System Decomposition for Distributed Multivariate Statistical Process Monitoring by Performance Driven Agglomerative Clustering

Conventional multivariate statistical process monitoring methods such as Principal Component Analysis perform poorly in detecting faults in large systems. Partitioning the system and implementing a multivariate statistical process monitoring method in a distributed manner improves monitoring performance. A simulation optimization method is proposed whose objective is to find the system decomposition for which the performance of a distributed multivariate statistical process monitoring method is optimal. The proposed method uses the search strategy used in agglomerative clustering in finding the optimal system decomposition. To demonstrate its effectiveness, the proposed method is incorporated into a distributed principal component analysis based monitoring scheme and applied to the benchmark Tennessee Eastman Process case study.