Early diagnosis of bearing faults using decomposition and reconstruction stochastic resonance system

Weak vibration signals generated by faults in rolling element bearings are often masked by high levels of background noise, which leads to inaccurate fault detection. To overcome the masking, a stochastic resonance based on decomposition and reconstruction method is employed for early bearing fault diagnosis. First, theoretical analysis is performed to confirm the requirement for signal preprocessing. Subsequently, a novel cross correlation kurtosis is defined as an index to obtain an optimal intrinsic mode function after empirical mode decomposition. Finally, the preprocessed signal is clarified via a stochastic resonance method and reconstructed for bearing fault diagnosis. Combined with ensemble bagged trees-based machine learning, the decomposition and reconstruction stochastic resonance method was beneficial for detecting fault signals from incoming noisy vibration signals using simulated and actual vibration signals. According to the results, the proposed method was superior to traditional stochastic resonance method and variational mode decomposition with the k-number method.