Proceedings DSCC2009 Proceedings of the ASME 2009 Dynamic Systems and ControlofConference 2009 ASME Dynamic Systems and Control Conference DSCC2009 12-14,Hollywood, 2009, Hollywood, CA, USA USA OctoberOctober 12-14, 2009, California,

DSCC2009-2737

NONLINEAR-FEEDBACK VEHICLE TRACTION FORCE CONTROL WITH LOAD TRANSFER

Annalisa Scacchioli∗ Panagiotis Tsiotras† Jianbo Lu‡ Department of Aerospace Engineering Department of Aerospace Engineering Research and Advanced Engineering Georgia Institute of Technology Georgia Institute of Technology Ford Motor Company Atlanta, Georgia,30309 Atlanta, Georgia, 30309 Dearborn, Michigan, 48124

ABSTRACT This article deals with the nonlinear feedback regulation of the longitudinal traction forces for high-speed vehicles, possibly over a low friction surface. Hybrid models of the longitudinal vehicle dynamics incorporating load transfer effects, a crucial element in advanced driving techniques, are derived. The designed hybrid regulator allows the tracking of a given friction force profile in the presence of known disturbances and unknown model uncertainties. Simulations show good performance of the proposed hybrid regulator under all operating conditions.

best of our knowledge, none of these works deals with expert driver-inspired vehicle active safety systems for vehicles in the limit of their handling conditions. In this article, following our previous works [7, 9, 10], we develop a hybrid model of a vehicle during “left foot braking” a specific driving technique used by rally drivers, which takes advantage of the load transfer effect from front to rear axles and vice versa, to control the friction forces at the front and rear wheels. Specifically, in this technique the driver applies braking and acceleration commands simultaneously in order to fine tune the total torque applied to the wheels. We describe the system’s behavior at two levels of abstraction (the wheels and the vehicle) through a finite number of discrete states and state transitions. We use a family of continuous models incorporating the load transfer effect, within the individual discrete states. Continuous control laws, based on the nonlinear output regulation theory and tracking a given friction force profile, are designed at the highest abstraction level, but implemented at the more accurate, lower wheel abstraction level.

INTRODUCTION Accident avoidance is the primary task of automotive active safety systems. During an imminent collision, when the driver’s reaction time is rather limited, the full utilization of the vehicle’s handling capability becomes critical. Only expert drivers are masters of controlling the vehicle at these extreme regimes. Incorporating expert driving skills in an active safety system is an important step forward in the technology of future accident avoidance systems. By fully utilizing the vehicle’s handling capability one should be able to stop the vehicle faster or to move the vehicle away from a hazard, thus improving vehicle response to levels that are not possible by novice drivers. Several contributions have been developed in the last years to improve vehicle control stability. Anti-lock braking systems (ABS) and traction control or acceleration slip regulation (ASR), are examples of the current state of technology. However, to the

PROBLEM FORMULATION Vehicle Model We consider a vehicle in a straight driving condition as shown in Figure 1. We assume that the vehicle lateral load transfer is zero and the road surface friction is evenly distributed on the right and left wheels. The four wheel vehicle can, then, be modeled through a two-wheel bicycle model as shown in Figure 1. Let ξ = (v, ωF , ωR )T be the state of the vehicle, where v is the translational velocity (measured along the longitudinal direction of the vehicle) of the center of mass, with ωF and ωR the

∗ Email

address: [email protected]. address: [email protected]. ‡ Email address: [email protected]. † Email

1

Copyright © 2009 by ASME

L r fR Figure 1.

assume that the friction coefficients depend on the slip ratios as follows

lF

lR CoG

v mg

x

  µ∗ (λ∗ ) = D∗ sin C∗ arctan(B∗ λ∗ ) ,

fd TF

TR h fF

NR

∗ = F, R,

(4)

where B∗ , C∗ , and D∗ are parameters fitting experimental data. For notational convenience, the argument in µF and µR will be dropped in the sequel. The normal loads NF and NR satisfy the following relations

NF

LONGITUDINAL FORCE DISTRIBUTION ON A TWO-AXLE

VEHICLE ON A LEVEL ROAD.

NF = mg angular velocities of the front and rear wheels, respectively. The equations of motion of the vehicle are mv˙ = fF + fR − fd , ˙ F = TF − r fF , IF ω ˙ R = TR − r fR , IR ω

(1c)

where fF [N] and fR [N] are the tractive friction forces acting on the front and rear wheels, respectively, fd [N] is the external disturbance, m [kg] is the mass of the vehicle body, IF [kgm2] and IR [kgm2 ] denote the rotational moments of inertia of the front and rear wheels, respectively, and r [m] is the rolling radius of the wheels. Furthermore, TF [Nm] is the total torque (engine drive torque and brake torque) acting on each of the front wheels, while TR [Nm] is the total torque acting on each of the rear wheels. Without loss of generality, we assume a FWD (Front Wheel Drive) vehicle, for which a possible distribution of the driving torque Tshaft [Nm], produced by the engine, and the brake torque Tbrake [Nm], produced by the braking system, on the front and rear wheels is given by [5]: (2a)

TR = −0.2Tbrake .

