Domain Adversarial Enhanced Multi-Channel Graph Networks for Aeroengine Gas Path Fault Diagnosis

The gas path system is an important part of aero-engine, and its fault will bring terrible suffering, therefore the research on fault diagnosis method of the gas path has always been a concern. However, the existing diagnosis methods have the problems of low diagnostic accuracy and poor adaptability to performance degradation. This paper proposes the domain adversarial enhanced multi-channel graph networks to improve the above shortages. In this method, the spatiotemporal series is embedded by multi-head attention enhanced temporal and spatial graph networks, combining domain adversarial training supplemented by spectral norm regularization. On the one hand, this solution can fully mine the spatiotemporal relationship of multivariate sequences in aeroengine gas path, on the other hand, it can improve the difference in data distribution and model generalization. According to the comparison experiment, the proposed model performs better fault feature extraction ability and domain transfer ability, which demonstrates its potential application in gas path fault diagnosis.