Multimode process monitoring using improved dynamic neighborhood preserving embedding

• A new process monitoring method for multimode industry processes is proposed.
• A novel distance is proposed to consider the scales of variables.
• Only one model is needed for monitoring in the proposed method.
• The process prior knowledge is not needed.

Complex processes often have multiple operating modes due to different manufacturing strategies. Meanwhile, within-mode process data usually exhibit dynamic behaviors and the data sample obtained at the present time may be correlated with those sampled for the previous and the next moment. In this paper, a novel improved dynamic neighborhood preserving embedding (IDNPE) algorithm is put forward and a new monitoring approach is proposed based on IDNPE. Different from the conventional principal component analysis (PCA) which aims at preserving the global structure of the data set, the proposed IDNPE tries to preserve the local neighborhood structure of the data set. In order to consider the scales of different variables within-mode and those of the same variables mode-to-mode, a novel distance which contains the local standard deviation information is employed in the IDNPE method. Moreover, for the dynamic behaviors of a single mode, the serial correlation is taken into account. Instead of constructing multiple monitoring models for multimode processes, the proposed IDNPE method builds only one global model without priori process knowledge. Finally, to test the modeling and monitoring performance of the proposed method, a numerical example and the Tennessee Eastman (TE) benchmark case studies are provided.