(2b)

fR = µR (λR )NR ,

NR = mg

lF

h + mv, ˙ L L

NF + NR = mg,

Hybrid Vehicle Model with Load Transfer In this section, we use the formalism of hybrid systems [8] to model the longitudinal vehicle dynamics of a FWD vehicle, including the effect of longitudinal load transfer. We propose two different abstraction levels for the hybrid system modeling a FWD vehicle: the Vehicle-level Hybrid System (VHS) and the Wheel-level Hybrid System (WHS), as depicted in Figure 2. The difference between the two levels stems from the effects of the time delay between the driver’s command (accelerate or brake) and the reactions of the wheels in the model description. For convenience of notation, in Figure 2 we denote with the superscript V the elements related to the VHS model abstraction and with superscript W the elements related to the WHS model abstraction. We will use the VHS model for designing the controller and the WHS for evaluating the efficiency of the designed hybrid controller. Both the abstraction levels are described using the following definitions of the wheels’ and vehicle’s operating conditions. Definition 1: Wheel Modes. Given the wheels’ slip ratios for a FWD vehicle [2]:

The disturbance due to the rolling wheel resistance is given by fd = croll mg, where croll is the rolling resistance coefficient and g [m/s2 ] is the gravitational acceleration. Following [2, 6], the longitudinal forces acting on the front and rear tires are given by fF = µF (λF )NF ,

h − mv, ˙ L L

(5) with h [m] the vertical distance to the vehicle’s center of mass, lF [m] the distance from the front axle to the vehicle’s center of mass, lR [m] the distance from the rear axle to the vehicle’s center of mass and L the distance from the front to the rear axle, given by L = lF + lR . As described in the next section, the vehicle has an intrinsic hybrid nature that leads to a hybrid model abstraction.

(1a) (1b)

TF = 0.5Tshaft − 0.3Tbrake ,

lR

λdF =

(3)

rωdF − vd rωdF

λdR = 0 where µF (λF ) and µR (λR ) are the friction coefficients, while NF and NR are the normal loads on the front and rear wheels, respectively. Typically, µF (λF ) and µR (λR ) are nonlinear functions, depending on the slip ratios of the front and rear wheels λF and λR , respectively. Using the Bakker-Pacjeka model [1], we may

λbF = λbR =

2

for vd < rωdF , for vd = rωdR ,

rωbF

−v b

(6b)

b

v b rωR − vb vb

(6a)

for vb > rωbF ,

(6c)

for vb > rωbR ,

(6d)

Copyright © 2009 by ASME

ing/braking mode (λdF ≥ 0 and λbR < 0):

we say that the front (rear) wheel is in wheel driving mode if λdF ≥ 0 (λdR = 0) and it is in wheel braking mode if λbF < 0 (λbR < 0), where the superscribed index d denotes the driving mode and the superscribed index b denotes the braking mode for the corresponding wheel. Definition 2: Vehicle Modes. Given the total force acting between the vehicle’s tires and the road as ftot = fF + fR − fd , we say that a vehicle is in vehicle driving mode if ftot ≥ 0 and in vehicle braking mode if ftot < 0, where the forces ( fFd , fRd ) for the vehicle driving mode and ( fFb , fRb ) for the vehicle braking mode, given, respectively, by

 fFd = µdF mg

lR L + hµdF

+

h

fd d d L + hµF

fRd = 0,



,

   k0 fFd + fRb − fdd    d   d d d d  −k fF + fR − fd γd − k fF ηd + k TF ηd  W f2 (x, u, d) =  0 1 d F 2 d F F d v v v    b    fF + fRb − fdb b TRb fRb −k0 + k γ − k 4 3 R vb vb vb (10) b < 0) with wheels in wheel driv3. Vehicle braking mode ( ftot d ing/braking mode (λF ≥ 0 and λbR < 0): 

   k0 fFd + fRb − fdb    d   d d d d  −k fF + fR − fd γd − k fF ηd + k TF ηd  W f3 (x, u, d) =  0 1 d F 2 d F F d v v v    b    fF + fRb − fdb b fRb TRb −k0 − k γ + k 3 4 R vb vb vb (11) b < 0) with wheels in wheel brak4. Vehicle braking mode ( ftot b ing/braking mode (λF < 0 and λbR < 0): 

(7a) (7b)

 lR − hµbR h + fdb , L + h(µbF − µbR) L + h(µbF − µbR)   h lF + hµbF µbR mg − fdb , L + h(µbF − µbR) L + h(µbF − µbR)

 fFb = µbF mg

(7c)

fRb =

(7d)

   k0 fFb + fRb − fdb    b  b b b  b  −k fF + fR − fd γb − k fF + k TF  W f4 (x, u, d) =  0 2 b  1 b F vb v v    b   b b b fF + fR − fd b fR TRb − k γ + k −k0 3 4 R vb vb vb (12) b < 0) with wheels in wheel brak5. Vehicle braking mode ( ftot ing/driving mode (λbF < 0 and λdR = 0): 

where fdd = fdb = fd and   µdF = DF sin CF arctan(BF λdF ) ,

for 0 ≤ λdF ≤ 1,

(8a)

µdR = 0,

for λdR = 0, (8b)   µbF = DF sin CF arctan(BF λbF ) , for − 1 ≤ λbF < 0,(8c)   µbR = DR sin CR arctan(BR λbR ) , for − 1 ≤ λbR < 0. (8d)

   k0 fFb + fRd − fdb   f5W (x, u, d) =    b fF + fRb − fdb b fFb TFb  γF − k1 b + k2 b −k0 vb v v (13) 

