An Iterative Mode Merging Method For Multimode Process Monitoring With Transition Mode

An industrial process typically consists of multiple production modes, and transitions between two modes often cause difficulties for process monitoring. For the transitions between steady-state modes, it is common to ignore or combine them with other modes for modeling. The former is likely to cause false alarms, and the latter reduces the detection rate of the monitoring process. Efficient and accurate demarcation of the transition mode from the steady mode presents an intriguing research topic. To address this issue, this paper presents an innovative iterative mode merging approach for multimode process monitoring with transition mode. The data is segmented according to the proportion of principal components to the total space, and then iterative mode merging (the merging of data of the same mode) is performed according to the reconstruction error of the probabilistic principal component analysis (PPCA) model. A case study using a multiphase flow dataset demonstrates the effectiveness of the proposed method.