Advanced bearing fault detection at varying rotational speeds using PSO-optimized SVM and CDET feature selection

Abstract In this exploration, we introduce a novel intelligent hybrid methodology for multiclass bearing fault detection under varying rotational speeds. Unlike previous studies, our approach combines empirical mode decomposition (EMD), wavelet packet decomposition (WPD), and a comprehensive attribute retrieval strategy from both time and frequency domains. Critically, we present an enhanced distance compensation evaluation technique (CDET) for feature selection, in which the threshold parameter is automatically optimized using particle swarm optimization (PSO) rather than being set arbitrarily. Furthermore, the parameters of both CDET and support vector machine (SVM) classifiers are jointly optimized by PSO, resulting in enhanced classification accuracy and computational efficiency. Extensive experiments on benchmark databases demonstrate that the recommended tactic surpasses conventional tactics regarding fault detection accuracy, stability, and feature compactness. From the original signals, 20 statistical features in the time domain and 6 in the frequency domain were extracted, along with 4 decomposition levels from WPD and the first 6 components from EMD. The outcomes demonstrate that specific features can effectively differentiate various bearing conditions at different speeds, highlighting the efficacy of the recommended approach compared to other fault detection methods.