Deriving and rearranging terms in (6) and using (1), and after denoting with x = (v, λF , λR )T the state of the vehicle in slip ratio variables, one obtains five dynamic systems, given by x˙ = fiW (x, u, d) (i = 1, . . . , 5)), which model the longitudinal vehicle behavior at the wheel level, as follows:

where k0 = 1/m, k1 = r2 /IF , k2 = r/IF , k3 = r2 /IR , k4 = r/IR and ηdF = (1 − λdF )2 , γdF = (1 − λdF ), γbF = (1 + λbF ), and γbR = (1 + λbR ). At this level, we assumed that both engine and brake torques do not have an immediate effect on the wheels, but there is a time delay between the driver’s command (accelerating or brakd ≥ 0) with wheels in wheel driv1. Vehicle driving mode ( ftot ing) and the effective reactions on the wheels. Using the notation ing/driving mode (λdF ≥ 0 and λdR = 0): described in [8], each of these five models corresponds to a disW W W W crete state qW ∈ QW = {qW 1 , q2 , q3 , q4 , q5 } of a finite state  automaton as depicted at the bottom in Figure 2: two for the    W d d d vehicle driving mode (qW k0 f F + f R − f d 1 and q2 ) referring to a total friction   force being nonnegative, and three for the vehicle braking mode   d f1W (x, u, d) =  f + fRd − fdd d fFd d TFd d  W and qW and qW ), where the total friction force is negative. γ − k η + k η −k0 F (q 1 d F 2 d F F 3 4 5 vd v v The state qW corresponds to (front and rear) wheels with friction (9) 1 d ≥ 0) with wheels in wheel drivslip ratios nonnegative, and qW 2. Vehicle driving mode ( ftot 4 corresponds to (front and rear)

3

Copyright © 2009 by ASME

wheels with friction slip ratios negative. The rest of the states W W (qW 2 , q3 , q5 ) describes mixed operating conditions, in which front and rear wheels have slip ratios with different signs. The transitions among these five models are regulated by the guard W W W W conditions GW 12 , G23 , G34 , G45 , G51 , which are forced when the invariant conditions are violated. After a transition, the state x W W W can be reset through the reset functions (RW 12 , R23 , R34 and R45 , W R51 ) described in Figure 2. At the VHS level, we assume that the time delay between the driver’s command to accelerate (brake) and the reactions of

GV 12

x˙ = f1V (x, u, d)

qV 1

x ∈ IV 1

RV 12

RV 21

GV 21

x˙ = f2V (x, u, d) x ∈ I2V

the wheels (front and rear) is zero. This allows a description at a higher level of abstraction of the vehicle through two models, V each corresponding to a state qV ∈ QV = {qV 1 , q2 } of a finite state automaton as depicted at the top of Figure 2: one for the vehicle driving mode (qV 1 ) and one for the vehicle braking mode (qV ). The transitions among these models, depending on the 2 value of the system state described by (9) and (12), are regulated V by the guard conditions (GV 12 , G21 ), stating when a transition can take place. They are forced when the invariant conditions (x ∈ V I1V , x ∈ I2V ) are violated. The reset functions are RV 12 , R21 . The VHS and the WHS models are connected through the following W V W V relations qV 1 ≡ q1 and q2 ≡ q4 and consequently f 1 (x, u, d) = W V W f1 (x, u, d) and f2 (x, u, d) = f4 (x, u, d).

qV 2

Control Problem Formulation In this section we formulate the control problem for the vehicle-level hybrid system ((9), (12)) to determine a hybrid controller such that the vehicle accelerates or brakes following ref (t), while the (known) a prescribed reference friction force ftot disturbances are rejected. This problem can be solved using the nonlinear output regulation theory [4], where the reference and the disturbance are produced by an autonomous dynamic system, called the “exosystem,” given by

I1V = {λF ≥ 0, λR = 0, ftot ≥ 0, u = ud , d = fdd }, I2V = {λF < 0, λR < 0, ftot < 0, u = ub , d = fdb }, V GV 12 = I2 ,

V GV 21 = I1 ,

RW 12

GW 12

x˙ = f1W (x, u, d) x ∈ I1W

qW 1

V RV 12 = {x := x}, R21 = {x := x}.

x˙ = f2W (x, u, d) x ∈ I2W GW 23

GW 13

RW 51

    w1 w˙ 1 0 −ωe 0 0  w˙ 2   ωe 0 0 0   w2     =  w˙ 3   0 0 0 0   w3  , w˙ 4 w4 0 0 00 

RW 23

GW 51 qW 5

qW 2

RW 13

x˙ = f5W (x, u, d) x ∈ I5W

x˙ = f3W (x, u, d) x ∈ I3W

qW 3

with initial conditions w(0) = (w1 (0), w2 (0), w3 (0), w4 (0))T = (1, 0, v0 , mg)T , where ωe [rad/s] is the known frequency of the reference signal and v0 [m/s] is the initial vehicle velocity. We consider a reference trajectory such that the total traction friction force is nonnegative for vehicle driving mode and strictly negative for vehicle braking mode. Denoting with fFd,ref (w) the reference friction force of the front wheel for driving, and with fFb,ref (w) and fRb,ref (w) the reference friction force of the front and rear wheel for braking, respectively, we define the reference trajectories as a combination of sinusoidal signals:

