“…Therefore, after acquiring the bearing signal feature matrix, it is necessary to be further extracted to remove the irrelevant and redundant features. At present, the methods to remove redundant and irrelevant features of bearing include auto-encoder [ 23 , 24 ], neural networks [ 25 , 26 ], Principal Component Analysis (PCA) [ 27 ], kernel PCA [ 28 ], and Singular Value Decomposition (SVD). However, although intelligent algorithms, such as self-encoding and neural networks, have been applied to diagnose bearing faults, they have disadvantages, such as low generalization, slow calculation speed, and higher requirements for hardware equipment.…”