Research on Feature Extraction Methods in Bearing Fault Diagnosis Algorithms

Abstract As a key component in the motor system, the stability and reliability of the entire system is directly influenced by the operating condition of the bearings. With the improvement of industrial automation, the importance of diagnosis of bearing faults technology is becoming more and more prominent. In this paper, for the diagnosis problem of bearing fault, the diagnostic efficacy of three feature extraction methods, namely, time-domain (TD), Discrete Wavelet Transform (DWT), and Fast Fourier Transform (FFT), is systematically investigated. For these three feature extraction methods, this paper proposes six fault diagnosis algorithms, which are decision tree (DT), random forest (RF), support vector machine (SVM), convolutional neural network (CNN), ensemble learning (EL), and extreme learning machine (ELM). The results indicate that the DWT-based feature extraction method is the best overall, achieving a diagnostic accuracy of 99.79% under the EL; the time-domain features provide stable diagnostic results, and the RF achieves an accuracy of 98.83%; whereas the FFT-based features obtain an accuracy of 97.58% under the ELM. Further analyses show that different feature extraction methods affect different algorithms to different degrees, and there is a significant difference in the sensitivity of the algorithms to the feature extraction methods: the traditional machine learning algorithm performs well with DWT features, while the deep learning method obtains better results with time-domain features. The research results provide a theoretical basis for the optimisation of bearing fault diagnosis algorithms, and at the same time provide principle support for the selection of systematic feature extraction methods.