Vehicle Parameter and States Estimation Via Sliding Mode Observers

Vehicle Parameter and States Estimation Via Sliding Mode Observers Hassan Shraim 1 , Mustapha Ouladsine 1 , and Leonid Fridman 2 1 Laboratory of sciences of informations and of systems, SIS UMR 6168 University of Paul C´ezanne, Aix-Marseille III, Av escadrille de Normandie Niemen 13397 Marseille Cedex 20 {hassan.shraim, mustapha.Ouladsine}@lsis.org 2 Department of Control, Engineering Faculty, National Autonomous University of Mexico (UNAM) lfridman@servidor.unam.mx 1 Introduction Preliminaries and motivations. In the recent years, important research has been undertaken to investigate safe driving conditions in both normal and critical situations. But due to the fact that safe driving requires the driver to react ex- tremely quickly in dangerous situations, which is generally very difficult unless for experts, then this reaction may influence the stability of the system. Con- sequently, the improvement of the vehicle dynamics by active chassis control is necessary for such catastrophic situations. Increasingly, commercial vehicles are being fitted with micro processor based systems to enhance the safety and to improve driving comfort, increase traffic circulation, and reduce environmental pollution associated with vehicles. Considerable attention has been given to the development of the control sys- tems over the past few years, authors have investigated and developed different methods and different strategies for enhancing the stability and the handling of the vehicle such as, the design of the active automatic steering [30] and the wheel ABS control [16], [25], or the concept of a four-wheel steering system (4WS) which has been introduced to enhance vehicle handling. Some researchers have shown disadvantages on (4WS) vehicles [15]. In terms of vehicle safety, and in order to develop a control law for the vehicle chassis, accurate and precise tools such as sensors should be implemented on the vehicle, to give a correct image of its comportment. Difficulties in measuring all vehicle states and forces, due to high costs of some sensors, or the non existence of others, make the design and the construction of observers necessary. In the vehicle dynamics, the estimations of vehicle side slip angle and wheel interaction forces with the ground are very important, because of their influence on the stability of the vehicle. Many researchers have studied and estimated vehicle side slip angle, using a bicycle model as in [23], or by using an observer with adaptation of a quality function as in [29] which requires a certain linearised form G. Bartolini et al. (Eds.): Modern Sliding Mode Control Theory, LNCIS 375, pp. 345–362, 2008. springerlink.com c  Springer-Verlag Berlin Heidelberg 2008 346 H. Shraim, M. Ouladsine, and L. Fridman of the model. Moreover, an extended Kalman filter is used for the estimation of wheel forces [19]. Methodology. In this chapter sliding mode observers are used for estimation of the vehicle parameters. These observers are developed and used (see, for example, the corresponding chapter in the textbooks by [5] and [26] and the recent tutorials by [4] and [17]. They are widely used due to their attractive features: • insensitivity (more than robustness) with respect to unknown inputs; • possibilities to use the values of the equivalent output injection for the un- known inputs identification ([26]); • finite time convergence to the reduced order manifold. In [12], [14] robust exact differentiators were designed ensuring finite time convergence to the real values of derivatives, as an application of super-twisting algorithm ([13]). New generation of observers based on the high order sliding mode differentiators are recently developed (see [3], [1], [2], [7] and [8]). Those observers: • provide a convergence to the exact values of states variables; • allow the exact identification the unknown inputs without