Self-adaptive deep learning for multimode process monitoring

Abstract

Automatic process monitoring is important for loss prevention in modern process industries. However, due to variations in equipment, organization, market and environment, the only constant in the processes is change. One chemical process generally experiences multiple operating modes. Process monitoring models built based on one mode might demonstrate high false alarm rate and therefore become useless in actual industrial applications. That is one of the root causes that have prevented many advanced process monitoring technologies from being widely accepted by the chemical industry. To overcome this issue, some multimode methods based on statistical models have been developed in the past twenty years. However, most of them can only be used for the modes that already exist in history. If the process operating condition shifts to a new mode that has never appeared before, the monitoring model inevitably fails. In this paper, a novel self-adaptive deep learning method based on local adaptive standardization and variational auto-encoder bidirectional long short-term memory (LAS-VB) is proposed for multimode process monitoring. In order to monitor the process under new operating modes, local adaptive standardization (LAS) is used to preprocess local moving window data and variational auto-encoder bidirectional long short-term memory (VAE-BiLSTM) is trained to detect the unstable deviation in the local moving window. Finally, the benchmark Tennessee Eastman process (TEP) is utilized to verify and illustrate the feasibility and efficiency of the proposed method. Compared with other multimode methods, our method shows its outstanding performance on the process monitoring task for new operating modes that never occurred in the process history.