An approach for mechanical fault classification based on generalized discriminant analysis

Abstract

To deal with pattern classification of complicated mechanical faults, an approach to multi-faults classification based on generalized discriminant analysis is presented. Compared with linear discriminant analysis (LDA), generalized discriminant analysis (GDA), one of nonlinear discriminant analysis methods, is more suitable for classifying the linear non-separable problem. The connection and difference between KPCA (Kernel Principal Component Analysis) and GDA is discussed. KPCA is good at detection of machine abnormality while GDA performs well in multi-faults classification based on the collection of historical faults symptoms. When the proposed method is applied to air compressor condition classification and gear fault classification, an excellent performance in complicated multi-faults classification is presented.