Machine Anomaly Detection under Changing Working Condition with Syncretic Self-Regression Auto-Encoder

Condition monitoring is one of the key tasks for the intelligent maintenance of high-end equipment. Facing the challenge of its changing working conditions, intelligent monitoring models that are built upon constant working conditions are not qualified for this task. To solve this problem, a syncretic self-regression variational auto-encoder (SSR-VAE) model is proposed to realize the parallel training of distribution learning and regression learning for machine anomaly detection. Among them, self-regression learning plays an auxiliary role in distribution learning. Furthermore, multi-sensor information fusion at the decision level is implemented to improve the robustness of the proposed model. The effectiveness of this model is evaluated on a gearbox test platform under changing working conditions.