Quality-Related Plant-Wide Process Monitoring Based on Mutual Information-Neighborhood Preserving Embedding-Partial Least Squares for Hot Strip Mill Process

Complex industrial processes usually consist of multiple interrelated operating units. It is of great theoretical and engineering significance to study the plant-wide process monitoring. In consideration of the complex coupling relationship among the sub-processes of the plant-wide process, a novel quality-related plant-wide process monitoring method based on mutual information-neighborhood preserving embedding-partial least squares (MI-NPE-PLS) is proposed. Firstly, the mutual information between different process variables and product quality variables are calculated, which are taken as the weight factor of process variables to enhance the quality-related features and weaken quality-unrelated features. Secondly, according to the mechanism knowledge, the plant-wide process is decomposed into several subblocks, and neighborhood preserving embedding algorithm is used to reconstruct the local features of subblocks. Then, the quality-related features of the upstream subblock are extracted based on PLS, and then the process data of the current subblock is augmented with the upstream features. In this way, the interaction information among each sub-block is considered and local quality monitoring models are established. Finally, the Bayesian fusion method is used to establish a global monitoring model to realize the quality-related process monitoring of the plant-wide process. The advantages and validity of the proposed scheme are verified on an actual hot strip mill process.