎥
⎣ 0.0000 0.5155 −57.8366⎦ ⎣ 0.0000 0.4475 −51.0409⎦

⎡ 0 0 ⎤ ⎡ 0 0 ⎤ ⎡− 1 0 ⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
G1 = ⎢ 0.000 72.96 ⎥ ; G2 = ⎢ 0.001 11.37 ⎥ ; H=⎢ 0 0 ⎥,
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣− 0.036 −7296.07⎦ ⎣− 0.036 −1137.11⎦ ⎣ 0 −100⎦

⎡ 0 0 ⎤ ⎡ 0 0 ⎤ ⎡ 0.2538 0.0000 −0.0000⎤


 = ⎢ 0.0000 0.2538 0.0023 ⎥ ,
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
L1 = ⎢ 0.261 50.00 ⎥ ; L2 = ⎢− 0.067 50.00 ⎥ ;
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣− 26.059 −5000.00⎦ ⎣ 6.678 −5000.00⎦ ⎣− 0.0000 0.0023 0.0256 ⎦

⎡ 1.6259 0.0000 −0.0000⎤ ⎡ 1.6259 0.0000 −0.0000⎤


1 = ⎢ 0.0000 1.6259 0.0026 ⎥ ; 2 = ⎢ 0.0000 1.6259 0.0026 ⎥ .
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎣− 0.0000 0.0026 1.3618 ⎦ ⎣− 0.0000 0.0026 1.3618 ⎦

The system of vehicle lateral dynamics was simulated choosing the fault presented in red in Figure 4. It is clear that
the observer’s estimate of the gains obtained by the previous theorem converges rapidly toward the fault, just after it

-2

-4
0 2 4 6 8 10 12 14 16 18 20

FIGURE 4 Fault f (t) and its estimates


EL YOUSSFI et al. 365

-1

-2

-3

0 2 4 6 8 10 12 14 16 18 20

FIGURE 5 Lateral velocity vy (t) and its estimates

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6
0 2 4 6 8 10 12 14 16 18 20

FIGURE 6 Yaw rate 𝜓(t)


̇ and its estimates

0.4

0.2

-0.2

-0.4

-0.6
0 2 4 6 8 10 12 14 16 18 20

FIGURE 7 Membership functions derivative


366 EL YOUSSFI et al.

[ ]T
has appeared, against the result obtained by Theorem 2 of Reference 13, initial conditions are: x(t) = 0 0 and ẑ (t) =
[ ]T
0.2 0.2 0.2 .
Figures 5 and 6 represent the lateral velocity vy (t) and the yaw rate 𝜓(t)
̇ with their estimates by different approaches,
respectively. We clearly see a relaxed estimation of states despite the presence of uncertainties. Noticing that the previous
theorem presents the swiftness compared to the theorem of Reference 13 during estimation of the vehicle lateral dynamics
system states.
Figure 7 shows the membership function derivative 𝜌̇ 1 (𝜉(t)) and 𝜌̇ 2 (𝜉(t)) as considered in Assumption 2.
Remark 4. From Figure 7, we can see clearly that the conditions proposed in Assumption 2 are satisfied and the derivative
of membership functions did not overcome its upper bound 𝜆1 and 𝜆2 . This means that the proposed conditions respect
its hypothesis.

6 CO N C LU S I O N

In this article, we have proposed a new UIO design to estimate actuator/sensors faults and states of uncertain vehicle
lateral dynamics system, which is represented by T-S fuzzy systems. The coincidence of the estimation error to zero is
studied with the Lyapunov approach and LMI constraints, which are provided to design the matrices of the different
components of the UIO. Refinement has been proposed through Finsler’s lemma. The vehicle simulations show clearly
the quality of states and fault estimation of the vehicle dynamics and the proposed approach can be adapted to driving
conditions.

AC K N OW L E D G M E N T S
Hicham El Aiss and Taha Zoulagh have been supported by “Fondo Nacional de Desar-rollo Cient𝚤fico y
Tecnologico”-Fondecyt, Chile under grant Nos. 3190378 and 3200854, respectively.

CONFLICT OF INTEREST
The authors declare no potential conflict of interests.

