Abstract: A novel fault diagnosis method based on feature compression with normalized Laplacian-based supervised optimal locality preserving projection (NL-SOLPP) is proposed. The time-frequency domain feature set is first constructed to completely characterize the property of each fault. NL-SOLPP is introduced to automatically compress the high-dimensional time-frequency domain feature sets of training and test samples into the low-dimensional eigenvectors which have better discrimination. The low-dimensional eigenvectors of training and test samples are input into Shannon wavelet support vector machine (SWSVM) to carry out fault identification. NL-SOLPP considers both local information and class labels in designing the similarity weight matrix and requires the output basis vectors to be statistically uncorrelated and orthogonal, therefore, it achieves higher fault identification accuracy. Fault diagnosis example on deep groove ball bearings and life state identification example on one type of space bearing demonstrated the effectivity of proposed method.

Key words: Dimension reduction, Fault identification, Locality preserving projection, Manifold learning, Time-frequency domain feature set