filtration; • guarantee the best possible accuracy of the state estimation w.r.t. to the sampling steps and deterministic noises. The results of the chapter are validated by the simulator ve-dyna (see [24]) developed by the group of companies TESIS, which is an independent expert team for the simulation of virtual vehicles in real time. Their software comprises higher precision models for simulation of vehicle dynamics ve-dyna,enginedy- namics en-dyna and brake hydraulics (RT Brake Hydraulics). The simulator ve-dyna that we use in our study is a software especially designed for the fast simulation of vehicle dynamics both in real-time applications (Hardware- in-the-Loop, Software-in-the-Loop) and concept studies on a standard desktop PC, its computational performance enables its usage for the optimization e.g. for the parameter identification. ve-dyna vehicle model is fully parametric and has a modular architecture with the following program modules: Vehicle multi- body system (chassis), axles (axle kinematics, compliance), tire model TM-Easy, transmission and drive line, engine, aerodynamics and braking system. All ve- hicle degrees of freedom (DoF) are nonlinearly modeled. The number of overall DoF depends on the number of additional bodies and the suspension type. The global vehicle model and each sub model are validated by real experiments with different operation conditions. Contribution. The main contributions of this work reside in the estimations of wheels contact forces with the ground, side slip angle and the velocity of the vehicle. By these estimations, we avoid the use of expensive sensors and we can preview some critical situations that may occur while driving, such as exces- sive rotation around Z axis and also excessive side slipping, inappropriate lateral Vehicle Parameter and States Estimation Via Sliding Mode Observers 347 acceleration, The proposed observers are characterized by the fast convergence to real values, robustness and they don’t require extensive computation load. As we have mentioned before, the proposed observers are validated by the simulator ve-dyna. Several validations were made to cover most of the driving cases, such as a double lane trajectory, straight line motion with strong variation in acceleration and deceleration, strong change in the steering angle. Structure of the chapter. This chapter is organized as follows: In section 2, prob- lem statement is proposed and the model used for the vehicle is shown. In sec- tion 3, two steps are presented. In the first one, we present the estimation of the longitudinal forces and wheels angular velocities. In this step, comparison of two high order sliding mode observers is made and the third order one is chosen for the estimation. In the second step, a sliding mode observer based on the super-twisting algorithm for the estimation of the side slip angle and the velocity of the vehicle is proposed. Finally a conclusion for the work is presented. 2 Problem Statement The presented problem is how to assist the driver in critical situations, i.e., when the vehicle goes out from its safety regions and enters in the dangerous situations. So accurate tools to represent vehicle states and parameters are required. These accurate representations need many precise and expensive sensors. The use of many sensors requires an important diagnosis system to avoid false data. To overcome these problems, robust virtual sensors are proposed which estimate vehicle parameters, forces and states. The model used for estimation and control is a non-linear one obtained by applying the fundamental principles of dynamics at the center of gravity on figure 1 (see [29]): ˙v COG = 1 M  cos(β)  F L − sin(β)  F S  (1) ˙ β = 1 Mv COG  cos(β)  F S − sin(β)  F L  − ˙ ψ (2) with  F L = F xwind +cos(δ f )(Fx 1 +Fx 2 )+cos(δ r )(Fx 3 +Fx 4 )−sin(δ f )(Fy 1 + Fy 2 ) − sin(δ r )(Fy 3 + Fy 4 ), and  F S = sin(δ f )(Fx 1 + Fx 2 )+sin(δ r )(Fx 