Open-Set Fault Diagnosis Method for Industrial Process Based on Semi-supervised Learning

Aiming at the inconsistent distribution of labeled and unlabeled data categories in the actual industrial production process, this paper proposes an openset semi-supervised process fault diagnosis method based on uncertainty distribution alignment. Firstly, the proposed method forces the matching of the distribution of labeled data and unlabeled data. Then it combines a semi-supervised fault diagnosis model with the anomaly detection of one-vs-all classifier. The interior point (unlabeled samples in known class) is correctly classified while rejecting outliers to realize the fault diagnosis of open-set industrial process data. Finally, fault diagnosis experiments are carried out through numerical simulation and Tennessee-Eastman chemical process to verify the effectiveness and feasibility of the proposed method. Compared with temporal ensembling-dual student (TE-DS) and other semi-supervised fault diagnosis methods, it is proved that the proposed method is suitable for open-set fault diagnosis.