Neural network prediction of disc brake performance

Abstract

An automotive brake's performance results from the complex interrelated phenomena occurring at the contact of the friction pair. These complex braking phenomena are mostly affected by the tribochemical properties of the friction material's ingredients, the brake disc properties, and the brake's operating regimes. In this paper, the synergistic effects of the friction material's properties, defined by its composition and manufacturing conditions, and the brake's operating regimes on the disc brake factor C variation have been modelled by means of artificial neural networks. The influences of 26 input parameters, determined by the friction material composition (18 ingredients), its manufacturing conditions (5 parameters), and the brake's operating regimes (3 parameters) on the brake factor C variation, have been predicted. The neural model of the disc brake cold performance has been developed by training 18 different neural network architectures with the five different learning algorithms. The optimal neural model of disc brake operation has been shown to be valid for predicting the brake factor C variation of the cold disc brake over a wide range of brake's operating regimes and for different types of friction material.