VEHICLE ROLLOVER IN OFF-ROAD CONDITIONS

Transcription

1 VEHICLE ROLLOVER IN OFF-ROAD CONDITIONS Georg Rill, Holger Faisst, Johannes Hautmann, Rudolf Ertlmeier University of Applied Sciences Regensburg, Galgenbergstr. 3, 9353 Regensburg, Germany Continental, Chassis & Safety, Siemensstr. 1, 9355 Regensburg, Germany University of Applied Sciences Ingolstadt, Esplanade 1, 8549 Ingolstadt, Germany address of lead author: Abstract Almost 45 people died in road accidents in Germany in 8. According to NHTSA statistics about one third of the fatalities are caused by rollover accidents. Therefore, there is still a huge potential to save lives by improving the vehicles rollover safety systems. To approach this goal, a multibody system simulation for soil trip rollover is set up and validated with a series of measured soil trip events of a scaled rollover test vehicle. Preliminary results will be shown in this paper. 1 INTRODUCTION Besides the obvious need for improved rollover avoidance by active safety systems, it is important to increase the passive safety for rollover crashes, which includes the improvement of crash detection for these types of accidents. Within Continental the project ContiGuard R embraces the networking of active and passive safety systems and the integration of environmental sensors and telematics for best possible protection of road users. This includes the functional improvement of the airbag control system by using driving dynamics sensor signals that come with an electronic stability control (ESC). Processing these chassis signals within a vehicle state observer, the vehicle s lateral dynamics is available as pre-crash information to the so-called Active Rollover Detection algorithm [1]. The crash severity can then be predicted in an earlier phase of the accident leading to up to 5% faster deployment times of restraints systems like airbags, belt tensioners, or rollover bars. This translates into a benefit for occupant safety which is most prominent in soil trip rollover situations, where the vehicle leaves the road and begins to roll when the tires laterally dig into soft soil. This soil trip rollover situation, which is the topic of this paper, is particularly challenging as the dynamics of the initial crash phase is rather close to sportive driving dynamics. Therefore, a deeper understanding of the dynamics of the soil trip is necessary. SCALED TEST VEHICLE The scaled test vehicle shown in Fig. 1 is based on a 1:5 scaled commercial car model [1]. The chassis was modified in order to match the layout of Sport-Utility-Vehicles (SUV). By providing a steering system, detailed double wishbone suspension systems and some sort of pneumatic tires it represents a realistic copy of a real vehicle. Figure 1: Scaled test vehicle with rollover chassis and detail of the front suspension

2 A differential gear distributes the drive torque generated by an electric motor equally to the rear wheels. The vehicle is operated by a remote control acting on the steering system and the electric motor. The model is equipped with sensors measuring the wheel speeds, the steering angle as well as the accelerations and the angular velocities of the chassis. 3 SIMULATION MODEL 3.1 Model Structure The multibody system approach is most appropriate to investigate the handling properties of vehicles by computer simulations, [5] and [6]. As the topology of the scaled test vehicle corresponds with the standard layout of a real car, the detailed multibody vehicle model shown in Fig. will represent the scaled test vehicle as well as a real vehicle. Figure : SIMPACK Simulation model The multibody software (MBS) package SIMPACK [8] was used to generate the simulation model. SIMPACK offers several templates for various vehicle substructures, for example a double wishbone suspension, the steering system, or the drive line. Here the simulation model consists of 4 parts and has 16 degrees of freedom (DOF). The chassis is represented by one rigid body. The double wishbone suspension system for the front and the rear suspension are modeled by rigid links. The standard double wishbone suspension consists of a wheel plate and two control arms. Ball-joints attach the control arms to the chassis and to the wheel plate. The scaled test vehicle is also equipped with coil springs and pneumatic dampers which are modeled by nonlinear force elements. The anti-roll bar is approximated by two rigid links attached at both ends of a torsion bar and connected via drop links to the lower control arms. The substructure wheel was supplemented with the TMeasy tire model [] which provides an internal as well as an external interface to SIMPACK. The steering system consists of servo-driven lever arms which transmit the steering motion via steering rods to the wheels. The steering angle is controlled by pre-defined input functions. The driving velocity of the simulation model is controlled by the driving torque applied to the drive shaft. A differential transmits the driving torque via half-shafts to the wheels. Within the simulation model the drive torque is applied directly to the wheels by drive line. The drive line is a mechanism with two DOF. It consists of an input shaft, a differential box, and two output shafts left and right. 3. Model Data All these structures are fully parameterized, so the simulation vehicle can easily be adopted to different vehicles. The geometry and the mass and inertia properties of the scaled test vehicle with respect to the center of gravity (CoG) were carefully measured. The characteristics of the small coil springs and the pneumatic damper elements were recorded separately on a small test rig. The main data of the scaled test vehicle are compared in Table 1 with a downsized data set of a typical SUV. According to the geometric scaling factor of 1:5 the factors 5 3 and 5 5 have to be used to scale the mass and the inertias, respectively.

