Wavelet-based compressed sensing for motor bearing fault diagnosis

Abstract

Motor bearings are key components of rotating machinery that operate under high-speed, heavy-load, and strong noise environments for a long time. Early fault signals are weak and easily disturbed. Traditional compressed sensing methods have high requirements for dictionary design, low reconstruction accuracy, poor signal recovery, and slow computation. Pure artificial intelligence algorithms rely heavily on massive data, with complex models and poor real-time performance. To solve these problems, a wavelet-based compressed sensing method for motor bearing fault diagnosis is proposed in this paper. It combines the advantages of wavelet multi-scale time-frequency analysis and compressed sensing low-rate sampling to build a complete fault diagnosis framework. A multi-scale feature enhancement model is established to amplify fault impact components. An adaptive regularized sparse reconstruction model is designed to protect key fault features. A global joint optimization model is constructed to reduce the dependence on dictionary design and decrease reconstruction errors. Experiments on the HUST bearing dataset under various speeds and fault types show that the proposed method achieves high-precision signal recovery with better RMSE, PRD, and SNR than traditional methods. It has low sampling demand, strong anti-interference ability, and good engineering practicability for online monitoring and intelligent diagnosis of early weak faults in high-speed heavy-duty motor bearings.