Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Transportation Research Part C 21 (2012) 181195

Contents lists available at SciVerse ScienceDirect

Transportation Research Part C


journal homepage: www.elsevier.com/locate/trc

Surface prediction and control algorithms for anti-lock brake system


Rishabh Bhandari, Sangram Patil, Ramesh K. Singh
Department of Mechanical Engineering, Indian Institute of Technology Bombay, Mumbai, India

a r t i c l e

i n f o

a b s t r a c t
Anti-lock brake system (ABS) has been designed to achieve maximum deceleration by preventing the wheels from locking. The friction coefcient between tyre and road is a nonlinear function of slip ratio and varies for different road surfaces. In this paper, methods have been developed to predict these different surfaces and accordingly control the wheel slip to achieve maximum friction coefcient for different road surfaces. The surface prediction and control methods are based on a half car model to simulate high speed braking performance. The prediction methods have been compared with the results available in the literature. The results show the advantage of ABS with surface prediction as compared to ABS without proper surface identication. Finally, the performance of the controller developed in this paper has been compared with four different ABS control algorithms reported in the literature. The accuracy of prediction by the proposed methods is very high with error in prediction in a range of 0.172.4%. The stopping distance is reduced by more than 3% as a result of prediction for all surfaces. 2011 Elsevier Ltd. All rights reserved.

Article history: Received 29 May 2010 Received in revised form 8 September 2011 Accepted 8 September 2011

Keywords: Anti-lock brake system Surface identication ABS control Sliding mode controller

1. Introduction Anti-lock brake system or ABS is a safety system that preserves the tyreroad rolling friction by preventing the wheels from locking up. Therefore, the driver can control the direction of the vehicle by steering input, which is extremely difcult with the locked wheels. While ABS typically decreases the stopping distances on dry and slippery surfaces, it can also increase the braking distance under certain conditions. Specically, ABS is known to increase the braking distance on loose surfaces such as snow and gravel. However, this shortcoming could potentially be mitigated by developing new control strategies for ABS systems. In order to develop effective control strategies, the vehicle dynamics and the tyresurface interaction need to be considered. A number of vehicle dynamics models for braking simulations have been reported in the literature. Notable among them are quarter car model used by Baslamisli et al. (2007, p. 217232) and half car model used by Hari et al. (2008, p. 731741). A full car model can be used to predict braking on a curved path, i.e., it can capture the effect of side slip and lateral load transfer. However, high speed braking and slipping is generally observed in straight line braking. Consequently, a half car model would sufce in such cases. Various tyre models have been proposed in the literature for longitudinal motion of the car. Hari et al. (2008, p. 731741) have proposed a model that describes the coefcient of friction between tyre and road as a function of slip ratio and velocity. Park and Lim (2008, p. 290295) used a simplied version of this model and expressed coefcient of friction as a function of slip ratio only. Anwar (2006, p. 11011117) has used piecewise linear approximation as a tyre model and Oniz et al. (2007, p. 9095) have also used a model with coefcient of friction expressed in terms of slip ratio. Different control strategies have been employed to control the slip ratio of the vehicle. Anwar (2006, p. 11011117) has proposed a nonlinear sliding mode type controller for slip regulation. Hari et al. (2008, p. 731741) have used a sliding
Corresponding author.
E-mail address: ramesh@me.iitb.ac.in (R.K. Singh). 0968-090X/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.trc.2011.09.004

182

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Nomenclature g a b H mtot m mf mr J R T k acceleration due to gravity distance from CG of front axel distance from CG of rear axel height of mass total mass of vehicle mass front unsprung mass of vehicle rear unsprung mass of vehicle moment of inertia of wheel radius of tyre torque slip ratio coefcient of friction velocity of vehicle wheel rpm longitudinal displacement longitudinal force normal force constant

l
v

x
x Fx Fz K

Subscripts F front R rear S sprung D desired B braking P proportional I integral D derivative