3 + Fx 4 )+cos(δ f )(Fy 1 + Fy 2 )+ cos(δ r )(Fy 3 + Fy 4 ). I Z ¨ ψ = t f {cos(δ f )(Fx 2 − Fx 1 )+sin(δ f )(Fy 1 − Fy 2 )} + L 1 {sin(δ f ) (Fx 2 + Fx 1 )+cos(δ f )(Fy 1 + Fy 2 )} + L 2 {sin(δ r )(Fx 3 + Fx 4 )− cos(δ r )(Fy 3 + Fy 4 )} + t r {cos(δ r )(Fx 4 − Fx 3 )+sin(δ r )(Fy 3 − Fy 4 )} (3) 348 H. Shraim, M. Ouladsine, and L. Fridman Table 1. Nomenclature Symbol Physical signification Ω i angular velocity of the wheel i M total mass of the vehicle r i radius of the wheel i COG centre of gravity of the vehicle r 1i dynamical radius of the wheel i Fz i vertical force at wheel i Fx i longitudinal force applied at the wheel i Fy i lateral force applied at the wheel i C fi braking torque applied at wheel i C mi motor torque applied at wheel i torque i C mi − C fi I Z moment of inertia around the Z axis ψ yaw angle ˙ ψ yaw velocity δ f front steering angle δ r rear steering angle δ i deflection in the tire i Vx longitudinal velocity of the center of gravity Vy lateral velocity of the center of gravity I ri moment of inertia of the wheel i v COG total velocity of the center of gravity L 1 distance between COG and the front axis L 2 distance between COG and the rear axis LL 1 + L 2 h COG height of COG t f front half gauge t r rear half gauge lt f + t r F xwind air resistance in the longitudinal direction F ywind air resistance in the lateral direction A L front vehicle Area ρ air density C aer coefficient of aerodynamic drag α i slip angle at the wheel i β side slip angle at the COG μ i friction coefficient at the wheel i X t length of the contact patch for the wheel i X ad,i length of the adhesion patch for the i-th wheel X sl,i length of the sliding patch for the i-th wheel Vx i Longitudinal velocity of the wheel i p i inflation pressure of the tire i K 1i constant depending on the deformation of the tire K 2i constant depending on the deformation of the tire Vehicle Parameter and States Estimation Via Sliding Mode Observers 349 1 L 2 L 1x F 2x F 3x F 4x F 3y F 1y F 2y F 2z F 1z F 3z F 4z F OG C r t×2 f t×2 ψ o x o y 1 αδ + f 2 αδ + f 4y F Fig. 1. 2D vehicle representation The model representing the dynamics of each wheel i is found by applying New- ton’s law to the wheel and vehicle dynamics figure 2: Fig. 2. Wheel and its contact with the ground I ri ˙ Ω i = −r 1i Fx i + torque i i =1:4 (4) In this paper, the task is to design a virtual sensors (observer) for the vehicle to estimate the states, parameters and forces which need expensive sensors for their measurement. But due to the fact that it is not easy to apply an observer for the global model, equations 4 are taken at first, a high order sliding mode 350 H. Shraim, M. Ouladsine, and L. Fridman observer is proposed for each equation 4 to estimate the longitudinal force of each wheel. After having the longitudinal forces, we substitute their values in equations 1, 2 and 3. From these equations, it is seen that if we substitute the estimated longitudinal forces, we will still have as complex terms the lateral forces. In order to model the lateral force, we use a brush model for the contact with the ground [21]. The brush model divides the surface of contact into two parts: a sliding part and an adhesion part. Then the lateral force generated at the surface of contact will be the sum of forces generated at each part of the surface. These lateral forces are represented as in [21]: Fy i = tan(α i )μ i Fz i X sl,i   tan(α i ) 2 +  Ω i Vx i  K 1i + K 2i p i  × Fz i 1000  − 1  2  × X ti + A i 3 × X 3 ad,i (5) with A i = 1 B 2K i tan(α i ) − Fy sl,i × ϕ X ad,i X sl,i (6) and B = C + D (7) C = 1 β 1 (1 + exp (−β 1 X ad,i )(sin(β 1 X ad,i )) −cos(β 1 X ad,i )) (8) D = X (1 + exp (−β 1 X ad,i ) cos(β 1 X ad,i )) (9) with β 1 = 4  K i E ×I ri (10) and ϕ =1+F + G + H (11) F = −exp (−β 1 X sl,i ) cos(β 1 X sl,i ) (12) G = −exp (−β 1 X ad,i ) cos(β 1 X ad,i ) (13) H = exp (−β 1 X ti ) cos(β 1 X ti ) (14) Fy sl,i = tan(α i )μ i Fz i X sl,i   tan(α