Multimode process monitoring based on correlative principal components and differential geometry feature extraction

Since the changes of raw material properties, external environment and other conditions, during practical industrial processes, multiple stable operation modes may arise, and between any two stable modes may undergo slowly changing transition modes. The existing multimode process monitoring methods haven´t monitored dynamic characteristics of the transition modes efficiently. This paper adopts differential geometry feature extraction method to extract the dynamic characteristics of transition modes, uses geometric elements, such as slope, curvature etc, to display the dynamic curve characteristics of transition modes, and then establishes the anomaly detection model of transition mode based on rolling balls to monitor the transition modes. The online data driven CPCA method is used for the anomaly detection of stable modes. Comparing this method with the global PCA and global CPCA, the experimental results show that the proposed method is efficient.