Fault diagnosis for small samples based on attention mechanism

Abstract

Aiming at the application of deep learning in fault diagnosis, mechanical rotating equipment components are prone to failure under complex working environment, and the industrial big data suffers from limited labeled samples, different working conditions and noises. In order to explore the problems above, a small sample fault diagnosis method is proposed based on dual path convolution with attention mechanism (DCA) and Bidirectional Gated Recurrent Unit (DCA-BiGRU), whose performance can be effectively mined by the latest regularization training strategies. BiGRU is utilized to realize spatiotemporal feature fusion, where vibration signal fused features with attention weight are extracted by DCA. Besides, global average pooling (GAP) is applied to dimension reduction and fault diagnosis. It is indicated that DCA-BiGRU has exceptional capacities of generalization and robustness by experiments, and can effectively carry out diagnosis under various complicated situations.