DATA AVA I L A B I L I T Y STAT E M E N T


Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

ORCID
Naoufal El Youssfi https://orcid.org/0000-0003-0430-1743
Taha Zoulagh https://orcid.org/0000-0003-4796-2744
Hicham El Aiss https://orcid.org/0000-0003-1098-275X

REFERENCES
1. Perales MA, Kebriaei P, Kean LS, Sadelain M. Building a safer and faster CAR: seatbelts, airbags, and CRISPR. Biol Blood Marrow
Transplant. 2018;24(1):27-31.
2. Plessen MG, Bernardini D, Esen H, Bemporad A. Spatial-based predictive control and geometric corridor planning for adaptive
cruise control coupled with obstacle avoidance. IEEE Trans Control Syst Technol. 2018;26(1):38-50. https://dx.doi.org/10.1109/tcst.2017.
2664722
3. Li L, Li X, Wang X, Liu Y, Song J, Ran X. Transient switching control strategy from regenerative braking to anti-lock braking with a
semi-brake-by-wire system. Veh Syst Dyn. 2016;54(2):231-257.
4. Aksjonov A, Augsburg K, Vodovozov V. Design and simulation of the robust ABS and ESP fuzzy logic controller on the complex braking
maneuvers. Appl Sci. 2016;6(12):382.
5. Hartani K., Aouadj N., Merah A., Mankour M., Mohammed Chikouche T. Control of the Lateral Dynamics of Electric Vehicle Using
Active Security System. In: Hatti M, ed. Renewable Energy for Smart and Sustainable Cities. ICAIRES 2018. Lecture Notes in Networks
and Systems, vol 62. Springer: 2019.
6. Monot N, Moreau X, Benine-Neto A, Rizzo A, Aioun F. Comparison of observers for vehicle yaw rate estimation. Proceedings of the 2018
IEEE Intelligent Vehicles Symposium (IV); 2018:1925-1930; IEEE.
7. Van Dong N, Thai PQ, Duc PM, Thuan NV. Estimation of vehicle dynamics states using Luenberger observer. Int J Mech Eng Robot Res.
2019;8(3):430-436. https://dx.doi.org/10.18178/ijmerr.8.3.430-436
EL YOUSSFI et al. 367

