Estimation of brake force on an open wheel racing car using artificial neural networks

The nature of automobile dynamics is complex. While they might not be aware of it, the driver of a vehicle is making many complex decisions producing a complex series of actions that effect the motion of the vehicle. Usually the driver can perform a sequence of actions which move the vehicle in a way in which the driver intends, however, occasionally all drivers find themselves having to correct the vehicle in a way that they did not expect. The problem here is in the control system and the high performance available in the vehicles that are driven on the roads. With the brake pedal linked directly to the force on the brake disks, the driver of the vehicle simply applies pressure that corresponds to the rate at which they intend to stop. At the limits of tyre adhesion this breaks down as more brake pressure fails the slow the vehicle quicker, and the vehicle actually takes longer to stop. To produce safer vehicles, car developers and manufacturers have developed anti-lock brakes and stability control systems. These state of the art systems monitor driver commands that inherently reflect their intention and the behaviour of the vehicle. When the vehicle behaves in a way that does not follow the driver's intent the system intervenes and selectively applies braking, limits engine power or changes other relevant parameters to assist the driver in retaining control of the vehicle. Systems such as these use mathematical models based on simplified assumptions of vehicle behaviour. Because of this they are built to be robust, commercially available systems fail to capitalise on the full performance potential of the vehicle. Since systems such as these become active in emergency situations, every small gain in performance can make up the difference between life and death. Neural networks, as emerging decision making tools offer another approach to the problem of modelling non-linear dynamic, multi variable vehicle physics. Neural networks use artificial intelligence to find relationships between inputs and outputs. These relationships are not assumed or based upon a simplified physical analysis, but are built based on the past experiences of the network. For automobile dynamic prediction a special vehicle 'Intelligent Car' was conceived, constructed and tested in real world driving conditions. Structured driving tests were carried out gathering sufficient data to train ant test the potential of neural networks for the application. The results from these tests represent some of the first outcomes from preliminary research in the 'Intelligent Car' Project. This study outlines the design and development of the University of Tasmania's Intelligent Car together with results from training various neural network models to predict its brake forces and comparison with measured values