3 scaled test vehicle real vehicle (SUV) downsized values Wheel base [m] /5 =.564 Track width [m] /5 =.37 Height CoG [m] /5 =.14 Mass [kg] /5 3 = Inertia (long.) [kgm ] /5 5 =.364 (lat.) [kgm ] /5 5 = (vert.) [kgm ] /5 5 = 1.44 Table 1: Main data of the scaled test vehicle compared to the corresponding downsized values of a real vehicle Nearly all data of the scaled test vehicle match quite well with the corresponding downsized values of a real vehicle. Only the inertias around the lateral and the vertical vehicle axis differ somehow. This is caused by a more compact mass distribution in the scaled test vehicle. However, the inertia of the scaled test vehicle around the longitudinal axis which has an significant influence to rollover motions match perfectly with the corresponding downsized value. The risk of rollover may be judged by the Static Stability Factor s F = s/ h where h denotes the height of the CoG and s names the track width. Values of s F 1 indicate a high rollover risk on paved dry roads. The static stability factors for the scaled test vehicle and a real vehicle Scaled Test Vehicle: s F =.31/.136 = 1.18 ; Real Vehicle: s F = 1.535/.7 = 1.96 calculated with the data given in Table 1 differ only a little. In both cases the factor is slightly larger than one which grants stability on dry paved roads on one hand but on the other hand will involve a risk to rollover in off-road conditions. 4 TIRE/ROAD MODEL 4.1 TMeasy Tire Model The TMeasy tire model was successfully applied to many different tires ranging from large agricultural tires over heavy truck tires to passenger car tires. The test vehicle is equipped with small pneumatic tires which although not inflated and not reinforced by steel, rayon or nylon should have a typical tire performance. So, there was a confident hope that these tires can be described by classical semi-physical tire models. TMeasy is a typical handling tire model, [3]. It provides all forces and torques generated in the contact patch of the tire. The normal force or wheel load is provided by a static and a dynamic part F z = F S z + F D z () The static part Fz S is described as a nonlinear function of the tire deflection and the dynamic part Fz D is roughly approximated by a damping force proportional to the time derivative of the tire deflection. Because the tire can only apply pressure forces to the road the normal force is restricted to F z. The longitudinal force as a function of the longitudinal slip F x = F x (s x ) and the lateral force depending on the lateral slip F y = F y (s y ) are defined by characteristic parameters, Fig. 3: the initial inclinations df x, dfy, the locations s M x, sy M and the magnitude of the maximum forces Fx M, Fy M as well as the sliding limits s S x, s S y and the sliding forces F S x, Fy S. These data can easily be assigned to particular tires even if no or only few measurements are available, [4]. The force characteristics of the tires of the scaled test vehicle are defined by the two sets of data shown in Table. Here the subscripts 1 and indicate the data provided for the payload F z = Fz N = 4 N and the doubled payload F z = Fz N = 8 N. Thus, the influence of the wheel load to the tire forces and torques is taken into account. The tires of the scaled test vehicle have large bristles and consist of genuine rubber. So, the traction coefficient in lateral direction was set to a rather low value, µ =.875. At the pay load of Fz N = 4 N this will result in a maximum lateral force of Fy M = N = 35 N. Furthermore a strong degressive influence of the wheel load to the (1)

