An Adaptive Feature Enhancement Method Based on Optimized Spectrum Period Estimation and Sparse Bayesian for Bearing Fault Identification

Under complex working conditions, it is difficult to accurately obtain the fault feature period of bearings in complex equipment for fault diagnosis. Period estimation based on peak detection and feature parameters is prone to high false detection rates due to periodic interference and random pulse effects, which pose a challenge for fault diagnosis. Therefore, this paper proposes an adaptive feature enhancement method based on optimized spectrum period estimation and sparse Bayesian for bearing fault identification (OSPES). First, adaptive weighted spectrum optimization (AWSO) calculates the weights of fault data spectrum lines based on healthy data to obtain a weighted optimized spectrum (WOS), thereby suppressing the impact of periodic interference components. Second, low-amplitude components are removed from the WOS and harmonic product spectrum analysis is performed to suppress the effects of random pulses and noise, and obtain the fault pulse period. Finally, a sparse Bayesian learning (SBL) framework incorporating period constraints and group sparsity constraints is employed to reconstruct sparse signals and achieve adaptive feature enhancement. OSPES suppresses periodic and random interference based on the characteristics of fault data and impact sparsity, thereby improving the problem of inaccurate period estimation that arises from peak detection and feature parameters. It also eliminates the need for manual parameter adjustment and enhances universality. Simulation, vibration, and acoustic signal experiments, as well as comparative experiments, verify the effectiveness and superiority of OSPES. A Periodic-to-Interference ratio (PIR) is designed to evaluate the superiority of OSPES quantitatively.