Fault Detection of Rolling Bearings Using an Ensemble Average Autocorrelation Based Stochastic Subspace Identification

Rolling bearings are the crucial parts of rotating machines. The detection and diagnoses of their defects at early stages are significant for ensuring safety and efficient operations. Usually, the vibration feature associated with bearing faults are submerged by the heavy background noise and nonstationary impacts. To enhance detection performance, this pa-per proposes a novel method developed based on ensemble average autocorrelation and stochastic subspace identification (SSI) techniques. It establishes the theoretical basis of the method based on the general characteristics of bearing vibration signals under faults. Then it examines the robustness of techniques under different level noise, which leads to an optimal selection of centre frequencies that have high signal to noise ratio and thereby high accuracy of envelope analysis for fault diagnosis. Both simulation and experimental results show that the proposed method is able to extract bearing fault signatures at very low signal to noise ratio (<-20dB) and consequently produces accurate detection.