An identification and prediction model of wear-out fault based on oil monitoring data using PSO-SVM method

This paper builds an identification and prediction model of wear-out faults with particle swarm optimization (PSO) in support vector machine (SVM) model. The empirical data indicates the wear-out faults, which take great proportion in fault types of diesel engine. The identification and prediction of wear-out status in time are significant to enhance reliability of diesel engine and reduce economic loss. Oil monitoring data through spectral analysis technology can reflect the wear-out status, but the data analysis is deficient and susceptible to external factors. Given the small sample size of the wear-out faults, the SVM model is proposed to identify and predict wear-out fault in this paper. Redundant attributes in primary oil monitoring data are reduced before establishing the model. In order to enhance the identification and prediction accuracy of wear-out fault, parameters are optimized through particle swarm optimization (PSO) and features are selected by recursive feature elimination process (RFE). In case experiment, the model using PSO-SVM can achieve greater accuracy than original model and grid-search optimization SVM.