Multimode Anomaly Detection Under the Multiscale Framework

Multiscale characteristic of fault signals determines the necessity of multiscale analysis in multimode anomaly detection. Firstly, mode identification is implemented. Then, multiscale MSPCA detection modeling and multiscale time-varying rolling balls anomaly detection model is, respectively, established to each steady state and transition mode. Simulation result shows that the multimode monitoring method under multiscale framework can significantly reduce the missing detection rate of critical faults, which can effectively avoid major accidents.