Abstract: An approach to gear fault diagnosis is presented, which bases on kernel principal component analysis (KPCA). In this approach, the integral operator kernel functions is used to realize the nonlinear map from the raw feature space of gear vibration signals to high dimensional feature space. By performing PCA on the high dimensional feature sets, the nonlinear principal components of raw feature space are obtained. In succession, the selected nonlinear principal components are used to construct the feature subspace for classification of gearbox working conditions. The experimental data sets of gearbox working under three conditions： normal, tooth cracked and tooth broken are used to test the KPCA based method. The classification effect of KPCA based method is compared with that of PCA based method. The results indicate that the method can perform gear crack detection efficiently and can fulfill fault classification accurately, and it is more suitable for nonlinear feature extraction from fault signals.

Key words: Feature extraction, Fault diagnosis, Kernel principal component analysis, Pattern classification