An Adaptive Data Decoupling Model for Rotating Machinery Multiple Faults Diagnosis

Single-point fault diagnosis of rotating machinery is a matter of great concern at present, and the mature study findings are fruitful in this field. However, since the interrelationship of subunits and the faults features correlation, the multiple faults diagnosis for rotating machinery commonly existing in actual industrial systems is still an intractable problem. Therefore, in order to solve the multiple faults diagnosis problem, an adaptive data decoupling model is proposed, and then, a solution programme for rotating machinery multiple faults diagnosis is constructed in this paper. Specifically, 1) A novel loss function and an initial weights constraint are formulated to improve the feature learning diversity of sparse autoencoder. 2) A decoupling constraint term is constructed to reduce the harmful influence of data coupling. Finally, comparison tests and trial results analysis are developed to verify the effectiveness and practicability of this proposed model.