Multimode process identification and monitoring based on hierarchical fluctuation window strategy

Abstract

Modern industrial processes generally have multiple operation modes, and the transition between modes cannot be ignored. Accurate mode division and effective monitoring models are crucial. This paper proposes a multimode process identification and monitoring based on a hierarchical fluctuation window strategy (HFWS). Firstly, Gaussian mixture model (GMM) is adopted to identify mode information roughly. Secondly, HFWS is used to identify stable modes and transition modes accurately. HFWS includes two parts: the initial level and the terminal level. In the initial level, neighborhood characteristics of samples are divided by the sliding window, and key windows are selected and transmitted to the terminal level. Next, variable fluctuation is introduced to accurately determine the boundary of transition modes in the terminal level. Thirdly, stable modes are monitored by GMM based on Mahalanobis distance. Finally, a new neighborhood shift model is established to monitor transition modes. The proposed HFWS can accurately divide stable modes and transition modes from the perspective of samples and variables. Additionally, it builds appropriate monitoring models for stable modes and transition modes, respectively, which reduces the monitoring complexity and improves the fault detection rate. The effectiveness of HFWS is verified by a numerical case and a new standard Tennessee Eastman process.