2001 IEEE Intelligent Transportation Systems Conference Proceedings - Oakland (CA),USA - August 25-29, 2001

A Dynamic Tire/Road Friction Model for 3D

Vehicle Control and Simulation
Xavier Claeys, Jingang Yi, Luis Alvarez, Roberto Horowitz, Carlos Canudas de Wit
ric and are based on empirical curve fitting for the friction forces and torque. The parameters are calibrated through data collected in suitable experiments. Similar parametric approaches are presented in [5] and [lo]. In particular, the model proposed in [2], commonly called the magic formula, has been used in industry as a bench mark for tire/road friction models. However, these models have disadvantages. They are over-parameterized, difficult,to calibrate and are not suitable for direct identification. In addition, both models ignore the physics and dyI. INTRODUCTION namics of the friction process. Their parameters lack One of the main arguments for deploying intelli- any physical interpretation, making it difficult to ingent transportation systems (ITS) is that the capac- tegrate model dependencies on specific factors such ity of highways will be increased while maintaining or as humidity of the road or tire tread temperature. increasing actual levels of safety. A very important Also, these models need to be re-calibrated for difissue, which requires careful analysis, is the influence ferent road and vehicle conditions. There are some of tire/road friction on individual vehicle safety and identifiable approximations to the models, proposed therefore on the overall highway safety. This fric- recently in [11]and [I]. The parameters in these modtion is a key factor in determining vehicles braking els can be estimated in real-time. The model in [I] capabilities. The tirejroad interaction depends on has an interesting estimation property, in that it can many factors at the vehicle and roadside levels: ve- guarantee the underestimation of the tire/road cohicle conditions, type and quality of each tire, condi- efficient of friction, which helps to maintain vehicle tion of pavement, weather, etc. If the goal of ITS safety. is to maintain high levels of safety, it is desirable In an effort to overcome the limitations of the to make vehicles that are able to adjust their bepseudo-static friction models, [8] and [4] recently prohavior to accommodate for changes in the road or a vehicle conditions. For example, in the case of wet posed dynamic friction models. It w s claimed that pavement, inter-vehicle spacing should be increased, these models capture most of the important phenomas the tire/road friction is expected to decrease in ena that occur during the friction process. These this situation. An accurate and identifiable model of models are suitable for identification and friction tire/road friction is therefore of prime importance for compensation in mechanical systems. [9] extended the use of the dynamic friction models to tire/road vehicle control and simulation. interactions. The authors of [9] introduced a LuGre Research in the area of tirejroad friction models is significant. [2] and [SI proposed two pseudo- first-order dynamic friction model. In [7]this model static frictions models that are widely used by re- was parameterized to allow direct identification of searchers in the field. These models are paramet- road conditions. Adaptive emergency braking control and comparison of this model with the magic Research supported by California PATH grant MOU 373 and formula are discussed in [13]. All the dynamic moda research gift from Renault Inc. els that have been applied to tire/road friction are Xavier Claeys and Canudas Canudas de Wit are with constrained to consider only longitudinal motion, i.e., the Laboratoire dAutomatique de Grenoble, ENSIEG-INPG, France. Xavier Claeys is currently a visiting scholar they disregard the lateral forces. at the University of California at Berkeley. E-mail: This paper extends the work in [13] by proposXavier.ClaeysQinpg.fr. Jingang Yi and Roberto Horowitz are with the Department ing a three dimensional LuGre-type dynamic friction of Mechanical Engineering, University of California at Berke- model. This kind of model is required for new develley, CA, USA. opments in vehicle control design, for example, adLuis Alvarea is with the Instituto de Ingenieria, Universidad Nacional Autonoma de Mexico, Mexico. vanced braking control or observer-based road monAbstract- A tire/road friction model based on the LuGre dry friction model and on tire dynamics is presented. The dynamics of the longitudinal and lateral forces, and the self-aligning torque are described by a set of flrst order differential equations. This model is suitable for 3D vehicle traction/braking simulation and control. A comparison of the forces and torque produced by this dynamic model with the well known magic formula i s presented. Keywords-Tire/road friction, LuGre model, vehicle control and simulation.

0-7803-7194-1/01/$10.00 02001 IEEE

483

. .