mode control along with an integral switching surface for chatter reduction. Neural networks and Fuzzy logic has also been implemented to control the slip ratio in ABS control strategies in various cases. Braking under changing road conditions requires a methodology for identifying the transition and adequately modifying the control strategy. Zhang et al. (2008, p. 493497) have proposed a neuro-adaptive unit to compensate external disturbances and uncertainty in vehicle dynamics due to the variation in road conditions. They have studied the braking performance of the proposed control strategy under changing road conditions. Various methods have been reported in the literature for prediction of road conditions. Choi (2008, p.9961003) used a Proportional Differential (PD) controller with a phase lag on rear wheel to predict the slip ratio for maximum friction coefcient and the reference angular velocity for the front wheel. Once the angular velocity is determined, a PD control is used to continuously control the front wheel velocity. Note that the above method involves oscillation about the peak for checking the peak which results in loss of braking force during prediction. Oniz et al. (2007, p. 9095) used a Grey predictor along with a sliding mode control to predict the road conditions. Such methods are computationally intensive and need higher processing power. Tanelli et al. (2009, p.891) have proposed some other prediction methods and presented results for various surface conditions. Based on the above discussion, it can be seen that limited work has been reported in the literature for a vehicle transitioning from one road condition to another during braking. Consequently, this paper is focused on implementing computationally efcient algorithms for identication of surface condition followed by an effective anti-lock braking control strategy for changing surface conditions. This paper presents the vehicle dynamics and tyresurface interaction models following which the methods to identify surface characteristics are described. Based on these models, the control strategies required to implement the ABS system have been developed. Finally, the results obtained from the simulations are presented and compared with experimental and simulation results reported in other studies on anti-lock brake systems (see Fig. 1). 2. Vehicle dynamics The development of a vehicle and tyre model is the rst step towards the development of ABS strategy. Once the model is ready, a sliding mode control is used to control the wheel slip ratio at a desired position. The desired position varies for different surfaces and this position is determined by the prediction algorithm. This gives the desired slip ratio for maximum

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

183

Dynamics and Tyre Model

Prediction Algorithm

Controller

Response

Fig. 1. Steps used to develop ABS strategy with prediction.

friction coefcient (kmax) via a PID controller as input to the sliding mode controller as shown in Fig. 2. The sliding mode control then gives braking torque for front and rear wheels (TBf and TBr). Detailed description of the dynamics and controls has been given in following sections. 2.1. Half car model As mentioned previously, half car model is adequate for straight line braking. It also takes into account the longitudinal weight transfer. As shown in Fig. 3, the half car or bicycle model divides the car into two parts and therefore does not take into account side slip and cornering forces. The model has been used by Hari et al. (2008, p. 731741) for the development of ABS strategies. This model assumes longitudinal braking conditions and steering effects are neglected. The equations demonstrating the vehicle dynamics involved in the half car model are,

lkf m1 lkr m2 mtot lkf m3 lkr m3 _ T lkf m1 Rg lkf m3 Rv _ xf Bf 2J f _ T Bf lkrm2 Rg lkrm3 Rv _ xr 2J f


_ v g
where R is the radius of wheels and Jf and Jr are the moment of inertia of front and rear wheels respectively.

1 2 3

m1

bmtot ; a b xf R ; kf 1

m2

amtot ; a b xr R kr 1

m3

mf hf ms hs mr hr a b

4 5

2.2. Friction coefcient as a function of wheel slip The model used by Park and Lim (2008, p. 290295) is based on a tyre friction model developed by Burckhardt and Reimpell (1993). It gives the friction coefcient between the tyre road as a function of wheel slip as:

lk C1 1 ec2 k C3 k

where C1, C2 and C3 are constants for characterizing different road conditions. C1 is the maximum value of the friction curve, C2 gives the friction curve shape and C3 represents the difference between the maximum value of the friction curve and the value when slip ratio is one. The values of the constants for various surfaces are given in Table 1. These values have been obtained from the data used by Hari et al. (2008, p. 731741) and some additional values have been used for the same model by Oudghiri et al. (2007, p. 1328). Fig. 4 shows a plot of the friction coefcient according to this model for all the surfaces mentioned in Table 1.

Vehicle dynamics

Prediction Algorithm

max

(Sliding Mode Control) Controller

Prediction (PID) Control

TBf , TBr
Fig. 2. Flow chart of representing prediction and control loop.

184

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Fig. 3. Half car model, free body diagram (Hari et al., 2008, p. 731741).