GW 34

RW 45 x˙ = f4W (x, u, d) x ∈ I4W

GW 45

(14)

RW 34

qW 4 I1W = {λF ≥ 0, λR = 0, ftot ≥ 0, u = ud , d = fdd }, I2W = {λF ≥ 0, λR < 0, ftot ≥ 0, u = udb, d = fdd }, I3W = {λF ≥ 0, λR < 0, ftot < 0, u = udb , d = fdb }, I4W = {λF < 0, λR < 0, ftot < 0, u = ub , d = fdb }, I5W = {λF < 0, λR = 0, ftot < 0, u = ubd , d = fdb }, W = I W , GW = I W , G12 13 3 2

W W W W W RW 12 = R13 = R23 = R34 = R45 = R51 = {x := x}.

ud

= TFd ,

ub

= (TFb , TRb )T ,

ubd

= TFb ,

udb

4

4

4

i=1

i=1

i=1

fFd,ref (w) = ∑ ai wi , fFb,ref (w) = ∑ bi wi , fRb,ref (w) = ∑ ci wi ,

W W W W W W W GW 23 = I3 , G34 = I4 , G45 = I5 , G51 = I1 ,

(15) where the coefficients ai , bi , ci , i = 1, · · · , 4 are determined such that the physical constraints of friction forces are satisfied. Here, we have a1 ≤ fFd,max − croll mg, b1 ≤ fFb,max − (croll /2)mg, c1 ≤ fRb,max − (croll /2)mg, where fFd,max , fFb,max , and fRb,max are derived from (7) and a2 = a3 = b2 = b3 = c2 = c3 = 0, a4 = croll , b4 =

= (TFd , TRb)T

Figure 2. GRAPHICAL REPRESENTATION OF THE HYBRID SYSTEM MODELING THE LONGITUDINAL VEHICLE DYNAMICS AT VEHICLE (V ) AND WHEEL (W ) LEVEL.

4

Copyright © 2009 by ASME

c4 = croll /2. The disturbance fd (w) acting on the system is fd (w) = a4 w4 .

in the open left-half complex plane and such that the regulator equations: (16) ∂πd Sw = f V (πd (w), cd (w), w), 1 ∂w d 0 = hV 1 (π (w), w),

After setting ed = edF and eb = (ebF , ebR )T , the output error equations for the vehicle driving mode and the vehicle braking mode are: ed = eb =



 (x, w), fFd − fFd,ref (w) = hV ! 1 b,ref fFb − fF (w) = hV 2 (x, w). fRb − fRb,ref (w)

CONTROL LAW DESIGN We design the hybrid regulator for the tracking of the traction friction forces of a longitudinal vehicle model in its hybrid formulation (Figure 2) using the nonlinear output regulation [4].

(17)

x˙ = f2V (x, u, w),

ed = h V 1 (x, w), IV

Exosystem (14) − (16)

Ad0 =

1

∂x

∂fV Ab0 = 2 ∂x

, (x0 ,0,0) , (x0 ,0,0)

Bd0 =

1

∂u

∂fV Bb0 = 2 ∂u

, (x0 ,0,0) , (x0 ,0,0)

α (x, w)

= cb (w) + K b (x − πb (w)),

Kb

Kd

fd (w) (9) − (13) Plant

Figure 3. HYBRID REGULATOR SCHEME OF LONGITUDINAL VEHICLE DYNAMICS WITH LOAD TRANSFER EFFECTS.

∂hV

Vehicle Driving Regulator We solve the regulator equation (22) for the vehicle driving mode.

(0,0)

(19) where the initial state x0 is given by x0 = (v0 , 0, 0)T and the pairs (Ad0 , Bd0 ), (Ab0 , Bb0 ) are stabilizable, and given an autonomous linear system, Poisson stable, w˙ = Sw with w(0) ∈ W o ⊂ R4 (with W o neighborhood of the equilibrium point), generating (17), design controllers of the form

b

αd (x, w)

x

C0d = 1 , ∂x (0,0) V ∂h , C0b = 2 ∂x

αd (x, w) = cd (w) + K d (x − πd (w)),

Supervisor

Hybrid Controller

IV ,

∂fV

w

Braking Regulator

with its invariants 1 and 2 respectively, as reported in Figure 2, and with nominal values of their linear parts around the equilibrium point (x0 , 0, 0) ∂fV

αb(x, w)

Driving Regulator

(18)

eb = h V 2 (x, w),

f2V (πb (w), cb (w), w),

b hV 2 (π (w), w), (21) are satisfied for all w ∈ W o and for all admissible values of the parameters of the plant and the exosystem.

