Fast general normalized convolutional sparse filtering via L1-L2 mixed norm for rotating machinery fault diagnosis

Abstract

Intelligent fault diagnosis in mechanical condition monitoring has emerged in recent years. Consequently, the training efficiency and diagnostic accuracy of fault diagnosis are urgent research topics for real applications. On this basis, this paper proposes fast general normalized convolutional sparse filtering (FGNC-SF) via the L1-L2 mixed norm for intelligent fault diagnosis. The contributions of this paper are as follows. The L1-L2 mix norm is used to penalize the objective function of the algorithm, generalized normalization is used to achieve the row and column normalization of sparse filtering, convolutional activation is used to improve the diagnostic efficiency, and pseudo-normalization is used to improve the test feature distribution. The L1-L2 mixed norm can have two different functions in the algorithm by adjusting different normalization parameters. Generalized normalization no longer limits the normalization parameters to specific values, the characteristic of the proposed approach is illustrated for different normalization parameters. Our method is characterized under different normalization parameters. The selection of optimal parameters is studied in terms of the diagnostic accuracy, the computing time and the standard deviation. The proposed FGNC-SF is validated through two collected rolling bearing datasets. Results show that FGNC-SF exhibits a strong learning ability and is superior to the existing methods for rotating machinery fault diagnosis, obviously improved diagnosis accuracy, efficiency and robustness, reduces the need of priori knowledge and makes intelligent fault diagnosis handle big data more easily.