Contamlnallon (water, mow) Tire "design"

Inflating prcssure,

Road texture
Wear

Temperature Iintcmal) Tire geometry angles,

f
Tire m
Temperature (tread

Fig. 1. Schematic representation summarizing the different problems in tire modeling.

itoring. To analyze the tirelroad interface, a description that separates the different elements affecting the dynamic behavior is proposed. Fig. 1 shows the three main factors: friction between tire and road, defor- 0 In the braking case, longitudinal slip S, and lateral slip S, are given by mation of the shell, and contact patch area. This paper focuses only on tire/road friction. Although friction properties are only one part of the tirelroad interaction, a thorough understanding is still necessary. For this reason a three dimensional model of tire/road contact forces and momentum is proposed. In braking V, - V, > 0, V, # 0 , then 1 2 S, > 0. This model is compared numerically with the wellknown "magic formula" and shows a very good fitting. In the traction case, longitudinal slip S, and latThis model is a dynamic description of the contact eral slip S, are given by properties; and therefore it should be able to describe richer behavior than the pseudo-static models. The paper is arranged as follows: in section I1 we define the coordinate systems and longitudinal and S, = 2 = (1 - S,)ltan cy1 lateral slips which are used for modeling. A disv c tributed friction model is presented in section I11 In traction, if V,-Vc < 0, w # 0, then 1 2 S, > 0. based on the LuGre model and the physical dynamics of the tire. Stationary properties of such a model Other conventions, like those used for the "magic are also presented. A lumped LuGre model is dis- formula" (see [3]), can easily be employed and do not cussed in section IV. This lumped friction model change the final results. In this particular case, a is equivalent to the distributed model with certain specific definition needs to be considered when the boundary conditions but is easier to use for estima- longitudinal speed V, or Vc tend to zero, in order to tion and control purposes. Parameter calibration and prevent a singularity in the definition of the slip. a comparison with the "magic formula" are included DYNAMIC TIRE/ROAD FRICTION in section V. Concluding remarks and future work 111. DISTRIBUTED MODEL axe discussed in section VI. Several dynamic tire/road friction models have al11. SLIPDEFINITIONS ready been studied which are restricted to only longiIn In this section we recall the definitions of slip that tudinal motion [9], [4]. this section we propose to' are commonly used in tire/road friction modeling. extend the models to consider both longitudinal and. We use V = [V,,V,] to denote the translational lon- lateral motions, and investigate the resulting forces gitudinal velocity at the center of the contact patch and torque at the center of the contact patch. 0 (see Fig. 2) in the wheel plane X O Y . The angular velocity of the wheel is denoted by w, V, is the A . Two dimensional model Let St represent a small slice of the deformed belt equivalent wheel translational velocity at the point 0 and V, = pw where p is the free radius of the tire. crossing the contact patch at position E in coordi.-

When w is positive (the vehicle is moving forward), the wheel equivalent longitudinal velocity Vc is positive in the coordinate system &, where & is defined as a moving frame with velocity V, with the origin 0 in the center contact patch. We define the slip velocity of the point 0 as V, = [V,, ,V,,] in the wheel plane X O Y . The slip angle is denoted by cy. The slip ratios used to parameterize the friction model are defined as S, and S,, for longitudinal and lateral directions, respectively. Two conventions will be used to separate the braking and traction cases (see Fig. 2), since as usual, the pseudo-static braking curves are developed under constant velocity whereas pseudo-static traction curves are given for constant angular velocity. These conventions also prevent the slip from becoming undefined when either the wheel speed or the longitudinal speed reach zero.

484

Traction
Y

., * ,

.........

. . .

*.._...*
............

v c t,,
'.,

' r ' j v vx

coefficients of the LuGre friction model for lateral and longitudinal directions, known as the normalized rubber stiffness (uo;), normalized rubber dampthe ing ( n l , ) , and the normalized viscous relative damping ( 0 2 ; ) . The normal load 6Fn is considered uniformly distributed over the patch along ( direction, thus6Fn = F n / L , and

I 1 ofmotion

Direction

vy

vx

Braking
:...-J, ....:
X

Contact patch

Fig. 2. Convention for the slip definition for both braking and traction cases

nate & frame (see Fig. 2). The slice 6[ is moving at the speed = [Vz,Vy [@] with @ the yaw speed of the rim [12]. We can model the dry friction present in each slice using the LuGre dynamic friction model. The contact between the two surfaces can be represented by microscopic bristle deflections with the coordinates Sz(<,t ) = [z(, t ) ,6z,([, t)], S,[ and the relative velocity of each slice at [ with respect to 0 is given as Vr(Elt) [V,,([, t ) ,VTY(El ]= = t) [-VsZ(t),-Vsy(t) [+] (the direction of the total force is opposite to the slip vector). However, in this paper we treat the tire belt as a rigid body. The model might be extended in order to include dynamic properties of the rubber belt. In this case, the longitudinal velocity of each slice Vac of the contact patch would have a more complex expression that might include camber angle dependencies or other factors. These notions have already been explained in the literature [12], [lo]. For the rigid tire belt model, the extended two dimensional distributed tire/road friction model is given by

are two functions that characterize the steady state properties of the friction] where pc,, pus,, are, reus, spectively, the Coulomb friction coefficient, the normalized static friction and the Stribeck relative velocity. The two LuGre models for longitudinal and lateral motions use different parameters since the friction properties of the contact tire/road are different in longitudinal and lateral directions. The fact that the tire has non-isotropic properties is well known in this area and has already been introduced in most of the current models [3], [5]. The system given in Eqs. (1) and (2) is both time and space dependent, and as a consequence the derivative of szi(E,t), i = z,y, for longitudinal and lateral directions is a full derivative given by

The system composed of (l),(2) and (4) is difficult to solve analytically. However, the stationary case (i.e. pseudo-static case, V, and 20 are constant) can be studied and compared with the available stationary tire models in this research area. This will be discussed in section 111-C.

