FEATURE EXTRACTION AND PATTERN RECOGNITION ON THE FRICTIONAL FAULTS OF DISC BRAKE

The frictional state of disc brake reflects and determines its braking performance directly. Therefore, the intelligent recognition of frictional faults according to frictional states must be quite valuable for improving its reliability. Firstly, in this paper a characteristic parameter set for characterizing the frictional faults of disc brake was extracted from its dynamic frictional features. The set contains two subsets: friction coefficient subset, temperature rise subset, which is composed of 10 fault character parameters. Then the frictional faults were defined as four fault patterns: normal, symptomatic, slight and serious fault. Secondly, by taking the braking of automobiles as an engineering background, the data samples of frictional faults were acquired by simulated braking experiments. Finally, a pattern recognition model for the frictional faults of disc brake was established by artificial neural network (ANN) technology. The input layer of the ANN model is the characteristic parameter set of frictional faults, while the output layer is the faults patterns. What is more, the fault pattern recognition experiments were performed with the built model. It is concluded that the characteristic parameter set and frictional fault patterns, established in this paper, can extract and recognize intelligently all kinds of frictional faults on disc brake. After learning the complex nonlinear mapping relationships between the fault patterns and characteristic parameters, which are contained in experimental data samples, the ANN can recognize exactly the frictional fault patterns of disc brakes.