Table 1 Values of constants for various surfaces (Hari et al., 2008, p. 731741, (2007, p. 1328)). Surface conditions Dry asphalt Wet asphalt Dry concrete Cobble wet Cobble dry Snow Ice C1 1.2801 0.857 1.1973 0.4004 1.3713 0.1946 0.05 C2 23.99 33.822 25.168 33.708 6.4565 94.129 306.39 C3 0.52 0.347 0.5373 0.1204 0.6691 0.0646 0

Coefficient of friction ()

Slip Ratio ()
Fig. 4. Friction coefcient as a function of slip ratio for different surfaces.

3. Surface prediction algorithms Prediction algorithms are required to identify the surface conditions. A robust ABS controller has been proposed by Baslamisli et al. (2007, p. 217232) using the quarter car model. The proposed method uses the half car model and prior information (shown in Fig. 4) about the nature of friction curves for prediction of the surface the vehicle is treading on at a given instance. The rst step is to move to predetermined values of slip ratios and then compare the measured coefcient of friction with the value of coefcient of friction based on tyre friction model given in Fig. 4 to determine the existing surface conditions. The two methods have been explained in detail below.

3.1. One point prediction method In the one point prediction method, a reference value of slip ratio (k0) is selected initially. At this value of slip ratio, the deceleration of the vehicle is measured using an accelerometer. This deceleration at a particular slip ratio is then related to different surfaces. Once the surface is known, the slip ratio for maximum friction coefcient is also known.

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

185

There are three major constraints while selecting a reference wheel slip ratio, k0. Firstly, it should be close to the range of slip ratio for maximum friction coefcient of all surfaces. This ensures that the loss of braking during prediction is minimal irrespective of road surface. Secondly, the prediction of slip ratio for maximum friction coefcient (kmax) should result in minimal error for all surfaces. Finally, since there is a sharp drop in coefcient of friction before the slip ratio for maximum friction coefcient in all curves (except dry cobblestone), therefore, it should be ensured that the selected value of slip ratio is greater than that for all curves. If the example of dry asphalt is considered then, for a slip ratio value which is 0.1 less than the optimal the reduction in friction is 14.5% as compared to only 2.7% reduction when the slip ratio value is 0.1 more than optimal. Using the data for surfaces available in the tyre model and considering the above constraints a suitable range of slip ratio (0.20.25) has been selected. The values of friction coefcient for this range are computed for all the road surfaces. These friction coefcients when related to the slip ratio for peak friction coefcient give a linear relation for each value of k as shown in Fig. 5. A best t line is then used to nd the value of k with minimum standard deviation. This value is used as k0. The standard deviations for the given range are shown in Table 2. 3.1.1. Relations between kmax and deceleration at k0 Practical measurement of acceleration is pretty straightforward with the available accelerometers in any vehicle. This deceleration can be easily related to l0 using the vehicle dynamics model (Eqs. (1)(3)).

g x

lkf m1 lkr m2 mtot lkf m3 lkr m3

Setting both kf and kr as k0, the equation simplies to

glk0 x

By assuming linear variation between the values of l0 observed at k0 for different surfaces and the peak coefcient of friction lmax for these surfaces a linear relation of the form lmax = A l0 + B (Eq. (9)) can be t, where the constants A and B are found by tting a best t line to the data points. Similarly, a second linear relation (Eq. (10)) can be obtained for each of the individual surfaces, which relates the peak coefcient of friction to the peak slip ratio. The following relations can be obtained for the four surfaces by assuming linear variation

lmax A l0 B lmax C kmax D

9 10

where l0 is the coefcient of friction at k0, lmax is the maximum coefcient of friction for a given surface and kmax is the slip ratio at which this coefcient of friction is achieved. The constants obtained after tting a linear best curve for the four surfaces are: A = 0.9973, B = 0.0111, C = 8.9430 and D = 0.3516. By combining the Eqs. (8)(10), slip ratio for peak coefcient of friction can be related to deceleration as:

kmax Ao Bo ; x

where A0 0:041 and B0 0:0114

11

The prediction method works well for most surfaces, however for surfaces like, cobblestone; the prediction method is not suitable. Considering that anti-lock brakes are useful at high speeds, the method is very useful. Table 3 shows the accuracy of prediction by the algorithm for all surfaces. As shown in Fig. 6, the range of the slope of the best t equation with variation in the reference slip ratio used for one point prediction is 0.0024 which corresponds to a percentage change of 2.15% which is not signicant. Hence it can be said

Slip ratio at peak coefficient of friction (max )

Coefficient of friction at = 0.2 (0 )


Fig. 5. Slip ratio at peak coefcient of friction vs. coefcient of friction at k0.