4 F x F x M F x S df x s x M s x S s x F y s y df F M s M F S ss s ϕ s x F(s) F x s y S s y S F y M s y F y M df F y y Figure 3: Longitudinal, lateral and generalized tire characteristics DFY_1 = 75. $ initial FZ_N [N] FYMAX_1= 35. $ max FZ_N [N] SYMAX_1=.15 $ sy where fy(sy)=fymax_1 FYSLD_1= 34. $ sliding FZ_N [N] SYSLD_1=.6 $ sy where fy(sy)=fysld_1 DFY_ =1. $ initial *FZ_N [N] FYMAX_= 6. $ max *FZ_N [N] SYMAX_=.18 $ sy where fy(sy)=fymax_ FYSLD_= 59. $ sliding *FZ_N [N] SYSLD_=.7 $ sy where fy(sy)=fysld_ Table : TMeasy model data for the lateral tire force characteristics of the scaled test vehicle lateral force characteristic was assumed, which reduces the traction coefficient in lateral direction from µ =.875 to µ = 6/( 4) =.75 at double the payload. The TMeasy model approach requires an initial inclination (cornering stiffness) of the lateral tire characteristics which is defined by dfy min = Fy M /sy M. As the slip value where the lateral force reaches its maximum was set to sy M =.15 the minimum possible cornering stiffness will be given by dfy min = 35 N/.15 = 433 N. The chosen value of df y = 75 N is slightly larger and hence will represent a rather soft tire performance. Similar data sets are provided for the longitudinal tire force characteristics. By combining the longitudinal and lateral slip to a generalized slip s the combined force characteristic F = F(s) can be automatically generated by the characteristic tire parameter in longitudinal and lateral direction, Fig. 3. The self-aligning torque is approximated via the pneumatic trail which again is described by a characteristic parameter. The influence of the camber angle to the lateral tire force and the self-aligning torque is modeled by an equivalent lateral slip and by a bore torque which is generated by the component of the wheel rotation around an axis perpendicular to the local track plane. The TMeasy model data are completed by geometric data as well as stiffness and damping properties of the tire. By taking the tire deformation into account the TMeasy approach to steady state tire forces can easily be extended to dynamic tire forces, [7]. 4. Soft Soil Extension A soft soil environment is characterized by a significant soil deformation and additional forces caused by the strength of the soil, Fig.4. a) b) z R c z S S d S z S F Sx v x FT M T F Sy v y Figure 4: Soft soil approach: a) deformation, b) forces

