基于对抗熵的转子系统跨工况故障诊断方法^*

Traditional data-driven fault diagnosis methods often rely on the availability of test condition data, but the actual operating conditions of rotor system are changeable, and the prior data distribution of test condition is difficult to obtain, which increase the difficulty of fault diagnosis across different working conditions. To solve this problem, an adversarial entropy-based domain generalization network (AEDG) is proposed for fault diagnosis of rotor system. Inspired by information bottleneck theory and generative adversarial network, this method achieves the antagonistic disturbance of potential data distribution through minimax entropy, which aims at improving the generalization ability of diagnostic model under unknown conditions. First, through multi-linear mapping fusion of deep embedding feature and the prediction output of classifier, the conditional adversarial domain adaptation network is established to realize the deep fusion of multi-source domain diagnosis knowledge. To further improve the generalization performance of the model under unknown working conditions, the entropy of prediction output of multi-source joint embedding features was minimized to realize the disturbance of the underlying data, which enhances the adaptability to the distribution shift under unknown working conditions. Finally, two fault datasets of rotor system are used to verify the effectiveness of the proposed method, and the results show that the proposed method has good identification accuracy and generalization ability across different working conditions.