Application of recursive statistical-based techniques for industrial process monitoring

This paper discusses the monitoring of large-scale processes that exhibit non-stationary and/or time-varying behavior. The work considers statistically-based monitoring techniques that relate to the multivariate statistical process control (MSPC) framework. Particular focus receives partial least squares (PLS), as it allows distinguishing between cause and effect variables. To demonstrate the difficulties of monitoring processes that show non-stationary and time-varying behavior, the use of conventional PLS is contrasted with its recursive and moving-window counterparts. An application study to a realistic simulation of a fluid catalytic cracking unit (FCCU) (i) shows the difficulties of monitoring non-stationary and time-varying process behavior and (ii) raises attention on how to apply recursive and moving window PLS correctly.