5 In a first approach the soil deformation is simply described by a linear spring damper element. If F z names the wheel load, then F z = c S z S + d S ż S (3) will hold, where z S denotes the soil deformation and the constants c S and d S characterize the stiffness and damping properties of the soil. Finally, the force balance (3) results in a first order differential equation d S ż S = F z c S z S (4) According to () the actual wheel load F z is calculated within TMeasy via the tire deflection and its derivatives which result from the momentary position and velocity of the wheel center and the road height. By replacing the road height z = z R with z = z R z S the soft soil deformation z S is automatically taken into account. If the wheel is sunken and moves in longitudinal and lateral direction, then additional forces F S x and F S y caused by the strength of the soil will certainly occur. A precise model of the interaction between a tire and soft soil will be very complicated and will therefore lengthen the simulation time extremely. Again, a quite simple approach was used. The strength of the soil is characterized by a sheer friction number µ S which depends linearly on the soil deformation z S and is defined for the payload F z = F zn by the value µ S N. Then, the maximum sheer force acting at the wheel modeled by c S z S F S = µ S N F z = µ S (z S ) F z (5) F zn is equivalent to Coulomb s dry friction law. The forces in longitudinal and lateral direction are finally obtained from v x v y F S x = F S and F S y = F S (6) v x + v y v x + v y where v x and v y denote the components of the contact point velocity in longitudinal and lateral direction. The working points of the longitudinal and lateral components of the soil force were assumed at 1 z S and 3 z S, respectively. 5 VERIFICATION 5.1 On-Road Performance In practice, standard driving maneuvers like steady state cornering or a step steer input are used to judge the dynamics of a vehicle on road. As the step steer input is an open-loop maneuver it was used here to verify the data of the simulation model. Via the remote control the scaled test vehicle was at first accelerated on a dry paved surface to a pre-defined velocity and then, a step-like steering input was performed. In order to get results which cover the whole range of lateral acceleration, different vehicle velocities were combined with various steering angles. For the verification the measured yaw rate ω z and lateral acceleration a y were compared with simulation results, Fig. 5. On real vehicles an ideal step input is not possible. Even if the input, in this case the switch of a remote control, is changed all of a sudden, the dynamics of the steering servo (on a real vehicle the dynamics of the power-assistance) and the inertia the steering system will cause a somewhat slower and smoothed steering motion at the wheels. SIMPACK offers a smoothed step-function where the input changes in a given time-interval without discontinuities from one value to another. For the left turn the input at the steering servo was changed from δ = to δ = 5 and the right turn was performed by changing the steering input from δ = to δ = 19. The vehicle velocity was kept for the left turn at v =.63 m/s and for the right turn at v = 3.75 m/s respectively. The tests were performed on a parking lot where the road surface had a small inclination which causes the scaled test vehicle to slightly accelerate or decelerate up and down the hill. This causes periodic variations in the steady state test results. The simulation results show a good conformity to the measurements in a wide range of driving situations, Fig 6. Even the first data guess of the tire characteristics of the scaled test vehicle was quite acceptable. 5. Rollover Test Laboratory tests for soil trip rollover are typically carried out on so called flying floors. The vehicle is positioned on a horizontal sled, which is accelerated laterally towards the crash target, like, for instance, a sand bed, which the vehicle will hit at a defined speed. These tests focus on the pure crash phase, thus ignoring the previous history of the driving and skidding situation. In turn they have highest reproducibility which is why they are used for the validation of the multibody system simulation.

6 ω z [rad/s].5 left turn a y [m/s ] ω z [rad/s] right turn a y [m/s ] t [s] t [s] 8 1 Figure 5: Measurements (thin solid line) and simulation results (thick broken line) ω z [rad/s] v [m/s] a y [m/s ] v [m/s] Figure 6: Steady state results: measurements, simulation The movie frames in Fig. 7 illustrate a typical rollover scenario. The sled is hereby driven with a speed of v = 4 km/h. At first the wheels are nearly sliding over the surface. Due to the compliance of the soft soil the vehicle starts to roll. Then, the outer wheels dig into the ground thus generating additional lateral forces which cause the vehicle to roll more and decelerate. Finally, the lateral movement of the outer wheel comes close to a stop forcing the vehicle to roll over. If the velocity of the sled is reduced to v=km/h no rollover occurs, Fig. 8. The whole manoeuver takes longer because the backwards roll motion of the vehicle is now included. As the vehicle is still close to a rollover situation it is more sensitive to the soft soil condition and hence to the parameters of the soft soil model. The simulation results can be judged not only by inspecting the screen shots in Figs. 7 and 8, but also by comparing the measured and computed time histories of the roll rate ω x shown in Fig. 9. The simulation also provides the roll angle, Fig. 1. In case of rollover the simulation results match very well with the measurements. In the simulation the contact elements at the chassis interact with the solid ground and not with the soft soil. That is why the roll angle α is restricted to 9 and the roll rate ω x drops down a little bit earlier compared to the measurements where of course the chassis of the scaled test vehicle contacts the soft soil. Reducing the sled velocity from v = 4km/h to v = km/h still brings the vehicle close to a rollover situation. In the simulation the maximum roll angle of α occurs at t 7ms, Fig. 1. It coincides with the zero-crossing of the roll rate ω x in Fig. 9. At the end of the simulation the roll angle remains at a small negative value because the backwards motion of the vehicle is finally stopped when the wheels at the right side of the vehicle come in contact to the soft soil again. The momentum of the backwards rolling vehicle results in a soil compression which then is slightly larger on the right than on the left side of the vehicle.