186

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Table 2 Standard deviation from best t line with different slip ratios. k Std. deviation 0.2 0.0041 0.21 0.0046 0.22 0.0051 0.23 0.0054 0.24 0.0057 0.25 0.006

Table 3 Predicted values using one point prediction method. Surface conditions Dry asphalt Dry concrete Wet asphalt Cobble wet Cobble dry Snow k For peak l 0.1700 0.1600 0.1308 0.1400 0.4000 0.0600 Predicted value of k 0.1716 0.1623 0.1291 0.0830 0.1361 0.0613 Error in prediction (%) 0.94 1.44 1.30 40 65.97 2.17 Peak l 1.1700 1.0900 0.8013 0.3800 1.0000 0.1900

l At predicted k
1.1700 1.0900 0.8013 0.3660 0.7107 0.1900

Error in l (%) 0.01 0.00 0.00 3.66 28.93 0.00

Fig. 6. Sensitivity analysis on the slope of the best t line.

that the choice of the reference slip ratio at 0.2 is relatively robust. The slope referred to in Fig. 6 corresponds to the curve parameter A in Eq. (9). Although, this method could sufce for practical applications but a generalized approach which could account for all surfaces needs to be developed. This is because the method does not work well on surfaces like wet and dry cobblestone. The next section presents a generalized prediction methodology based on three point prediction. 3.2. Three point prediction method To overcome some of the limitations of the one point method, a three point method has been developed. This algorithm tests the deceleration at three different predened slip ratios and uses these values to predict the surface characteristics. Each surface can be dened by the equation as shown in the tyre model by Burckhardt and Reimpell (1993).

lk C 1 1 eC2 k C 3 k

12

If the values of these three constants are known, the peak slip can be located. At the maxima and hence the peak of the slip ratio curve

_ lkmax C 1 C 2 eC2 kmax C 3 0


Therefore, slip ratio for maximum friction coefcient is given by

13

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195 Table 4 Predicted values using three-point prediction method. Surface conditions Dry asphalt Dry concrete Wet asphalt Cobble wet Cobble dry Snow k For peak l 0.1700 0.1600 0.1308 0.1400 0.4000 0.0600 Predicted value of k 0.1683 0.1587 0.1299 0.1396 0.4096 0.0601 Error in prediction (%) 1.00 0.81 0.69 0.29 2.4 0.17 Peak l 1.1700 1.0900 0.8013 0.3800 1.0000 0.1900

187

l At predicted k
1.1700 1.0900 0.8013 0.3800 0.9998 0.1900

Error in l (%) 0.00 0.00 0.00 0.00 0.02 0.00

kmax

lnC 1 C 2 =C 3 C2

14

For large values of C2k, the exponential term in can be neglected, hence the rst two points are chosen at high slip ratio values. The values of k1 and k2 chosen for simulation are 0.6 and 0.5 respectively.

C1

lk1k2 lk2k1

k2 k1 lk1 lk2 C3 k2 k1

15 16

Now the third point is chosen at a small value of k and is used to evaluate C2. The value of k3 used for simulations is 0.1. A lower value needs to be chosen for k3 to capture the effect of the exponential term in Eq. (12).
C1 ln C 1 lk3C 3 k3

C2

k3

17

where l(k1), l(k2) and l(k3) are coefcient of friction at the respective slip ratios. Using this method, prediction can be done for all surfaces; however the accuracy of prediction depends upon the values of k1, k2 and k3 chosen. Table 4 presents the range of errors in prediction for the above selection of k for all surfaces. Note that this prediction methodology yields better prediction than one point method for all surfaces. The three point method yields signicantly reduced prediction errors (e.g. 400.29% for cobblestone wet and 65.972.4% for cobblestone dry). The only drawback is that the number of cycles used for prediction results in a loss in braking and hence in instances when surface does not change there may be an unnecessary loss of braking while checking for the surface change. The next section deals with the development of controllers used to achieve the desired values of slip ratio.

