Feature Extraction And Pattern Recognition Of Friction Faults For Disc Brake

Disc brake is the most important machinery safety device, which affects theproductivity and personal safety directly. And, the brake friction state is the most directexpression of braking performance. So the monitoring of brake friction state is animportant basis for brake faults diagnosis.Currently, the brake faults diagnoses mainlyfocus on the monitoring or analysis of non-friction state like hydraulic system, brake shoeclearance or idle motion time, etc. And the studies of brake friction performance areconfined to the static friction parameters of friction coefficient and wear rate, whichcannot reflect the dynamic friction characteristics of braking process accurately.Therefore, the study of brake dynamic friction feature extraction method and the researchof feature extraction and pattern recognition of friction faults for disc brake can provide atheoretical basis for faults diagnosis and prediction of friction brake, which will haveimportant theoretical and practical value for improving braking performance and itsreliability.This paper was funded by the National Natural Science Foundation of China. Withdisc brakes as the research object, the dynamic features of brake friction statusparameters were extracted, and the brake friction faults feature extraction and patternrecognition methods which based on the dynamic friction characteristics parameters werestudied. Firstly, in view of the deficiency of the existing brake disc brake simulation testequipment, the disc brake simulation test rig under the project background of automotivedisc brake was designed and built, and also the brake friction test which based on the testrig. Secondly, multiple data processing methods like statistical theory, data fitting andwavelet analysis were used to analysis and transform the friction state parametersmathematically. The dynamic friction state parameters (including the end-point of frictionfactor surge, friction factor trend coefficient, the mean coefficient of friction, the frictionfactor stability factor, the energy coefficient of friction factor, the average surfacetemperature of the friction and the friction surface maximum temperature, etc.) wereextracted, and the extraction methods of dynamic characteristic parameter of the brakefriction state were established. Later, the brake friction tests were carried out to study theaffection of braking condition (initial braking speed and brake pressure) to the dynamiccharacteristic parameters of friction state. Finally, the brake friction fault was defined andclassified, and the friction fault characteristic parameters which based on the brake friction state dynamic characteristic parameter extraction method were extracted. The BPneural network model for friction faults pattern recognition was established, and faultspattern recognition software system was written by VB and MATLAB. The results ofpattern recognition of friction faults simulation test shows that BP neural network hasgood recognition effect on friction faults. Researches showed as followings: the dynamicfriction parameters set of features extracted from this paper could characterize the frictionstate comprehensively and objectively; the friction faults features which based on thedynamic friction parameters extraction could be used as an index of brake friction faultspattern recognition and the correction of friction faults pattern recognition system thatestablished on the BP neural network was high.