7 t = 1 ms t = 5 ms t = 1 ms t = 15 ms t = ms t = 5 ms t = 3 ms t = 35 ms t = 4 ms t = 45 ms t = 5 ms t = 55 ms Figure 7: Movie frames of a rollover test-rig experiment and screen-shots of animated simulation results t = 1 ms t = 6 ms t = 1 ms t = 7 ms t = ms t = 8 ms t = 3 ms t = 9 ms t = 4 ms t = 1 ms t = 5 ms t = 11 ms Figure 8: Movie frames of test-rig experiment without rollover and screen-shots of animated simulation results It turns out that this situation is more delicate to reproduce by the simulation model. That is why the simulation results match a little less with the measurements. A first analysis indicates that the additional soft soil force which at present is modeled just as a function of the soil deformation may also depend on the sliding velocity. The investigation of additional tests where the vehicle is orientated no longer perpendicular to the sled motion should bring more insight and the chance to improve the soft soil model.

8 8 roll-over at v=4 km/h 4 no roll-over at v= km/h 6 ωx [rad/s] 4 ωx [rad/s] t [ms] t [ms] Figure 9: Time history of the roll rate ω x ( Measurements, Simulation) 9 α [ o ] 6 roll-over at v=4 km/h 9 α [ o ] 6 no roll-over at v= km/h t [ms] t [ms] Figure 1: Simulated roll angles 6 CONCLUSION At first measurements and computer calculations on a paved dry surface were performed in order to verify the simulation model data with the scaled test vehicle. Then, the tire model TMeasy was extended by a soft soil model where the soil deformation and additional soil forces were taken into account. Although the impact of a soft soil surface to the tire was modeled quite simply, the comparison between test-rig experiments with the scaled test vehicle and corresponding simulation results show a good conformity. Further improvements in the soft soil model such as a nonlinear soil deformation, and soil forces which also depend on the sliding velocity should result in a model which is able to simulate manouevers with and without rollover in reasonable quality. It will then be possible to do systematic variations of the parameters that define the macroscopic behavior of the soft soil and the initial conditions of the vehicle. The findings from these investigations will be integrated into Continental s next generation Active Rollover Detection algorithm. References [1] R. Ertlmeier, H. Faisst, T. Kiefer. Active Crash Detection, to appear in: 14. Internationaler Kongress Elektronik im Kraftfahrzeug, Baden-Baden, Germany, Oktober 9. [] W. Hirschberg, G. Rill, H. Weinfurter, User-Appropriate Tyre-Modeling for Vehicle Dynamics in Standard and Limit Situations, Vehicle System Dynamics, Vol. 38, No., pp , Lisse: Swets & Zeitlinger,. [3] W. Hirschberg, G. Rill, H. Weinfurter, Tyre Model TMeasy, Vehicle System Dynamics, Volume 45, Issue S1 7, pages [4] W. Hirschberg, F. Palèák, G. Rill, J. Łotník, Reliable Vehicle Dynamics Simulation in Spite of Uncertain Input Data, in: Proceedings of 1 th EAEC European Automotive Congress, Bratislava, 9. [5] R. N. Jazar. Vehicle Dynamics, Theory and Applications. Springer, New York, 8.

9 [6] G. Rill. Simulation von Kraftfahrzeugen (in German). Vieweg, Braunschweig, Reprint: rig39165 (called ). [7] G. Rill, First Order Tire Dynamics, in: Proceedings of the III European Conference on Computational Mechanics Solids, Structures and Coupled Problems in Engineering, Lisbon, Portugal, 58 June 6. [8] URL: (called ), Webpage of Intec GmbH.