Weak Fault Detection for Rolling Bearings in Varying Working Conditions through the Second-Order Stochastic Resonance Method with Barrier Height Optimization

The stochastic resonance (SR) method is widely applied to fault feature extraction of rotary machines, which is capable of improving the weak fault detection performance by energy transformation through the potential well function. The potential well functions are mostly set fixed to reduce computational complexity, and the SR methods with fixed potential well parameters have better performances in stable working conditions. When the fault frequency changes in variable working conditions, the signal processing effect becomes different with fixed parameters, leading to errors in fault detection. In this paper, an underdamped second-order adaptive general variable-scale stochastic resonance (USAGVSR) method with potential well parameters’ optimization is put forward. For input signals with different fault frequencies, the potential well parameters related to the barrier height are figured out and optimized through the ant colony algorithm. On this basis, further optimization is carried out on undamped factor and step size for better fault detection performance. Cases with diverse fault types and in different working conditions are studied, and the performance of the proposed method is validated through experiments. The results testify that this method has better performances of weak fault feature extraction and can accurately identify different fault types in the input signals. The method proves to be effective in the weak fault extraction and classification and has a good application prospect in rolling bearings’ fault feature recognition.