Fault Diagnosis Method for Rolling Bearings Based on Wavelet Packet Analysis and Adaboost.M2 Algorithm

Abstract

Aiming at the precise diagnosis requirements for multiple types of rolling bearing faults and reducing the impact of bearing faults on the operational performance of mechanical equipment, an intelligent fault diagnosis method combining wavelet packet transform (WPT) energy feature extraction and AdaBoost.M2 is proposed. First, four-layer wavelet packet decomposition is performed on bearing vibration signals to extract full-band normalized energy features, which can fully mine the time-frequency transient feature vectors of non-stationary vibration signals induced by single and composite faults. Then, the feature vector is fed into the ensemble learning model to identify eight typical faults including inner race, outer race, and rolling element coupled faults, and the adaptive weight updating mechanism is adopted to enhance the model's ability to distinguish easily confused fault samples. Experimental results show that the overall diagnosis accuracy of this method reaches 99.15%, and it maintains stable recognition performance under small sample conditions, providing an effective and reliable technical solution for online fault monitoring and intelligent diagnosis of rotating machinery.