i ) 2 +  Ω i Vx i  K 1 + K 2 p i  × Fz i 1000  − 1  2  × X ti (15) From this representation of the force, we need to define some of the variables, all the equations for these variables are detailed in [22] and [28], and they are given by: Vehicle Parameter and States Estimation Via Sliding Mode Observers 351 • The coefficient of adherence μ i : μ i = Fx i Fz i (16) • The side slip angle of each wheel: α 1 = tan −1  v COG sin(β)+L 1 ˙ ψ v COG cos(β) −t f ˙ ψ  − δ f (17) α 2 = tan −1  v COG sin(β)+L 1 ˙ ψ v COG cos(β)+t f ˙ ψ  − δ f (18) α 3 = tan −1  v COG sin(β) −L 2 ˙ ψ v COG cos(β) − t r ˙ ψ  − δ r (19) α 4 = tan −1  v COG sin(β) −L 2 ˙ ψ v COG cos(β)+t r ˙ ψ  − δ r (20) • The velocity of each wheel: Vx 1 =  v COG cos(β) −t f ˙ ψ  cos(δ f )+  v COG sin(β)+L 1 ˙ ψ  sin(δ f ) (21) Vx 2 =  v COG cos(β)+t f ˙ ψ  cos(δ f )+  v COG sin(β)+L 1 ˙ ψ  sin(δ f ) (22) Vx 3 =  v COG cos(β) −t r ˙ ψ  cos(δ r )+  v COG sin(β) −L 2 ˙ ψ  sin(δ r ) (23) Vx 4 =  v COG cos(β)+t r ˙ ψ  cos(δ r )+  v COG sin(β) −L 2 ˙ ψ  sin(δ r ) (24) For the vertical forces, cheap sensors can be found for their measure (or they can be estimated as shown in [28]), the determination of the contact patch and its repartition into a sliding part and adhesion part can be found as shown in [22]. The system described by the equations 1, 2 and 3 is observable if we consider that we measure only the yaw rate (and by supposing the longitudinal forces as inputs). A sliding mode observer based on hierarchical super twisting algorithm is used to estimate vehicle side slip angle and velocity. By these estimations, the longitudinal and lateral velocities of the center of gravity, the lateral forces of the wheels are then directly deduced. By these estimations, the driver (or the controller) knows if the vehicle is in the safe region or not. These regions depend on the velocity, the coefficient of friction and the steering angle [28] and [9]. In this study it is supposed that we have as inputs (estimated or measured): • angular positions of the wheels (used if angular positions sensors are imple- mented on the vehicle) 352 H. Shraim, M. Ouladsine, and L. Fridman • front wheel angle (measured) • yaw rate (measured) • torque applied at each wheel (estimated see [11] and [18]) And it is required to estimate: • angular velocity of the wheels (to be observed only if it is not measured) • contact forces • vehicle velocity • vehicle side slip angle If the wheels angular velocities are measured instead of wheels angular posi- tions, then in the first step we have to estimate only the longitudinal forces. For this goal, a sliding mode observer based on the super twisting algorithm may be used. 3 Observer Design 3.1 Estimation of Wheels Angular Velocities and Longitudinal Forces In this part, sliding mode observers are proposed to observe the angular velocity Ω i and to identify the longitudinal force of each wheel Fx i . Case when the angular positions of the wheels are measured Dynamical equations of wheels (4) are written in the following form: ˙x 1 = x 2 ˙x 2 = f(x 1 ,x 2 ,u) (25) where x 1 and x 2 are respectively θ i (which is measured) and Ω i (to be observed) (appears implicitly in Fx i ), and u is the torque i . In fact this torque may be measured as shown in [18] and it can also be estimated by estimating the motor and the braking torque, the motor torque may be estimated as in [11], while the braking torque is estimated by measuring the hydraulic pressure applied at each wheel (existing on the most of the vehicles). Broken super-twisting observer In the first part of this section, the so-called broken super-twisting algorithm observer proposed in [3] in the form: ˙ ˆx 1 =ˆx 2 + z 1 ˙ ˆx 2 = f 1 (x 1 , ˆx 2 ,u)+z 2 (26) where ˆx 1 and ˆx 2 are the state estimations of the angular positions and the angu- lar velocities of the four wheels respectively, f 1 is a nonlinear function containing only the known terms (which is only the torque in our case), z 1 and z 2 are the Vehicle Parameter and States Estimation Via Sliding Mode Observers 353 correction factors based on the super twisting algorithm having the following forms: z 1 = λ|x 1 − ˆx 1 | 1/2 sign(x 1 − ˆx 1 ) z 2 = λ 0 sign(x 1 − ˆx 1 ). (27) In the above equations the function |.