Unified discriminant manifold learning for rotating machinery fault diagnosis

Fault diagnosis is an important technology for performing intelligent manufacturing. To simultaneously maintain high manufacturing quality and low failure rate for manufacturing systems, it is of great value to accurately locate the fault element, evaluate the fault severity and find the fault root cause. In order to effectively and accurately perform fault diagnosis for rotating machinery, a novel feature selection method named unified discriminant manifold learning (UDML) is proposed in this research. To be specific, the local linear relationship, the distance between adjacent points, the intra-class and inter-class variance are unified in UDML. Based on these, the local structure, global information and label information of high-dimensional features are effectively preserved by UDML. Through this dimension reduction method, homogeneous features become more concentrated while heterogeneous features become more distant. Consequently, mechanical faults could be diagnosed accurately with the help of proposed UDML. More importantly, local linear embedding algorithm, locality preserving projections algorithm, and linear discriminant analysis algorithm could be regarded as a special form of UDML. Moreover, a novel weighted neighborhood graph is constructed to effectively reduce the interference of outliers and noise. The corresponding model parameters are dynamically adjusted by the gray wolf optimization algorithm to find a subspace that discovers the intrinsic manifold structure for classification tasks. Based on the above innovations, a fault diagnosis method for rotating machinery is proposed. Through experimental verifications and comparisons with several classical feature selection algorithms, rotating machinery fault diagnosis can be more accurately performed by the proposed method.