Abstract:

In mechanical fault diagnosis, to address the problem that the weak fault features extracted by traditional methods are easily disturbed by strong background noise and have a low accuracy, a data-driven sparse features extraction method using the generalized minimax-concave penalty is developed. This method constructs an effective sparse optimization objective function for mechanical fault diagnosis in order to improve the accuracy of fault feature extraction. It is also proved that the non-convex controllable parameters can guarantee the overall convexity of the objective function under certain constraints. The proximal algorithm is used to solve the unconstrained optimization problem. In addition, the data-driven regularization parameter setting criteria are studied to ensure that the proposed sparse feature extraction method has parameter adaptability. Finally, simulation results and practical fault experiment verify the effectiveness of the proposed method in the machinery fault diagnosis.

Key words: machinery fault diagnosis, concave penalty, sparsity enhancement, parametric adaptation