Quality-Relevant Process Monitoring Based on Improved Concurrent Canonical Correlation Analysis

How to use existing technology to improve product quality while ensuring production stability and safety has become a hot topic in various fields. Being one of the most powerful multivariate analysis tools in process monitoring, canonical correlation analysis (CCA) is an exploratory statistical method that emphasizes the correlation between a set of poly dimensional data obtained at the same experimental level. For overcoming the drawbacks of CCA, a concurrent canonical correlation analysis (CCCA) modeling method was put forward, which decomposes the input and output variables into five segments. However, CCCA based quality-related process monitoring scheme would make ambiguous decisions due to the decomposition is incomplete. It is intended that this paper will explore an improved concurrent canonical correlation analysis (ICCCA) technique used for process monitoring. Specifically, a CCA model is established where the latent structure of process data is extracted. Subsequently, an algorithm of total principal component regression (TPCR) is implemented grounded on the established CCA model. Complete decomposition of the input and output variables into six parts was performed for quality-related process monitoring. Taking the Tennessee Eastman process (TEP) as an industry benchmark, the superiority of ICCCA is verified by comparing it with other monitoring methods.