Quality monitoring method based on enhanced canonical component analysis

In continuous processes, the quality variables generally can be interpreted by the process variables due to intercorrelation. However, in particular condition, the past quality trends may be responsible for interpretation due to the auto-correlation. The existing methods only reveal one of the correlations. Considering the effects of two types of correlations for quality monitoring, this study develops enhanced canonical component analysis (ECCoA) method based on Canonical Correlation Analysis (CCA). For revealing the intercorrelation, CCA is performed to extract the quality related features from the process variables. However, the components of CCA ignore the variance formation in the data. To retain both cross-data (process variables and quality variables) correlation information and the variance information within process variables, principle projective-CCA (PP-CCA) method is proposed, generating the primary feature subspace to capture the variation of quality variables. Moreover, as for the auto-correlation, on the residual obtained in PP-CCA method, a residual-CCA (R-CCA) method is proposed for modelling and generating the complementary feature subspace, reflecting the trends of quality variables. Sequentially, statistical indexes and decision-making logic are established for online monitoring. A numerical case and the Tennessee Eastman process are tested for validation. The achieved results indicate the feasibility and efficiency of the proposed enhanced canonical component analysis method.