B. Self-aligning torque
The self-aligning torque is an important part of the tire model because the reaction force applied to the vehicle (steering wheel feedback force) is strongly dependent upon it. The self-aligning torque consists of two important elements, the yaw motion of the tire that creates a friction torque 6Mz,(<, t ) ,and the moment of the friction forces about the center of wheel frame 6Mz, ((E, t). Both of these effects generate the torque at the center of patch, known as the selfaligning torque. In the current paper we neglect the yaw motion of the tire 6Mz, (<, t ) ,due to the fact that the yaw motion effect is small [12]. Denoting 'p as the yaw angle, this torque could easily be described by using a third dynamic friction model similar to the LuGre model given in section I11 (Eqs. (5) and (6)). Given a yaw bristle deformation]

and the friction forces

485

the friction torque could be expressed by the following equation

conditions, we have

6 M z , = { 00,6zz

(5,t ) + O i z 6 i z(5,t ) + ~2~ $1 6Fn

(6)

The tire/road forces and torque are always calculated at the center of the patch 0 in Fig. 2. Consequently, a self-aligning torque is produced resulting from the non-symmetry of the contact patch deformation bz([,t ) , or forces bF,(c, t ) , over the contact patch length L. The equivalent forces and torque produced by a slice 65 at position t with respect to the center of the patch 0 is given by [6F,, 6F,, 6Mz = [SF,] and the total tirefroad interaction in the patch center is now expressed by two forces Fx, and the F, self-aligning torque M z :

with spatial coordinates. As a consequence, the stationary bristle model with spatial coordinates becomes

with Szi = 0 as the boundary condition. The spatial solution is given by

(-4)

Integrating the forces and torque along the contact patch (using the formula (7)), we obtiin three components for the stationary tire model F,, FY, z for M the traction case:

Remark 1 Eventually, we could also look for the non-symmetric force distribution along the width of the patch and add other components to the model. In particular, the self-aligning torque will then depend upon the longitudinal force 6 F z ( [ ,t ) as confirmed by the experiments. 0

C. Stationary properties
The stationary characteristics of the tire are widely applied in this research area. To produce these characteristics, a complex experimental setup is usually utilized. These conditions are hard to obtain on a real vehicle, since the required maneuvers would be very severe for the passengers. Each point on the stationary curve is given for a constant slip and a constant wheel velocity or wheel angular velocity, therefore V , and V remain constant and a slip angle a is obtained. The yaw motion of the rim is also not considered, i.e. (L, = 0. Therefore, during the stationary conditions, 5 and t are no longer independent because of constant velocity, thus we have Szi(5, t ) = Szi(t) if we desire a time varying solution, or 6zi([,t) = Szi(S) if we require a spatial solution. We choose to calculate the spatial solution .in the frame &, defined previously. Notice that, if [ = V is constant during stationary ,

The function ~i defined as 7i(Vsi) = 1- -, for i = 2,y . Calibration of parameters and comparison with the "magic formula" are presented later in section V. For the braking case, we can find similar formulae for Fx, y , and M,. F

Iv. LUMPED DYNAMIC TIRE/ROAD

MODEL

FRICTION

Distributed models are difficult to use for estima.. tion and control purposes. Thus, we will now develop a simplified lumped parametric representation. An approach has been given in [9] for deriving a lumped model, assuming null boundary conditions for the internal state are not possible as the deflection is not symmetrical with respect to the center of the patch 0. This is an essential property t o guarantee the existence of a self-aligning torque. In this paper, we

486

obtained the lumped model by defining lumped variables Zi as follows,

we have

where L is defined as an "elementary surface length", which could be a tread block element or the full contact patch length between the tire and the road. Neglecting the yaw motion of the rim, i.e. CL, = 0, the distributed friction model becomes

where = V, is assumed uniformly along the patch t ) = 0, the system and i = z, y. Assuming 6z turns into two first order differential equations which # are similar to [13] (6z(!j,t) 0 because the solution is not symmetric). Finally the lumped model can be defined by 0 Internal states

(-4,

Remark 2 The dynamic tire/road friction model has several interesting properties: The model is an average over the patch length and depends only upon time. The model is not limited to stationary curves, we can describe the system when V, and V are not constant. The bristle dynamics are relatively fast with respect to vehicle dynamics when slip velocity v, is large. However, for small values of the slip velocities V,, and V,, the friction dynamics become slower and should be considered in vehicle control and simula0 tions.

