Research on sparsity indexes for fault diagnosis of rotating machinery

This paper originated from an investigation of sparsity indexes for fault diagnosis of rotating machinery. Although various sparsity indexes have been widely applied in machinery fault feature extraction, there is little information on the guideline available for the selection of the best sparsity index for the specified scenarios with different interferences. To solve the problem, this article firstly analyzes the performance of the representative sparsity indexes, containing Gini index, l(2)/l(1) norm, Hoyer measure and kurtosis. Aiming at the feature of the machinery fault signal, three performance attributes, including data-length independency, random-impulse resistance and fault-impulse discernibility, are originally proposed to quantitatively evaluate the sparsity index. Based on the comparison results, a guideline for the selection of the optimal sparsity measure is summarized. After that, this guideline is used for the improvement of kurtogram and protrugram, and the results are evaluated. Finally, the comparison result, using both simulated and experimental bearing fault signals, confirms that an optimal scheme can be designed for the sparsity-based improvement under the proposed guideline. (C) 2020 Elsevier Ltd. All rights reserved.