8. Katzfuss M, Stroud JR, Wikle CK. Understanding the ensemble Kalman filter. Am Stat. 2016;70(4):350-357. https://dx.doi.org/10.1080/
00031305.2016.1141709
9. Reina G, Messina A. Vehicle dynamics estimation via augmented extended Kalman filtering. Measurement. 2019;133:383-395.
10. Liu M, Zhang L, Shi P, Zhao Y. Fault estimation sliding-mode observer with digital communication constraints. IEEE Trans Automat
Contr. 2018;63(10):3434-3441. https://dx.doi.org/10.1109/tac.2018.2794826
11. El Youssfi N, Oudghiri M, Aitouche A & El Bachtiri R Fuzzy sliding-mode observer for lateral dynamics of vehicles with consid-
eration of roll motion. Proceedings of the 2018 26th Mediterranean Conference on Control and Automation (MED); 2018:861-866;
IEEE.
12. Du D, Cocquempot V, Jiang B. Robust fault estimation observer design for switched systems with unknown input. Appl Math Comput.
2019;348:70-83. https://dx.doi.org/10.1016/j.amc.2018.11.034
13. Vu VP, Do TD. Fault/state estimation observer synthesis for uncertain TS fuzzy systems. IEEE Access. 2019;7:358-369. https://dx.doi.org/
10.1109/access.2018.2885379
14. Zhang K, Jiang B, Cocquempot V. Fuzzy unknown input observer-based robust fault estimation design for discrete-time fuzzy systems.
Signal Process. 2016;128:40-47. https://dx.doi.org/10.1016/j.sigpro.2016.03.015
15. Yeh SJ, Chang W, Wang WJ. Unknown input based observer synthesis for uncertain Takagi–Sugeno fuzzy systems. IET Control Theory
Appl. 2015;9(5):729-735. https://dx.doi.org/10.1049/iet-cta.2014.0705
16. Zhang H, Han J, Wang Y, Liu X. Sensor fault estimation of switched fuzzy systems with unknown input. IEEE Trans Fuzzy Syst.
2017;26(3):1114-1124. https://dx.doi.org/10.1109/tfuzz.2017.2704543
17. Gómez-Peñate S, Valencia-Palomo G, López-Estrada FR, Astorga-Zaragoza CM, Osornio-Rios RA, Santos-Ruiz I. Sensor fault diagnosis
based on a sliding mode and unknown input observer for Takagi-Sugeno systems with uncertain premise variables. Asian J Control.
2019;21(1):339-353. https://dx.doi.org/10.1002/asjc.1913
18. Louzimi A, El Assoudi A, Soulami J, El Yaagoubi EH. Unknown input observer design for a class of nonlinear descriptor systems: a
Takagi-Sugeno approach with Lipschitz constraints. Nonlinear Anal Differ Equ. 2017;5(3):99-116. https://dx.doi.org/10.12988/nade.2017.
61190
19. Hou Y, Zhu F, Zhao X, Guo S. Observer design and unknown input reconstruction for a class of switched descriptor systems. IEEE Trans
Syst Man Cybern Syst. 2018;48(8):1411-1419. https://dx.doi.org/10.1109/tsmc.2017.2679345
20. Nguyen AT, Guerra TM, Sentouh C, Zhang H. Unknown input observers for simultaneous estimation of vehicle dynamics and driver
torque: theoretical design and hardware experiments. IEEE/ASME Trans Mechatron. 2019;24(6):2508-2518. https://dx.doi.org/10.1109/
tmech.2019.2933744
21. Mammar S, Glaser S, Netto M. Vehicle lateral dynamics estimation using unknown input proportional-integral observers. Proceedings of
the 2006 American Control Conference; 2006:6; IEEE.
22. Hashemi E, Zarringhalam R, Khajepour A, Melek W, Kasaiezadeh A, Chen SK. Real-time estimation of the road bank and grade angles
with unknown input observers. Veh Syst Dyn. 2017;55(5):648-667.
23. Chen W, Xiao H, Wang Q, Zhao L, Zhu M. Integrated Vehicle Dynamics and Control. John Wiley & Sons; 2016:135-181.
24. Rajamani R. Vehicle Dynamics and Control. Mechanical Engineering Series 2nd. Springer Science & Business Media; 2011.
25. Polack P, Altché F, d’Andréa-Novel B, de La Fortelle A. The kinematic bicycle model: a consistent model for planning fea-
sible trajectories for autonomous vehicles? Proceedings of the 2017 IEEE Intelligent Vehicles Symposium (IV); 2017:812-818;
IEEE.
26. Bergsten P, Palm R, Driankov D. Observers for Takagi-Sugeno fuzzy systems. IEEE Trans Syst Man Cybern B Cybern. 2002;32(1):
114-121.
27. Jiang B, Gao Z, Shi P, Xu Y. Adaptive fault-tolerant tracking control of near-space vehicle using Takagi–Sugeno fuzzy models. IEEE Trans
Fuzzy Syst. 2010;18(5):1000-1007.
28. Lilly JH. Fuzzy Control and Identification. John Wiley & Sons; 2011:88-105.
29. de Oliveira MC, Skelton RE. Stability tests for constrained linear systems. Perspectives in Robust Control. Springer; 2001:
241-257.
30. Ohta S. Finsler interpolation inequalities. Calc Var. 2009;36(2):211-249.
31. Schramm D, Hiller M, Bardini R. Model of a typical complex complete vehicle. Vehicle Dynamics. Springer; 2018:351-371.
32. Doumiati M, Charara A, Victorino A, Lechner D. Vehicle Dynamics Estimation Using Kalman Filtering: Experimental Validation. John
Wiley & Sons; 2012.
33. El Youssfi N, Oudghiri M, El Bachtiri R. Vehicle lateral dynamics estimation using unknown input observer. Proc Comput Sci.
2019;148:502-511. https://dx.doi.org/10.1016/j.procs.2019.01.063
34. Ramirez Mendoza RA Sur la modélisation et la commande des véhicules automobiles. PhD thesis. Grenoble INPG; 1997.
35. El Youssfi N, El Bachtiri R, Chaibi R, Tissir EH. Static output-feedback H∞ control for T-S fuzzy vehicle lateral dynamics. SN Appl Sci.
2020;2(1):101. https://dx.doi.org/10.1007/s42452-019-1897-y
36. Aouaouda S, Bouarar T, Bouhali O. Fault tolerant tracking control using unmeasurable premise variables for vehicle dynamics subject to
time varying faults. J Franklin Inst. 2014;351(9):4514-4537. https://dx.doi.org/10.1016/j.jfranklin.2014.05.012
37. Tanaka K, Hori T, Wang HO. A multiple Lyapunov function approach to stabilization of fuzzy control systems. IEEE Trans Fuzzy Syst.
2003;11(4):582-589. https://dx.doi.org/10.1109/tfuzz.2003.814861
368 EL YOUSSFI et al.

38. Bærentzen JA, Gravesen J, Anton F, Aanæs H. Guide to Computational Geometry Processing: Foundations, Algorithms, and Methods.
Springer Science & Business Media.; 2012.
39. Agulhari CM, Felipe A, Oliveira RC, Peres PL. Algorithm 998: the robust LMI parser–a toolbox to construct LMI conditions for uncertain
systems. ACM Trans Math Softw (TOMS). 2019;45(3):1-25. https://dx.doi.org/10.1145/3323925

How to cite this article: El Youssfi N, El Bachtiri R, Zoulagh T, El Aiss H. Unknown input observer design for
vehicle lateral dynamics described by Takagi–Sugeno fuzzy systems. Optim Control Appl Meth. 2022;43(2):354-368.
doi: 10.1002/oca.2808

You might also like