4. Control algorithms There are two loops for the control algorithm and prediction algorithm as shown in the Fig. 7. The goal of the prediction loop is to converge and gather data at the desired slip ratio set points based on which method is being used. The acceleration values used in the prediction loop during simulations are obtained from the vehicle dynamics model. Switching between the two loops is generally based on time. Initially, the controller starts the prediction algorithm and if the error in slip ratio is less than a desired value, then it predicts the surface and switches to the control loop. Otherwise the prediction loop continues for a predetermined number of cycles and predicts the surface. Once the surface prediction is completed, a new value for kmax is determined. This value is now used as the desired value for the slip ratio in the control loop. Note that the slip ratios in both the loops are determined by solving the system of equations (Eqs. (1)(3))) which govern the vehicle dynamics. The selection of the number of cycles for which the prediction and control loop are executed is very important. The prediction loop should take as fewer cycles as possible. The minimum value of prediction cycles (M) is restricted by the settling time required. A higher number of control cycles ensures that braking efforts are not reduced due to regular predictions. However, a very large number may delay the prediction and even when the surface conditions change it might go undetected for some time. Thus, a number is chosen that results in acceptable loss in braking due to prediction but does not delay the prediction loop by more than a certain time interval. For this braking distance of a car going at 30 m per second when brakes are applied is compared as shown in Table 6. The vehicle used for all the simulations has the following parameters. Based on the results in Table 6, a smaller value of control cycles (N), results in better braking during road change, however, there is a loss in braking when there is no road change. 0.20% Loss in braking is not big trade off as compared to the advantages that come with quick predictions. Therefore, a smaller number of control cycles can be used during braking. Now there is a need to develop control strategies for both prediction loop and control loop. The following owchart represents the entire process. Sliding mode control has been widely used for ABS applications but it demands knowledge of vehicle dynamics. Therefore, it cannot be used for prediction but can be used in the control loop.

188

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Fig. 7. Prediction and control loop.

Table 5 Parameters of the vehicle used for simulations. Symbol G A B ha hf hr mtot ma mf mr Jf Jr R Tmax Quantity Acceleration due to gravity Distance from centre of gravity to front axle Distance from centre of gravity to rear axle Height of sprung mass Height of front unsprung mass Height of rear unsprung mass Total mass of vehicle Sprung mass of the vehicle Front unsprung mass of vehicle Rear unsprung mass of vehicle Moment of inertia of front wheel Moment of inertia of rear wheel Radius of tyre Constraint to the brake torque Value 9.8 m/s2 1.186 m 1.258 m 0.6 m 0.3 m 0.3 m 1500 kg 1285 kg 96 kg 119 kg 1.7 kg m2 1.7 kg m2 0.326 m 8000 N m

4.1. Sliding mode control Sliding mode control, or SMC, is a form of variable structure control (VSC). It is a nonlinear control method that alters the dynamics of a nonlinear system by application of a high-frequency switching control. The state-feedback control law is not a continuous function of time. Instead, it switches from one continuous structure to another based on the current position in the state space. The motion of the system as it slides along these boundaries is called a sliding mode. For the case of controlling the wheel slip, rst a sliding surface is dened

S k kd _ s gs
where g is the convergence factor

18 19

8 > 1 for s > 0 < sgns 0 for s 0 > : 1 for s < 0


_ Also assuming kd is constant, kd 0.

20

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195 Table 6 Effect of control cycles on the braking distance. Value of N Dry asphalt 500 4000 No change 500 4000 Delay between consecutive prediction (s) for 30 m and then wet asphalt 0.07 0.42 0.07 0.42 Braking distance with prediction 43.51 43.57 39.27 39.21 Braking distance without prediction 43.87 43.87 39.19 39.19

189

Loss in braking with prediction (%) 0.83 0.69 0.20 0.05

Now this value is replaced to get the brake torque in terms of sliding mode parameters. There are various ways to reduce chattering; one of them is to introduce the sat function instead of the sgn function. Therefore,

_ s g sats=u
where g is the convergence factor and u is the boundary layer thickness (u > 0)

21

sats=u

sgn s for jSj > u S=u for jSj 6 u

22

The major advantage of sliding mode for ABS applications is that it makes the system robust with respect to system parameter variations, un-modelled dynamics and external disturbances (Fig. 8). The control strategy can be modelled based on the sliding surface dened in the previous section and as given in by Tanelli et al. (2007, p. 593611) .

