A Stochastic Resonance Based Signal Reinforcement Approach for Early Fault Diagnosis of Rolling Bearings

The early identification of bearing defects has recently attracted increasing attention in the fields of condition monitoring due to the critical role of a bearing on the reliability and safety of turbomachines. The weak features representing early fault in the vibration signals are often submerged in environmental noise, which poses a major challenge for early fault monitoring of the bearing. This paper proposes an improved negative entropy of square envelope spectrum approach integrated with stochastic resonance-based signal enhancement for accurate early defect detection of bearings. This proposed method takes into account the cycle-stationarity, impulsivity and similarity of the original signal via signal enhancement approach, thus reinforcing characteristic frequency while integrating the regularity of the original signal in evaluating the stochastic resonance performance. The genetic algorithm associated with the improved negative entropy of square envelope spectrum metric is developed to optimize the parameters in stochastic resonance model, in order to enhance the potential components representing bearing defects in the original signal, thus improving defect detection accuracy. An experimental example is employed to illustrate the effectiveness and accuracy of the proposed method for defect detections of rolling element bearings in inner race. After using this method, the fault information in the noisy signal is dominant.