Recognition of life state for slewing bearings using probabilistic principal component analysis

A novel multi-domain feature extraction approach based on probabilistic principal component analysis (PPCA) is proposed to deal with the weak fault feature of slewing bearings. Several feature vectors are extracted to form a feature set with high dimension. Then the vectors that best reflect the slewing bearing life status are extracted from the feature set by applying PPCA. These vectors are then used as inputs of a support vector machine with particle swarm optimization to perform the life state recognition. It follows from the whole life experiment of slewing bearing that PPCA is better than the traditional PCA in reducing feature dimension, and its recognition accuracy of lifetime state increases by about 8%. A comparison with a single feature or single domain features shows that the multi-domain and multi-feature set reflects the degradation of slewing bearings more comprehensively and accurately. And a comparison with the traditional feature-extraction method shows that the proposed method reflects the fault of the slewing bearing that is running in a complex and harsh environment more effectively, thus, it can be applied in the area of slewing bearing health monitoring.