Diagnosis of Incipient Fault Based on Sliding-Scale Resampling Strategy and Improved Deep Autoencoder

Currently, the fault diagnosis with balanced data and distinct characteristics has received mass concern, and the related research achievements are remarkable. However, because of the weakness and scarcity of incipient fault signals, the diagnosis of incipient fault commonly existing in industrial systems is still an intractable problem. In order to solve the problem, an incipient fault diagnosis method based on a sliding-scale resampling strategy and improved sparse autoencoder with multi-particle noise addition (MpNA-SAE) is proposed in this paper. Firstly, the original time domain signals are preprocessed, and a sliding-scale resampling strategy is designed to construct the balanced sample sets. Secondly, a multi-particle noise addition strategy and an adaptive loss function are designed, and then an improved sparse autoencoder with multi-particle noise addition (MpNA-SAE) fault diagnosis model is constructed to identify the fault pattern and determine the severity degree. Thirdly, a diagnostic performance evaluation criterion is proposed to quantify the application range of the model. Finally, the effectiveness and practicability of the proposed method are verified by incipient artificial damage and real fault experiments, respectively.