An Improved DANN-based Mixed Gas Booster Station Fault Diagnosis Method

A by-product gas booster station plays an important role in the steel production process, however, due to environmental and other factors can occur jump machine failure, affecting the normal steel production. This study proposes a fault diagnosis method based on one improved domain adversarial neural network (IDANN) for the by-product gas booster station. The method considers the influence of hidden variable factors such as discharge gas pressure and resonance on fault identification and designs an improved domain-adaptive network structure. Assessing the impact of the implicit variable injection ratio on the recognition accuracy of the model, the inter-class distance is adopted to optimize the parameters of the injected implicit variable factor ratio to maximize the inter-class distance. To verify the effectiveness in this study, the operating data of a steel plant LDG booster station is selected for experiments and compared with deep neural networks (DNN) and other fault diagnosis methods. The experimental results show that the recognition rate of this paper can reach 95% for the jump machine faults and the method of this paper has good robustness and network generalization ability.