The control problem is then formulated as follows. Given the high-level hybrid system model of the longitudinal vehicle dynamics (top of Figure 2) of the form x˙ = f1V (x, u, w),

∂πb Sw = ∂w 0=

∂πvd S = k0 (π f d + π f d − π f d ) F R d ∂w c d  π d  π d + π d − π d  ∂πλd fd f f uF fF R F F πγd − k1 ηdF + k2 ηd S = −k0 F ∂w πvd πvd πvd F d,re f 0 = π f d − fF . F (22) Given (23), we compute the steady-state zero output manifold πd (w) = (πvd , πλd )T and the steady-state input cd (w) = cud , as F F follows

(20a) (20b)

where x = πd (w) ∈ I1V ⊂ R2 and x = πb (w) ∈ I2V ⊂ R3 with πd (0) = πb (0) = 0 are the steady-state zero output manifolds, and where u = cd (w) ∈ I1V ⊂ R and u = cb (w) ∈ I2V ⊂ R2 with cd (0) = cb (0) = 0 are the steady-state control inputs, for driving and braking dynamics, respectively, with K d and K b matrices such that the eigenvalues of (Ad0 +Bd0 K d ) and (Ab0 +Bb0 K b ) are

πvd = vd |x=πd (w) ,

(23a)

π f d = fdd |x=πd (w) = a4 w4 ,

(23b)

cud = udF |u=cd (w) ,

(23c)

d

F

πµd = DF sin(CF arctan(BF πλd )), F

5

F

(23d)

Copyright © 2009 by ASME

along with (20a), where K d is designed based on the linearized vehicle driving mode around the equilibrium point (x0 , 0, 0) with x0 = (v0 , 0, 0)T and described by the matrices Ad0 , Bd0 , C0d in the Appendix.

and h − mπ˙ vd , L L d = fF |x=πd (w) = πµd πN d ,

πN d = NFd |x=πd (w) = mg F

πfd F

lR

F

(24a) (24b)

F

πγd = γdF |x=πd (w) = (1 − πλd ),

(24c)

= (1 − πλd )2

(24d)

F

F

πηd = F

ηdF |x=πd (w)

F

Vehicle Braking Regulator We compute the steady-state zero output manifold πb (w) = (πvb , πλb , πλb )T and the steady-state input cb (w) = (cub , cub )T F R F R for the vehicle braking mode, solving the regulator equation (30)

From the last equation of (22), one obtains π f d = fFd,ref , and after F substituting this together with (15), (16), (23b), (24b) in the first ∂π d equation of (22), one obtains v S = a1 w1 , whose solution is ∂w m πdv = αw2 + w3 with α = a1 /mωe . By inversion of (23d), one obtains πλd = p0 tan(p1 arcsin(p3 (πµd ))), F

∂πvb S = k0 (π f b + π f b − π f b ) F R d ∂w π b  c b  π b +π b −π b  ∂πλb fd fR f fF uF F F − k1 + k2 S = −k0 (1 + πλb ) F ∂w πvb πvb πvb π b  c b  πfb + πfb − πfb ∂πλb fR uR R R d ) − k3 S = −k0 (1 + πλb )( F + k4 R ∂w πvb πvb πvb   b,re f ! π f b − fF 0 F . = b,re f 0 π f b − fR R (30) The terms of (30) are defined as

(25)

F

after combining (24a) and (24b). Given the previous expressions of π f d and π˙ vd , one gets πµd of the form F

F

π µd = L F

a1 w1 + a4w4 mglR − ha1w1

.

(26)

πvb = vb |x=πb (w) ,

π f b = fdb |x=πb (w) = a4 w4 ,

(31a)

cub = ubF |u=cb (w) ,

cub = ubR |u=cb (w) ,

(31b)

F

d

R

πµb = DF sin(CF arctan(BF πλb ),

(31c)

πµb = DR sin(CR arctan(BR πλb ),

(31d)

F

From the second equation of (22), one gets the steady-state control input cud (control input of front wheel):

R

F

F

IF

πvd

r (1 − πλd )2 F

Ls πλd + F

IF 1  π fFd − π fdd  1 − π λd

r m

+ rπ fF d

(27)

πN b

R

F

πfb with Ls πλd = F

∂πλd

F

∂w

F

h − mπ˙ vb , L L h l F = NRb |x=πb (w) = mg + mπ˙ vb , L L = fFb |x=πb (w) = (πµb )(πN b ), F

R

1

1

cos2 (p1 arcsin(p2 πµd )) (1 − (p2 πµd )2 )1/2 (mglR − ha1w1 )2 F F (28) and ρdF = −ωe a1 (mglR + ha4 w4 )w2 . Finally, the steady-state solution and the feedback control law for the vehicle driving mode are

πvb = βw2 + w3 ,

(31g) (31h)

F

πλb = q0 tan(ϑ1 ), F

πλb = z0 tan(ϑ2 ), (32) R

where ϑ1 = q1 arcsin(q2 πµb ), ϑ2 = z1 arcsin(p2 πµb ), and



F

(31f)

Analogously to the vehicle driving mode, we find the solution of (30), given by

,

αw2 + w3 (29a) p0 tan(p1 arcsin(p2 πµd )) , F πvd IF IF 1  π fFd − π fdd  Ls πλd + , , + rπ fF d(29b) cd (w) = F r (1 − πλd )2 r m 1 − π λd

πd (w) =

(31e)

F

F

F



lR

π f b = fRb |x=πb (w) = (πµb )(πN b ).