| 1 2 and sign(.) should be thought of as componentwise extensions of their traditional scalar coordinates. We will define the solutions of (26)-(27) and all equations below in the Filippov sense [6]. Taking ˜x 1 = x 1 − ˆx 1 and ˜x 2 = x 2 − ˆx 2 we obtain the equations for the error ˙ ˜x 1 =˜x 2 − λ|˜x 1 | 1/2 sign(˜x 1 ) ˙ ˜x 2 = F (t, x 1 ,x 2 , ˆx 2 ) −λ 0 sign(˜x 1 ) (28) where F (t, x 1 ,x 2 , ˆx 2 )=f(x 1 ,x 2 ,u) − f 1 (x 1 , ˆx 2 ,u) is the unknown function to be identified. Suppose that the system states can be assumed bounded, then the exists a such a constant f + that the inequality: |F (t, x 1 ,x 2 , ˆx 2 )| <f + (29) holds for any possible t, x 1 , x 2 and |ˆx 2 |≤2sup|x 2 |. As described in [3], it is sufficient to choose λ 0 =1.1f + and λ =1.5  f + . A sliding motion occurs in the error system (27) which makes ˜x 1 ≡ 0and˜x 2 ≡ 0 in finite time. Furthermore we get: αsign(x 1 − ˆx 1 )=F(t, x 1 ,x 2 , ˆx 2 ) (30) where the left hand of the above equation represents the average value of the discontinous term which must be taken in order to maintain a sliding motion. The value of sign(x 1 − ˆx 1 ) can be obtained by appropriate low-pass filtering of the discontinous injection signal sign(x 1 − ˆx 1 ) and so is available in real time. So in the broken super twisting observer suggested here allowing to estimate of the longitudinal forces by filtering the discontinuous injection signal (which may cause a certain delay). Third order sliding mode Observer To avoid filtration let us consider s third order sliding mode observer in the form: ˙ ˆ θ i = ˆ Ω i + λ 0 |θ i − ˆ θ i | 2/3 sign(θ i − ˆ θ)    v o ˙ ˆ Ω i = torque i I ri + z 1 (31) where ˆ θ i and ˆ Ω i are the state estimations of the angular positions and the angular velocities of the four wheels respectively, z 1 is the correction factor based on the super twisting algorithm having the following forms: 354 H. Shraim, M. Ouladsine, and L. Fridman z 1 = λ 1 | ˆ Ω i − v o | 1/2 sign( ˆ Ω i − v o )+Z 1    v 1 ˙ Z 1 = λ 2 sign(Z 1 − v 1 ). (32) with λ 2 =3 3 √ λ 0 , λ 1 =1.5 √ λ 0 et λ 0 =2f + . The second-order sliding-mode occurs in (32) in finite time Levant (2003). In particular, this means that the longitudinal forces can be estimated in a finite time without the need for low pass filter of a discontinuos switched signal. Case when the angular velocities of the wheels are measured In this case, we need to estimate only the longitudinal forces. The proposed observer in this case is: ˙ ˆx 2i = torque i I ri + λ 1 |x 2i − ˆx 2i | 1/2 sign(x 2i − ˆx 2i )+Z 1i ˙ Z 1i = λ 2 sign(x 2i − ˆx 2i ). (33) The second-order sliding-mode occurs in system (33) in finite time [14]. 0 10 20 30 40 50 60 70 80 -400 -200 0 200 400 600 800 time(sec) Fig. 3. Motor and braking torque (N.m) applied at the two rear wheels 0 10 20 30 40 50 60 70 80 -700 -600 -500 -400 -300 -200 -100 0 time(sec) Fig. 4. Motor and braking torque (N.m) applied at the two front wheels [...]... (N.m) 700 600 500 400 300 200 100 0 -100 0 5 10 15 time (sec) Fig 12 Motor and braking torque (N.m) applied at the two rear wheels Vehicle Parameter and States Estimation Via Sliding Mode Observers 359 18 16 Observ er Simulator 14 velocity (m/sec) 12 10 8 6 4 2 0 -2 0 5 10 15 20 25 time (sec) Fig 13 Estimated vehicle velocity (m/sec) and that of the simulator ve-dyna 0.015 Simulator Observ er side slip... Conclusions Sliding mode observers are proposed in this work to estimate vehicle parameters and states which are not easily measured These observers have shown the fast time of convergence and robustness in the automotive applications that we have proposed The validation of