Multimode Process Monitoring Based on Aligned Mixture Factor Analysis

Abstract

To meet the demand for diversified products, industrial processes typically have multiple operating modes. In this article, a novel scheme based on aligned mixture factor analysis (AMFA) is proposed for multimode process monitoring. First, mixture factor analysis (MFA) is applied to produce a statistical fingerprint of the training data that gives a detailed description of the multiple operating modes. Then, unlike conventional multimodal algorithms in which multiple models are constructed and monitoring results are softly combined, the proposed method aims at aligning the separated local models together and performing the monitoring behaviors based on the global model. Through this approach of dividing and integrating, both the within-mode and cross-mode correlations can be greatly preserved in the global model. Finally, the utility and feasibility of the proposed method are demonstrated through a numerical example, a nonisothermal continuous stirred-tank reactor (CSTR) model, and the Tennessee Eastman benchmark process.