Intelligent Fault Diagnosis Using Limited Data Under Different Working Conditions Based on SEflow Model and Data Augmentation

Accurate fault diagnosis of machine components is quite important for normal operation of equipment. Nowadays, artificial intelligent methods have been widely researched in fault diagnosis of rolling element bearings (REB). However, due to the variation of machine working conditions, the diagnosis accuracy always degrade seriously. Besides, as it is really hard to achieve large amounts of labeled health condition signals from real equipment, data deficiency is another trouble. Both issues impede the practical application of data-driven fault diagnosis. So as to solve the problems, a data augmentation method SEflow based on squeeze-and-excitation networks (SEnet) and flow-generative model is proposed. Proposed SEflow can learn the data distributions from limited data, then generate augmented signals among different machine working conditions. The experiments applied on bearing datasets and ball screw signals verify the effectiveness of proposed method on solving domain adaption and data deficiency.