Robust Online Monitoring for Multimode Processes Based on Nonlinear External Analysis

Abstract

A robust online monitoring approach based on nonlinear external analysis is proposed for monitoring multimode processes. External analysis was previously proposed to distinguish faults from normal changes in operating conditions. However, linear external analysis may not function well in nonlinear processes, in which correlations between external variables and main variables are generally nonlinear. Under the consideration of real-time monitoring, a moving window is used for sample selection and least-squares support vector regression is used as the model structure of nonlinear external analysis. When the influence of external variables is removed, the filtered information of main variables is extracted and monitored by a two-step independent component analysis−principal component analysis strategy. In addition, to improve the performance of modeling and monitoring, a robust scheme is developed. A benchmark study of the Tennessee Eastman process shows the efficiency of the proposed method.