A selective fuzzy ARTMAP ensemble and its application to the fault diagnosis of rolling element bearing

Abstract

A novel intelligent fault diagnosis method based on feature extraction methods, features selection using modified distance discriminant technique and selective ensemble of multiple fuzzy ARTMAP (FAM) classifiers is proposed in this paper. The method consists of three stages. Firstly, different features in multiple symptom domains, such as time-domain features, frequency-domain features, wavelet grey moments, wavelet packet energy spectrum and auto-regression model parameters, are extracted from the raw vibration signals. Secondly, with the modified distance discriminant technique five salient feature sets are selected from the five original feature sets in different symptom domains respectively. Finally, these optimal feature sets are input the selective ensemble of multiple FAM classifiers based on the correlation measure method and Bayesian belief method to identify different abnormal cases. The proposed method is applied to the fault diagnosis of rolling element bearings, the test result shows that the selective ensemble of four FAM classifiers can identify the different fault conditions accurately and has a better classification performance compared to the single FAM and ensemble of all FAM classifiers. Besides, the diagnosis performance of the selective ensemble is analyzed by the bootstrap method. All experiment results have demonstrated that the selective ensemble of FAM classifiers has the effectiveness, stability, generalization, reliability and robustness.