Nonlinear Feature Fusion Based on Kernel Fisher Discriminant Analysis for Machine Condition Monitoring

Abstract

A nonlinearly discriminant feature extraction and fusion scheme is proposed to recognize the different condition of mechanical faults. As most running statuses of machines are nonlinear and non-stationary, it is difficult to extract the effective features for fault diagnosis by linear feature extractor such as principal component analysis (PCA) and Fisher linear discriminant analysis (FLD). Therefore, kernel Fisher discriminant analysis (kernel FDA) is used to extract the nonlinear discriminant feature. In addition, the idea of feature fusion is introduced to improve the robustness of feature extractor, because feature fusion can synthesize complementary information of different feature variables from multi-signal sources. The heavy computational burden induced by the tremendous size of the feature space is also effectively settled by using a kernel function in the input space without explicit computation of the mapping in feature space, and the influence of the parameter in kernel function on recognition performance is also discussed. The results show that the proposed feature fusion scheme can effectively improve the robustness of feature extractor, and nonlinearly discriminant features perform better than kernel PCA features.