the proposed observers is realized by comparing the observers output with the outputs of the simulator ve-dyna, reasonable and acceptable... between the longitudinal force (computed from the observer) and that given by ve-dyna 3.2 Estimation of the Vehicle Side Slip Angle and Velocity In this part, a sliding mode observer based on the super twisting algorithm is used to estimate the velocity and the side slip angle at the center of gravity The model of the vehicle is a non-linear model and it can be written as follows: x = f (x, u) = A(x) +... steer- Vehicle Parameter and States Estimation Via Sliding Mode Observers 357 500 Observ er Simulator 0 -500 -1000 -1500 -2000 -2500 -3000 0 10 20 30 40 t ime(s ec ) 50 60 70 80 Fig 9 The unknown input after filtration (N) and the longitudinal force from the simulator ve-dyna Observ er Simulator 400 200 0 -200 -400 -600 -800 5 10 15 t ime(s ec ) 20 25 Fig 10 The unknown input after filtration (N) and the... Control 17, 1734–1753 (2007) Vehicle Parameter and States Estimation Via Sliding Mode Observers 361 3 Davila, J., Fridman, L., Levant, A.: Second-order sliding mode observer for mechanical systems IEEE Transactions on Automatic Control 50, 1785–1789 (2005) 4 Edwards, C., Spurgeon, S.K., Hebden, R.G.: On development and applications of sliding mode observers In: Xu, J., Yu, X (eds.) Variable Structure Systems:... A sliding mode approach of unknown input observers for linear systems In: Proc 43th IEEE Conference on Decision and Control, Nassau, Bahamas (2004) 2 Bejarano, F.J., Fridman, L., Poznyak, A.: Exact State Estimation for Linear Systems with Unknown Inputs based on a hierarchical super twisting algorithm International Journal of Robust And Nonlinear Control 17, 1734–1753 (2007) Vehicle Parameter and States. . .Vehicle Parameter and States Estimation Via Sliding Mode Observers 355 6000 5000 4000 Simulator Obs erv er 3000 2000 1000 0 -1000 0 10 20 30 40 t ime(s ec ) 50 60 70 80 Fig 5 Angular position (rad) by the simulator ve-dyna, and that estimated by the observer 2 1.5 Simulator Obs erv er 1 0.5 0 -0.5 0 0.1 0.2 0.3... Y.: Estimation of the instantaneous torque of a gasoline engine (in French) In: Proc Int Conf Francophone Aut CIFA 2006, Bordeaux, France (2006) 12 Levant, A.: Robust exact differentiation via sliding mode technique Automatica 34, 379–384 (1998) 13 Levant, A.: Sliding order and sliding accuracy in sliding mode control International Journal of Control 58, 1247–1263 (1993) 14 Levant, A.: Higher order sliding. .. in Control and Information Science, vol 274 Springer, Berlin (2002) 5 Edwards, C., Spurgeon, S.K.: Sliding Mode Control Taylor and Francis, London (1998) 6 Filippov, A.F.: Differential equations with discontinuous right-hand sides Kluwer Academic Publishers, Dordrecht (1988) 7 Fridman, L., Levant, A., Davila, J.: Observation of linear systems with unknown inputs via higher order sliding modes International... (rad) using sliding modes and that of the simulator ve-dyna Results of simulation ˆ Once again, the estimated variables, vCOG and β are compared to that provided ˆ by ve-dyna The operation conditions are given by a variation in δ f figure 11 and torquei figure 12, which constitute a significant driving situation F xi are estimated from the third order sliding mode observer In figure 13, 14 and 15 ˆ ˆ ˙ . of Robust And Nonlinear Control 17, 1734–1753 (2007) Vehicle Parameter and States Estimation Via Sliding Mode Observers 361 3. Davila, J., Fridman, L., Levant, A.: Second-order sliding mode observer. rotation around Z axis and also excessive side slipping, inappropriate lateral Vehicle Parameter and States Estimation Via Sliding Mode Observers 347 acceleration, The proposed observers are characterized. equations for these variables are detailed in [22] and [28], and they are given by: Vehicle Parameter and States Estimation Via Sliding Mode Observers 351 • The coefficient of adherence μ i : μ i = Fx i Fz i (16) •