Multiple Collaboration Preserving Projection for Monitoring of Complex Industrial Process

To fuse and mine multisource and multimodal data in monitoring complex industrial processes effectively, a new dimensionality reduction method called "Multiple Collaborative Preserving Projection" (MCPP) has been proposed. MCPP is innovative in the sense that it is capable of extracting and preserving the correspondences of heterogeneous data in temporal series and spatial structure. In temporal series, both local and global correlations of raw datasets are preserved; in spatial structure, the local structure of data is extracted based on the captured global structure, and a data projection method is integrated to reduce the adverse impact of industrial noise in raw data; moreover, the collaboration among multiple views is achieved by specifying the direction in projecting to subspaces; this maintains the consistency of common features in multiple views for a lower dimensional subspace. The effectiveness of MCPP has been verified in the case studies of the Tennessee Eastman process (TEP) and an industrial process to monitor the conditions of tuyere raceway in an actual blast furnace (BF).