_ _ _ S k kd
_ _ Here kd is the desired slip ratio, which being constant gives kd 0. As dened earlier in the section,

23

_ k g sats=/
Also,

24

kf

Rxf _ kf

v xf R v  v

25  26

_ _ xf v xf v

_ _ Putting x and v from the half car model into the above equation gives us

0 B Rxf B _ B kf v B @

1 _ T Bf lkf m1 Rg lkf m3 Rv C 2J f xf C C C A
g

27

lkf m1 lkr m2 mtot lkf m3 lkr m3

_ Now, using k g sats=u; T Bf in terms of sliding control parameter is found to be

x T Bf lkf Rm1 m3 2J f g
Similarly,

_ v 2v Jf g sat R v _ v 2v J r g R v

  kd k /   kd k sat /

28

T Br lkr Rm1 m3 2J r xr g x

29

These equations can now be used to control the slip ratio at the predicted values. The variable sliding mode parameters are / and g. These can be tuned to get optimum results. However, unless the surface is not known, applying sliding mode control is not possible as the desired torque value of Eq. (29) depends on the coefcient of friction. Consequently, PID control is used for predicting of surface and settling time is minimized to ensure minimal loss of braking. 4.2. PID control As mentioned previously, a relatively simple PID algorithm is used for predicting the surface upfront and then sliding mode controller will be used. The following graphs give an overview of the settling time and loss in braking due to PID

190

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Fig. 8. Sgn and sat function.

0.2 0.18 0.16

Slip Ratio

0.14 0.12 0.1 0.08 0.06 0.04 10 20 30 40 50


-4
PID Control SMC Control

60

70

80

90

100

Time (*10 Seconds)


Fig. 9. Slip ratio vs. time for PID and SMC control.

Table 7 Comparison of braking and settling time for PID and sliding mode control. PID Braking distance Braking time Settling time 4.553 m 1.006 s 85 cycles SMC 4.373 m 0.9 s 22 cycles

control being used in prediction over SMC control. On the same vehicle parameters, brakes are applied when the vehicle is at a speed of 10 m/s (see Fig. 9 and Table 7). The major disadvantage of PID control is the settling time. Since the prediction can only be done once the slip ratio has settled around a desired value, and the number of cycles needed for PID to settle is much higher than SMC, the prediction time increases and therefore there is a loss in braking. The loss is not only because of the use of PID over SMC as the braking distance here shows. It is also because during prediction, the vehicle is at a slip ratio at which the braking forces available are much lesser as compared to the peak braking forces.

4.3. Parameter selection The parameters for the PID and SMC control loops have been chosen by a manual tuning strategy to achieve a reasonably good performance in terms of rise time and settling time for a step input. For PID tuning, corresponding to each particular value of the proportional gain, the integral and derivative gains were varied in a suitable range and the performance was studied. As mentioned earlier for the sliding mode Eq. (21), g is the convergence factor and u is the boundary layer thickness (u > 0). It was found that similar performance results are obtained for g// values of the same order of magnitude. For higher values of g// (around 50,000) a lot of chattering is observed as the saturation function acts almost like the sgn function. On

