Adaptive Multiscale Noise Tuning Stochastic Resonance for Health Diagnosis of Rolling Element Bearings

The analysis of vibration or acoustic signals is most widely used in the health diagnosis of rolling element bearings. One of the main challenges for vibration or acoustic bearing diagnosis is that the weak signature of incipient defects is generally swamped by severe surrounding noise in the acquired signals. This problem can be solved by the stochastic resonance (SR) approach, which is to enhance the desired signal by the aid of noise. This paper presents an adaptive multiscale noise tuning SR (AMSTSR) for effective and efficient fault identification of rolling element bearings. A new criterion, called weighted power spectrum kurtosis (WPSK), is proposed as the optimization index without prior knowledge of the bearing fault condition. The WPSK concerns both the kurtosis in signal power spectrum and the similarity to a sinusoidal signal in signal waveform, thus it can balance the enhancement of possible characteristic frequency in the frequency domain and the regularity of the signal in the time domain for the SR performance. Two parameters in the AMSTSR, including the cutoff wavelet decomposition level and the tuning parameter, are simultaneously optimized based on the WPSK index through the artificial fish swarm algorithm. The AMSTSR is further applied to the health diagnosis of rolling element bearings and four experimental case studies verify the effectiveness of the proposed method in adaptive identification of the bearing characteristic frequencies.