S = Lp0 p1 p2 σdF ρdF , where

1

σdF =

R

πN b = NFb |x=πb (w) = mg F

cu d =

F

R

cub = IF πvb Ls πλ b + IF 1 (1 + πλ b )(π f b + π f b − π f b ) + rπ fF b , F  F F R F d r r m I I 1 R R cub = πvb Ls πλ b + (1 + πλ b ) π f b + π f b − π f b + rπ fRb , R R R R F d r r m (33)

F

6

Copyright © 2009 by ASME

where

R

c1 w1 + c4 w4 mglF + h(c1 + b1)w1

Ls πλb =

,

Ls πλb =

F

R

∂πλb

F

∂w ∂πλb

R

∂w

S = Lq0 q1 q2 σbF ρbF , S = Lz0 z1 z2 σbR ρbR ,

with ρbF = −ωe (mglR b1 − hb4(b1 + c1 )w4 )w2 , ρbR = ωe (mglF c1 + hc4(b1 + c1 )w4 )w1 , 1 1 1 , σbF = cos2 (ϑ1 ) (1 − (p2πµb )2 )1/2 (mglR − h(b1 + c1 )w1 )2 F 1 1 1 , σbR = cos2 (ϑ2 ) (1 − (z2πµb )2 )1/2 (mglF + h(b1 + c1)w1 )2 R (34) with β = (b1 + c1)/mωe , q0 = 1/BF , q1 = 1/CF , q2 = 1/DF , z0 = 1/BR, z1 = 1/CR , z2 = 1/DR . The feedback control law for the vehicle braking mode is given by (20(b)) where K b is designed based on the linearized vehicle braking dynamics around the equilibrium point (x0 , 0, 0) with x0 = (v0 , 0, 0)T and described by the matrices Ab0 , Bb0 , C0b in the Appendix.

tracking reference friction force on front wheel 4000

Kb =

  = − 7587 11983 ! ! 953 12683 1109 k1b k2b k3b =− . 594 1175 6800 k4b k5b k6b

k1d k2d



actual: f

F

friction force [N]

1.61

1.62

1.63

−200

0

0

−1000

0

2

4

6

8

−2000

10

1000 500

0 −500

ZOOM

0 −20

−2500 −3000

2

4

2

4

6

2

8

10

4

20

−1500

0 2

−2500

6

8

−3000

10

shaft

0

4

6

2

8

0.3 2

1000

0.2

1

−3

0 torque [Nm]

8

10

actual slip ratios

brake

0.25

200 150 100 50

x 10

λ

ZOOM 1

F

λR

0

0.15 ZOOM 1

−1

0.1

−2 1.6

0.05

1.61

1.62

1.63

0 −0.05

0

−0.1

−50 1.6

−6000

6 time [s]

2000

−5000

10

4

and T

3000

−1000

10

ZOOM

time [s]

−4000

8

40

−1000

−20

control inputs: T

−3000

6

0 −500

−2000

0

−2000

1.63

time [s]

500

20

1.62

tracking friction force error on rear wheel: fR−fR,ref 1500

−1500

0

1.61

time [s]

1000

40

1.6

tracking friction force error on front wheel: fF−fF,ref 1500

−1000

−100

1000

−1000

−2000

error friction force [N]

−10 1.6

R

0

2000

1000

desired: π

f

3000

F

0 2000

R

ZOOM 1

f

3000

NUMERICAL RESULTS In this section, we present simulation results, given in MATLAB, concerning a specific model of a vehicle, with the following nominal parameters (extracted from the “Big Sedan” vehicle in CarSim): m = 853.5 kg, IF = IR = 0.9 kgm2, h = 0.515 m, lF = 1.033 m, lR = 1.657 m, r = 0.278 m, croll = 0.01, B = 7, C = 1.6, D = 0.8. The reference friction forces on the front and rear wheel fFref (t) and fRref (t) and the disturbance fd (t) are assumed to be generated by a four-dimensional neutrally stable exosystem, with parameters ωe = 1 [rad/s] and initial conditions w(0) = (w1 (0), w2 (0), w3 (0), w4 (0))T = (1, 0, 15 [m/s], 8373 [N])T . Furthermore, the coefficients of the reference friction forces have been set to a1 = 0.65 fFd,max = 2333 [N], b1 = 0.30 fFb,max = 934 [N], c1 = 0.50 fRb,max = 1793 [N]. Following Section IV, the controller is designed, for each operating condition, on the basis of the vehicle model described by the VHS ((9), (12)), and the simulations are performed on the full nonlinear vehicle model described by the WHS ((9)-(13)). In this way, we obtain the regulation for the regulation error e(t), by the following suitable choice of the control parameters: 

actual: f

desired: π

−2000

Kd =

tracking reference friction force on rear wheel 4000

ZOOM 1 10

friction force [N]

π µb = L

mglR − h(b1 + c1)w1

,

error friction force [N]

F

b1 w1 + b4w4

slip ratio [adm]

π µb = L

Figure 4 presents simulation results for initial conditions (v(0), ωF (0), ωR (0)) = (15.25 [m/s], 70 [rad/s], 54.85 [rad/s]). re f The figure shows the tracking of reference friction forces fF (t) ref and fR (t) (first row of figure) and the related tracking errors (second row of figure). Starting from an initial mismatch of the current friction force with the desired friction force of the front wheel, the friction force tracking error of the front wheel reaches zero around t = 0.5 s. In the third row of Figure 4 the two control variables Tshaft and Tbrake , with the related switching variables λF and λR , are reported. It is clear that the controller allows

1.61

1.62

T

1.63

shaft

−0.15

T

brake

−7000

0

2

4

6

8

10

−0.2

0

2

4

time [s]

