Fault Detection of Rolling Bearing Using Sparse Representation-Based Adjacent Signal Difference

The early defects in bearings tend to result in periodic impacts and, thus, cause impulse components in vibration signals. However, these fault features are always submerged and distorted by the heavy noise. A new method called sparse regularization (SR)-based adjacent signal difference (SR-ASD) is proposed to extract fault features from bearing vibration signals. Impulsive sparse retention term and adjacent difference sparse retention term are proposed to extract the impulsive features based on the characteristic of the fault signal. They are capable of preserving the impulsive features and eliminating noise by retaining the sparsity of the signal and its difference, respectively. The adaptive setup of regularization parameters is developed to enhance the effect of noise removal according to the characteristics of the vibration signal. Majorization-minimization (MM) is adopted to solve the optimization problem of the objective in SR-ASD. Experimental results on simulation signals and bearing vibration signals illustrate that SR-ASD can effectively eliminate noise interference and then extract fault features of vibration signals. SR-ASD outperforms other typical methods on fault feature extraction from vibration signals.