Canonical Variate Analysis Based Regression for Monitoring of Process Correlation Structure

In the process of practical application, it is found that the typical method of process monitoring using the process data covariance matrix cannot effectively monitor the changes of the underlying structure of the system. In order to accurately detect and identify the faults caused by process structure changes, a state-space model based on canonical variable analysis (CVA) is proposed in this paper, which has good performance on the representation of process dynamics and the properties of global identifiability. In addition, our approach not only has a strong ability to capture potential connection configuration information, but also greatly simplifies and improves fault monitoring performance because it is more sensitive to fault monitoring in the regression subspace of unrelated variables (acquired CVA status) Is orthogonal). Applying the method proposed in this paper to the simulation study of the four-tank system, the effectiveness of detecting and identifying structural changes is proved by multiple faults.