6

8

Switching Discrete States of Hybrid Model:ZOOM 1

Switching Discrete States of Hybrid Model:ZOOM 2

1

1 W 1

State q

0.9

W 1

State q

0.9

State qW

State qW

3

0.8 0.7

3

State qW

0.8

State qW

State

0.7

State q

4 W q 5

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1 0 1.6

10

time [s]

4 W 5

0.1 1.605

1.61

1.615

1.62

1.625

1.63

0 4.66

4.665

4.67

4.675

4.68

4.685

4.69

Figure 4. REGULATION OF FRICTION FORCES FOR A VEHICLE MODEL: tracking friction forces on front fFref (t) and rear fRref (t) wheels

(first row), tracking errors fF (t) − fFref (t), fR (t) − fRref (t) (second row), control inputs Tshaft , Tbrake and slip ratios λF (t), λR (t) (third row) and, switching from qW to qW via qW (left, fourth row) and switching from

(35a)

1

(35b)

4

3

W W qW 4 to q1 via q5 (right, fourth row) .

7

Copyright © 2009 by ASME

[4] A. Isidori, Nonlinear Control Systems. Springer-Verlag, London, UK, 1995. [5] H. Lee and M. Tomizuka, “Adaptive vehicle traction force control for intelligent vehicle highway systems (IVHSs),” IEEE Transactions on Industrial Electronics, vol. 50, no. 1, pp. 37-47, February 2003. [6] R. N. Jazar, Vehicle Dynamics: Theory and Application. Springer, 2008. [7] A. Scacchioli, Hybrid Regulation of Electromagnetic Valves in Automotive Systems. EECS PhD thesis, University of L’Aquila, Italy, May 2005. [8] C. J. Tomlin, J. Lygeros and S. S. Sastry, “A game theoretic approach to controller design for hybrid systems,” in Proceedings of the IEEE, vol. 88, no. 7, pp. 949-970, July 2000. [9] E. Velenis and P. Tsiotras, “Minimum time travel for a vehicle with acceleration limits: theoretical analysis and receding horizon implementation,” in Journal of Optimization Theory and Applications, vol. 138, no. 2, pp. 275-296, August 2008. [10] E. Velenis. P. Tsiotras and J. Lu, “Optimality Properties and Driver Input Parameterization for Trail-Braking Cornering,” European Journal of Control, vol. 14, no. 4, pp. 308– 320, 2008.

the tracking of reference friction forces and the stabilization of the vehicle despite known disturbances acting on model and unknown model uncertainties. Finally, in the last row of Fig. 4 the W switching signals between qW 1 and q4 of the automaton of Fig. 2 are shown. As can be easily verified, the intermediate states qW 3 and qW 5 are indeed transient, having a very short duration. CONCLUSIONS AND FUTURE WORK We have applied the nonlinear output regulation theory to the design of a state-feedback vehicle control system, incorporating expert driving skills (ie, left foot braking) to adjust traction friction forces. Specifically, we considered the left-foot braking, where the load transfer effect plays a strategic role for accident avoidance in hazard scenarios. A longitudinal FWD vehicle model with load transfer dynamics has been derived and its formulation using hybrid models has been proposed. The designed hybrid controller, described in Figure 3, is given by a vehicle driving regulator αd (x, w), which computes, through (29), the control input torque for the front wheel during the vehicle driving mode and, a vehicle braking regulator αb (x, w), which computes, through (32), the control input torques for the front and rear wheels during the vehicle braking mode. The hybrid controller depends on eight design parameters (kid with i = 1, 2 and kbj with j = 1, · · · , 6.) We verified, via numerical simulations that, given a large compact set of initial conditions, it is possible to tune the design parameters in to achieve the desired control objective despite known disturbances and unknown model uncertainties. High-fidelity numerical simulations (using CarSim) of the presented algorithm are under investigation.

Appendix ! ! d d d  a a b 11 12 11 Ad0 = , Bd0 = , C0d = cd11 cd12 , d d d a21 a22 b21     b b b b a11 a12 a13 b11 bb12      Ab0 =  ab21 ab22 ab23 , Bb0 =  bb21 bb22 , C0b = cb11 cb12 cb13 ab31 ab32 ab33 bb31 bb32  g g with ad11 = − lR C f a,F , ad12 = g lR C f a,F , ad21 = 2 lR 1 + v0 L L v0 L   g lR m m 2 2 d d r 1+r C f a,F , a22 = − C f a,F , b11 = cd12 = 0, IF v0 L IF   g bd21 = 1 r , cd11 = 1, ab11 = − 12 C f a,F lR + C f a,R lF , v0 IF v0 L   g g g b b a12 = C f a,F lR , a13 = C f a,R lF , ab21 = 13 C f a,F lR 1 + L v0 L  L   g lR m 2 b r + C f a,R lF , C f a,F 1 + r2 m , a22 = − IF v0 L IF   g g ab23 = − lF C f a,R , ab31 = 13 C f a,F lR + C f a,R lF 1 + r2 m , v0 L IR  v0 L  g lR g lF b b a32 = − 1 + r2 m , C , a33 = − C v0 L f a,F v0 L f a,R IR r r, 1 1 b b b b b b b b ,b = b11 = b31 = b12 = b22 = c12 = c13 = 0, b21 = v0 IF 32 v0 IR cb11 = 1, where C f a,F = BF CF DF , C f a,R = BRCR DR and v0 = v(0).