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

191

the other hand for lower g// values (around 1000) the saturation function is weak and does not have a noticeable effect on the braking torque to achieve the desired convergence of slip ratio. 5. Results and discussion First, the advantage of ABS over normal braking is shown. Then the results obtained by the proposed methods are compared to the results available in literature for both prediction and overall braking performance to benchmark the performance. 5.1. Comparison of braking performance with and without ABS First, the results of ABS with sliding mode control are compared to a car without ABS. A clear advantage of ABS over locked wheels is visible. When brakes are applied to a car running at 20 m/s, the following results are obtained. As shown in Fig. 10, ABS reduces the braking distance substantially by not allowing the wheels to lock. In the above cases, the braking distance is reduced in a range of 14.5% for wet cobblestone to 30% for wet asphalt. Similarly, it is found that the stopping time is reduced in a range of 16.9% for wet cobblestone to 32.8% for wet asphalt. 5.2. Evaluation of surface identication methods To benchmark the results, the rst task is to compare the surface prediction and/or identication methods. Tanelli et al. (2009, p.891), Savaresi and Tanelli (2010) have presented results for their methods. The prediction has been done on noisy simulation data. Variances have been included in the velocity and wheel speed values to indicate estimation errors and measurement noise respectively. For noisy conditions it was found that a combination of the one point and three point methods performs better as compared to the individual algorithms. The one point prediction is found to be relatively accurate for surfaces with peak slip ratio less than 0.25. In the combination method, the one point prediction is used if both the predictions are below 0.25; otherwise the value predicted by the three point method is used. Figs. 11 and 12 compare the prediction by the combined method in noisy conditions to those obtained in Tanelli et al. (2009, p.891). Figs. 11 and 12 clearly show that the combined method gives good results for most of the surfaces. 5.3. Sensitivity analysis for testing robustness of control algorithms The vehicle parameters that need to be considered for sensitivity analysis are those that can vary during each run or those whose estimated values can be erroneous. The values that have been considered are total mass, height of CG and moment of inertia of wheel which have nominal values of mtot = 1500 kg, ha = 0.6 m, J = 1.8 kg m2 as indicated in Table 5. 10% Variation in each of these parameters has been considered and the effect on stopping distance for each case has been compared. The variance observed in the stopping distances for one point and three point prediction methods have been shown in Table 8. The initial velocity for this analysis is assumed to be 30 m/s. 5.4. Benchmarking the control strategy Results of the models presented in this paper are compared to the results presented by Hari et al. (2008, p. 731741) to benchmark the control strategies. The surface prediction method coupled with SMC control could potentially reduce the braking distance as compared to most of the other ABS strategies. The advantage of prediction methods over other control strategies can be seen from Fig. 13. One point method gives slightly better results (0.64.3%) than three point method since there is a substantial loss in braking during the prediction cycles of three point method and since the road conditions do not change, the accurate prediction

Braking Distance (metres)


Fig. 10. Comparison of braking distance with and without ABS.

192

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Fig. 11. Comparison of slip ratio prediction by different methods.

Fig. 12. Percentage error in slip ratio prediction.

Table 8 Sensitivity analysis based on variation in vehicle parameters. Varying parameter (10%) Variance in distance for each prediction method One point prediction Total mass CG height Wheel moment of inertia 1.6 106 4.9 109 5.13 107 Three point prediction 1.96 104 4.71 106 9.71 105

Distance (metres)
Fig. 13. Comparison of braking with different control strategies used in ABS.

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

193

Fig. 14. Simulation results for a surface change of dry to wet asphalt at 10 m.

Fig. 15. Comparison of braking distance for different ABS methods.

gives no major advantage in case of three point prediction method. Advantages of the proposed prediction methods over self learning fuzzy-sliding mode and neural hybrid network are evident from Fig. 13. However, SMC and genetic adjusted fuzzy method can reduce the braking distance as in case of snow by about 4.3% and 3.6% respectively. They can also increase by 3.4% and 7.3% for the same methods in case of Dry asphalt. However, the real advantage of prediction methods is visible in case of changing road conditions.

5.5. Single surface change during braking Figs. 14 and 15 compare results of one point and three point prediction methods with the method proposed by Hari et al. (2008, p. 731741) when surface changes from dry to wet asphalt, dry concrete to snow and dry to wet cobblestone. When

194

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

Fig. 16. Wheel velocity, vehicle velocity and stopping distance for all surfaces.

Fig. 17. Slip ratio for braking at all surfaces using three point method.

brakes are applied to a vehicle moving at 20 m/s, the results obtained if road condition changes after 10 m of braking can be seen in Figs. 14 and 15. This clearly highlights the advantages of the prediction methods. Since the number of prediction cycles is less in one point method, it gives better results as compared to both three point prediction method and simple SMC control. The stopping distance in case of dry to wet asphalt for three point method is 4.7% more and for dry concrete to snow it is 6.6% more than one point prediction. The same for SMC is 1% more and 3.7% more respectively. However, since the prediction by one point method is not accurate for cobblestone, the braking distance increases in case of one point method. The three point method with its accurate prediction gives better results in case of cobblestone and the stopping distance reduces by 19%. The stopping distance for one point method in the rst two cases is almost the same as that achieved in perfect maximal braking. In the third case, the stopping distance of the three point method is 6.4% more than the distance in perfect braking.

