Life grade recognition of rotating machinery based on Supervised Orthogonal Linear Local Tangent Space Alignment and Optimal Supervised Fuzzy C-Means Clustering

Abstract

A novel life grade recognition method based on Supervised Orthogonal Linear Local Tangent Space Alignment (SOLLTSA) and Optimal Supervised Fuzzy C-Means Clustering (OSFCM) is proposed for rotating machinery in this paper. Firstly, the time–frequency feature parameter sets are constructed to completely extract the features of different life grades. Then, SOLLTSA is proposed to compress the time–frequency feature parameter sets of testing and training samples into low-dimensional eigenvectors with clearer clustering. Finally, the low-dimensional eigenvectors of testing and training samples are put into OSFCM to realize life grade recognition. SOLLTSA not only combines the local geometry with class information for manifold decoupling, but also solves the optimal low dimensional embedding subspace by spectral regression and subspace orthonormalization approach, thus improving the life grade feature extraction power of LLTSA. Meanwhile, OSFCM defines an optimized objective function that adopts the average distance measure between training and testing samples to lead the clustering process, and further applies training samples to partition matrix initialization for raising pattern recognition efficiency and avoiding local minimum. This allows OSFCM to have a higher life grade recognition accuracy than Fuzzy C-Means Clustering (FCM) does. A life grade recognition example for deep groove ball bearings demonstrates that the proposed method is effective in life grade recognition.