ACKNOWLEDGMENT The authors gratefully acknowledge Ford Motor Company for their technical and financial support under the URP program. In addition, the first two authors would like to acknowledge the financial support from the US DoD SAVE program (contract no. 0603734A) and the NSF GOALI program (CMMI 0727768).

REFERENCES [1] E. Bakker, L. Nyborg and H. B. Pacejka, “Tyre modelling for use in vehicle dynamics studies,” Proceedings of Society of Automotive Engineers international congress and exposition, (Deatroit, Michigan), February 23, 1987. [2] C. Canudas de Wit, P. Tsiotras, E. Velenis, M. Basset, and G. Gissinger, “Dynamic friction models for road-tire longitudinal interaction,” Vehicle System Dynamics, vol. 39, no. 3, pp. 189-226, 2003. [3] T. D. Gillespie, Fundamentals of Vehicle Dynamics. International Society of Automotive Engineers, Warrendale, PA, USA, 1992.

8

Copyright © 2009 by ASME

Nonlinear-Feedback Vehicle Traction Force Control ...

are masters of controlling the vehicle at these extreme regimes. Incorporating expert .... with h [m] the vertical distance to the vehicle's center of mass,. lF [m] the ...... for their technical and financial support under the URP program. In addition ...

723KB Sizes 2 Downloads 200 Views

Recommend Documents

Online statistical estimation for vehicle control
Mar 6, 2006 - stochastic steepest gradient descent, possibly the simplest online estimation ... 3.2 Definition (Steepest gradient descent) Given a vector xt ∈ Rn, and f ..... It is of course possible to use other measures than the euclidean norm.

VEHICLE MOTION CONTROL-module4.pdf
the type of electronic feedback control system. ... such that throttle control reverts back to the accelerator ... The throttle actuator opens and closes the throttle in.

man-117\service-traction-control-2007-avalanche.pdf
man-117\service-traction-control-2007-avalanche.pdf. man-117\service-traction-control-2007-avalanche.pdf. Open. Extract. Open with. Sign In. Main menu.

Multiple Vehicle Driving Control for Traffic Flow Efficiency
challenges such as traffic flow efficiency, safety enhancement and environmental ..... 73 ms to post-process the vision data. Two LIDARs were used to maintain ...

Safe Receding Horizon Control of an Aerial Vehicle
real time high performance controller and trajectory generator for air vehicles. ... keeping control efforts, as well as total time of flight, small. Furthermore, we ...

Safe Receding Horizon Control of an Aerial Vehicle
ation ai ∈ R3. There is also a ... where pi,ai are defined above, b,c > 0 are scalar weights, g represents ..... http://www.cc.gatech.edu/projects/large models/ps.html.

Online Electric Vehicle Control Using Fuzzy Logic Controller.pdf ...
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. Online Electric ...

Online statistical estimation for vehicle control: A tutorial
a very detailed model has the disadvantage that it then becomes difficult to solve. In the end we ...... natural quantity to minimise and has the advantage that it does not rely on knowledge of the car. .... Dover, second edi- tion, 1994. Richard S.

Design Optimization of Vehicle Control Networks
Compared to the case of a conventional data network which typically considers ...... best swapping of tasks between ECUs needs to be found. This ..... London, UK: .... Conference, the IEEE Symposium on Personal, Indoor, Mobile, and Radio.

PDF Space Vehicle Dynamics and Control (AIAA ...
... Vehicle Dynamics And Control Aiaa space vehicle dynamics and control aiaa education series PDF co uk ebook php ebook php manual pdf vehicle dynamics ...

Proprioceptive control for a robotic vehicle over ... - IEEE Xplore
Inlematioasl Conference 00 Robotics & Automation. Taipei, Taiwan, September 14-19, 2003. Proprioceptive Control for a Robotic Vehicle Over Geometric ...

Safe Receding Horizon Control of an Aerial Vehicle
In the vehicle control domain, this often boils down to two things: not getting ... vehicles a guaranteed obstacle free circular loitering pattern ensures safety, [2], [5] ...

Robust Broadcast-Communication Control of Electric Vehicle ... - arXiv
Jun 1, 2010 - two-way energy flows to buffer time-variable renewables[1],. [2], [3] or to provide ... EV loads on the circuit, and multiple classes of EV charging powers. ..... source for electric utilities,” Transportation Research Part D: Transpo

Force Control of Flexible Catheter Robots for Beating ...
authors are with the Harvard School of Engineering and Applied Sciences, ... Division of Health Sciences & Technology, Cambridge, MA, 02139 USA.

Abnormal sensorimotor control, but intact force field ...
their adaptation capability. Recently, it has been shown that MS subjects with little or no disability have subtle problems in gait and balance [8] and in controlling arm move- ments [9]. Multiple sclerosis subjects with mild dis- ability also displa

Vehicle Reusability
Two round trips (same energy as getting to low Earth orbit = $26/kg. Factor of 60x electrical energy costs. Factor of 250x ..... MARYLAND. Solar Power Satellites?

Vehicle Dynamics
(c) Physics of tyre traction ... (c) Effect of wetness on the automobile designed for dry traction. ... (b) Differences in the tyre traction on dry and wet roads.