General normalized sparse filtering: A novel unsupervised learning method for rotating machinery fault diagnosis

摘要

In the era of data deluge, "big data" generated by mechanical equipment creates higher requirements for the field of mechanical fault diagnosis. Intelligent diagnosis methods have become a new focus for researchers. Deep learning-based intelligent diagnosis method for health monitoring is a promising tool that can get rid of the dependence of prior signal processing knowledge and diagnostic experience. In this paper, a novel unsupervised learning method called general normalized sparse filtering (GNSF) is proposed for intelligent fault diagnosis. The proposed algorithm realizes the feature sparsity measurement by optimizing the objective function based on the generalized l(r-p/q) norm of the feature matrix. The characteristic of the generalized l(r-p/q) norm as sparsity measure is discussed. Based on GNSF, a new intelligent fault diagnosis method is designed and applied on rolling bearing and planetary gear fault diagnosis under complex working conditions. The verification results confirm that the proposed method is a promising tool that can obtain higher diagnosis efficiency and accuracy with fewer training samples and has much better robustness than the previous methods. (C) 2019 Published by Elsevier Ltd.