On-line mode identification of transitional modes based on differential PCA for multimode processes

Abstract

Multimode is the general characteristic of complex industrial processes. Different modes have different process characteristics, so different models should be established to describe them. Therefore, on-line mode identification is necessary to choose corresponding model to realize process optimization, process monitoring and condition evaluation of multi-mode processes. If transitional mode has been identified during on-line mode identification, the next steady mode can be determined, so the mode identification of transitional modes is the key point to on-line mode identification of multi-mode processes. A method using differential PCA and dynamic trend match is proposed to identify the type of transitional modes. Dynamic information can be obtained by differential transform of transitional data. Dimensionality is reduced by using principal component analysis. The principal components containing much variation are chosen to analyze the dynamic change trend, and process characteristics which can identify the transitional mode are extracted. Dynamic information matrix of online data is matched with offline mode characteristic matrix to identify the mode of the online data. Feasibility and accuracy of the method are evaluated by the illustration.