Abstract
Friction of the vehicle to the road surface is a key issue in road safety, and many existing and future driver´s assistance systems expect friction to be known. Friction potential cannot be evaluated by any driver sensitive feeling, nor monitored on the dashboard with a single sensor.
This paper proposes an original approach to estimate friction potential, initiated in ARCOS 2004, and going-on in SARI, two French research programs associating private and public structures, in the PREDIT framework.
As friction can vary with many parameters (road surface and wear, water height, tires type and condition...), on-board Real Time estimation appears as the only efficient way. ESP/ABS systems lead to a proper estimation of friction, but only during heavy braking or accelerating phases. Our objective was to estimate friction also while low acceleration level driving.
Based on monitoring the effects of friction, i.e. slip, the method is called "model matching": an on-board model of the car (CALLAS) is computed in real time in any driving situation, and one continuously compares the real car response, provided by physical sensors, with the model response, which constitutes the friction estimation reference. The error between real slip and model slip is then processed by a fuzzy logic system. The basic principle can be summarized as follows: "If the vehicle slips more than the model, friction is over estimated and one gradually decreases it, until slip error is reduced to zero". So the model friction estimation converges toward the vehicle/road friction.
The first stage was on-table simulation and states that basic principle was both simple and natural, and that the results were very good. The second stage was the implementation into a real car. If the system requires only about 16 sensors, the laboratory car has over 100 sensors, including dynamometer wheels, to trace all the system variables, and to compare the friction estimation with the part of available friction really used. The real world application exhibits complexity and comes up against a three-fold difficulty:
- The model and its parameters must be very accurate in order to avoid systematic errors,
- Since several causes may lead to similar consequences, one must separate effects due to friction variation from those coming from road roughness, banking and longitudinal slope, wind, load condition...
- Sensors embedded in a road vehicle require complex signal conditioning.
However the technical feasibility of this method was established and the system appears as very powerful. The friction monitoring potentially works in all directions (longitudinal, lateral and coupled) with 16 sensors, but presently the best results are obtained in the longitudinal direction and minimal levels of acceleration (2 m/s²) and speed (15 km/h) are required. Improvements have still to be done: in the present state of development, the prototype requires a top-of-the-range PC and costly sensors, the response time is between 0.3 and two seconds, and the cover rate has still to be enlarged.
This approach has required a close collaboration between several experts: vehicle dynamics, applied mathematics, vehicle instrumentation and road testing. The paper describes the method, implementation in the car, and a selection of results. The potential evolutions of this ambitious application and requirements for industrialization are also discussed.
Keywords - friction, vehicle dynamics, simulation, driver assistance