Braking torque control using recurrent neural networks

The basic problem in the operation of automotive brakes is the unpredictable nature of the tribological processes that occur at the contact of the friction pair. The stochastic nature of the tribological contact of the disc brake is affected differently by the complex interaction between the brake disc and the friction material under different conditions because of the influences of the applied pressure, the speed and the brake interface temperature. Owing to the highly dynamic non-linear change in the braking torque induced by the complex situation at the contact of the disc brake, the braking torque could not be modelled, predicted and controlled using classical mathematical methods. This is related, in particular, to the dynamic change in the braking torque in a braking cycle. Dynamic modelling and prediction of the braking torque is very important for further improvement in the performance of the brakes of motor vehicles through more precise control of their performance with respect to the driver demands and the change in the adhesion between the tyre and the road. Recurrent dynamic neural networks were employed in this paper for modelling, prediction and control of the dynamic change in the braking torque during a braking cycle. The dynamic functional relationship between the changes in the applied pressure, the sliding speed, the brake interface temperature and the braking torque of the disc brake was established. The dynamic model developed was used to predict and control the braking torque during a braking cycle under different disc brake operation conditions.