On-line process monitoring using local measures of association: Part I — Detection performance

• A new on-line monitoring scheme for the process correlation structure is proposed.
• The scheme is based on a EWMA updating approach.
• Partial correlations consistently lead to stable and higher detection performances.

This article addresses the problem of monitoring on-line the correlation structure of process variables. Current solutions consist in adopting a moving window approach or an EWMA updating scheme for continuous estimation of the variance–covariance matrix, based on which several monitoring statistics can be derived. The last alternative (EWMA recursion) has attracted more interest, and monitoring statistics have been proposed for the trace of the updated covariance or regarding squared deviations from target. Even though better than the usual benchmark, PCA-MSPC, these methodologies are based only on the (marginal) covariance and therefore unable to discern local changes in the process structure. Partial correlations are a useful tool for enhancing the detection of such structural changes, since they convey more detailed information about the process structure. The results obtained confirm the improved monitoring performance gained with the use of partial correlations.