Multi-block dynamic weighted principal component regression strategy for dynamic plant-wide process monitoring

To properly monitor dynamic large-scale processes, a new distributed dynamic process monitoring strategy named multi-block dynamic weighted principal component regression (DWPCR) is developed in this paper. Because complex plant-wide processes have multiple operation units and complex correlations among variables, traditional global process monitoring models may suppress local fault information and fail to identify incipient faults and local faults for large-scale processes. Besides, product quality determines the economic benefits of the enterprise. Motivated by these problems, this work studies the distributed quality monitoring strategy. At first, the idea of community partition in complex networks is used for multiple subblock division for a large number of process variables in this new monitoring framework. Then, the monitoring model for each subblock is established by the proposed DWPCR approach. Moreover, a novel weighting key components strategy based on fault information is proposed to monitor the process. Finally, the comprehensive monitoring result is fused by Bayesian inference. The superiority of the proposed distributed DWPCR strategy is testified in the case study.

Automated summary

A new distributed dynamic process monitoring strategy named multi-block dynamic weighted principal component regression (DWPCR) is developed, which partitions process variables into multiple subblocks using the concept of community partition in complex networks. The monitoring model for each subblock is established and a novel weighting key components strategy based on fault information is proposed for process monitoring.