R. Bhandari et al. / Transportation Research Part C 21 (2012) 181195

195

5.6. Braking under changing road conditions Finally, the utility and accuracy of prediction in ABS systems can be highlighted by considering a situation when brakes are applied to a car running at 40 m/s and it is exposed to all road conditions as shown in Figs. 16 and 17. For this simulation, the vehicle under braking rst comes across (1) (2) (3) (4) (5) (6) Wet cobblestone for 10 m. Dry concrete for the next 20 m. Dry asphalt for the next 20 m. Snow for 10 m. Wet asphalt for 10 m. Dry cobblestone till it stops.

Such a simulation has been used to test the function of the three point prediction method under any change in condition it encounters. The results obtained show the robustness of the prediction method. The prediction starts for cobblestone wet gives 0.57% error in estimation of peak slip ratio. The error in estimation of peak slip ratio of the subsequent surfaces is 0.25% for dry concrete, 0.24% of dry asphalt, 7% for snow, 6.8% for wet cobblestone and 2.5% for dry cobblestone. Even under such changes, the accuracy of prediction does not suffer much as compared to the original single surface change predictions by the three point method. As shown in Fig. 16, the stopping distance under such conditions for a vehicle running at 40 m/s is just 98 m. 6. Conclusions From the above results it is clear that for off-road conditions, three point method is suitable and one point method is appropriate for normal driving conditions. A combination of both with the ability to switch as per the drivers requirements can also be used and would be the best as it can have the advantages of fast prediction (one point method) in normal driving conditions and better prediction (three point method) in off road driving condition. Following specic conclusions can be drawn from the current work:  Three point method give best predictions with an average error of around 2% for the surfaces used during simulation. By any other method, the average error in prediction for all surfaces is more than 6%.  One point method is optimized for working with normal road conditions that are regularly faced. The prediction for such surfaces by one point method also gives an average error of around 2% and it reduces the braking distance by 15% as compared to any other method.  One point method has serious shortcomings for unconventional roads. For such conditions, three point method has been proposed which reduces the braking distance by 19% as compared to one point method when surface changes from dry to wet cobblestone.  A combination of PID control for prediction and SMC control to achieve the predicted slip ratio for maximum friction coefcient, braking distances can be reduced by more than 3% if any surface change is encountered during braking.

References
Anwar, S., 2006. Anti-lock braking control of a hybrid brake-by-wire system. Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering 220 (8), 11011117. Baslamisli, S.C., Kose, I.E., Anlas, G., 2007. Robust control of anti-lock brake system. Vehicle System Dynamics 45 (3), 217232. Burckhardt, M., Reimpell, J., 1993. Fahrwerktechnik, Radschlupf-Regelsysteme, Vogel. Choi, S.B., 2008. Antilock brake system with a continuous wheel slip control to maximize the braking performance and the ride quality. IEEE Transactions on Control Systems Technology 16 (5), 9961003. Hari, A., Aghagolzadeh, A., Alizadeh, G., Sadeghi, M., 2008. Designing a sliding mode controller for slip control of antilock brake systems. Transportation Research Part C: Emerging Technologies 16 (6), 731741. Oniz, Y., Kayacan, E., Kaynak, O., 2007. Simulated and experimental study of antilock braking system using grey sliding mode control. In: IEEE International Conference on Systems, Man and Cybernetics, p. 9095. Oudghiri, M., Chadli, M., Hajjaji, A.E., 2007. Robust fuzzy sliding mode control for antilock braking system. International Journal on Sciences and Techniques of Automatic Control (IJ-STA) 1 (1), 1328. Park, K.-S., Lim, J.-T., 2008. Wheel slip control for ABS with time delay input using feedback linearization and adaptive sliding mode control. Electrical Engineering, 290295. Savaresi, S.M., Tanelli, M., 2010. Active Braking Control System Design for Vehicles. 1st ed.. Springer-Verlag. Tanelli, M., Piroddi, L., Savaresi, S.M., 2009. Real-time identication of tireroad friction conditions. IET Control Theory Applications 3 (7), 891. Tanelli, M., Sartori, R., Savaresi, M.S., Poussat-Vassal, C., 2007. Combining slip and deceleration control for brake-by-wire control systems: a sliding-mode approach. European Journal of Control 6 (1), 593611. Zhang, X., Lou, Y., Yang, X., 2008. Study of control logic for automobile anti-lock braking system. In: 2008 International Conference on Intelligent Computation Technology and Automation ICICTA, p. 493497.

You might also like