V. PARAMETER CALIBRATIONS AND

SIMULATIONS

NUMERICAL

with Zi(0) = 0 and Sz ( g , O ) = 0 because the patch is a t free when t = 0.

Lumped forces

Using the new lumped internal states Zi and 6zi ($, t ), we can obtain
0

Lumped self-aligning torque

Noticing that

We need to calibrate the dynamic friction model parameters before the model can be used for friction estimation and vehicle control. Moreover, we should validate the model with experimental data. In this paper, we use the typical tire model 165-65R14 as an example. The "magic formula" has been calibrated for dry surface conditions for tires given by Michelin Inc. under a pure braking/corning maneuver. Fig. 3 shows the test results from the calibrated "magic formula". We calibrated the parameters of our model using the nonlinear fitting methods in MATLAB. The model parameters are given in Table I and the force and torque curves in Fig. 3. The result shows a good fit between the model and the empirical approach. Moreover, we found that including the lateral force and the self-aligning torque facilitated the determination of some parameters, such as those in the y direction.

VI. CONCLUSION
In this paper we extended and derived a three dimensional dynamic tire/road friction model, based on previous work which had only considered longitudinal motions. Both distributed and lumped friction models were discussed. The lumped model can be used to identify the tire/road conditions and can be applied t o vehicle control. A numerical example was presented to calibrate the model parameters and

487

validate the model with respect t o the widely used "magic formula". From the analysis and numerical results obtained, we found that the proposed friction model can capture the tire/road friction characteristics and can easily be used for friction estimation and control purposes. Integrating yaw motion and the coupling dynamics between longitudinal and lateral friction into the dynamic friction model is a topic of current research.

REFERENCES
Luis Alvarez, Jingang Yi, and Roberto Horowitz. A d a p tive Emergency Brake Control in AHS with Underestimation of Friction Coefficient. In Proceedings of American Control Conference, pages 574-579, Chicago, IL, 2000. E. Bakker, L. Nyborg, and H.B. Pacejka. Tyre Modelling for Use in Vehicle Dynamic Studies. Society of Automotive Engineers Paper # 870421, 1987. Egbert Bakker, Lars Nyborg, and Hans B. Pacejka. Tyre Modeling for Use in Vehicle Dynamics Studies. SAE Transactions, (870421):190-204, 1987. P.A. Bliman, T . Bonald, and M. Sorine. Hysteresis Operators and Tyre Friction Models: Application t o Vehicle Dynamic Simulation. In Proceedings of ICIAM'95, Hamburg, Germany, 1995. M. Burckhardt. A B S und ASR, Sicherheitsreleuantes, Radschlupf-Regel System. Lecture Scriptum, University of Braunschweig, Germany, 1987. M. Burckhardt. Fahrwerktechnik: Radschlupfregelsysteme. Vogel-Verlag, Germany, 1993. Carlos Canudas de Wit and Roberto Horowitz. Observers for Tire/Road Contact Friction using only wheel angular velocity information. In Proceedings of 38th IEEE Conference of Decision and Control, Phoenix, AZ, 1999. Carlos Canudas de Wit, H. Olsson, K.J. Astram, and P. Lischinsky. A New Model for Control of Systems with Friction. IEEE Transactions on Automatic Control, 40(3) :419-425, 1995. Carlos Canudas de Wit and Panagiotis Tsiotras. Dynamic Tire Friction Models for Vehicle Traction Control. In Proceedings of 38th IEEE Conference of Decision and Control, Phoenix, AZ, 1999. Gwanghun Gim and P.E. Nikravesh. An Analytical Model of Pneumatic Tyres for Vehicle Dynamic Simulations. Part I: Pure Slim. International Journal of Vehicle Design, 11(6):589-'618, 1990. [ll] U. Kiencke. Realtime Estimation of Adhesion Characteristic Between Tyres and Road. In Proceedings of the WAC World Congress, volume 1, 1993. I121 M. Sorine and J. Szymanski. A new dynamic multi d.0.f. tire model. Preprint, 2000. [13] 3 . Yi, L. Alvarez, R. Horowitz, and C. Canudas de Wit. Adaptive Emergency Braking Control in Autcmated Highway System Using Dynamic Tire/Road Friction Model. In Proceedings of 3gth IEEE Conference of Decision and Control, Sydney, Australia, 2000.

, -

c '

Friction mode

slip angle (deg) Fig. 3. Comparison of the stationary tire/road friction model and the "magic formula" (constant velocity during braking U = 15m/s). TABLE I

FRICTION PARAMETERS

FOR

FZ A N D F,
2 = y

Coefficients

a=x .

U,.. ,

I.

21%

1.41 2.66

3.05 1.17