Design and Implementation of a Process Monitoring System in a Just in time PIMS Platform

Abstract

Process monitoring in complex multivariate systems is challenging due to numerous influencing factors, making it difficult to track process specifications or expected error rates. These systems often exhibit non-stationary behavior and operational modes with constant changes in key performance variables, complicating the use of Multivariate Statistical Process Control (MSPC) techniques, which require process stationarity. Here it’s proposed a methodology for designing PCA, T², SPE, and combined index control charts to handle small mean fluctuations. The methodology considers the quasi-stationary mean and variance estimated from typical operating data for each set point or operating region, enabling distinct designs for each. A first-order filter with a slow time constant estimates the mean of non-stationary process variables, allowing work with residuals that reflect true common-cause variability. The implementation of this methodology in a commercial PIMS platform for real-time monitoring of a thermoelectric power plant is discussed. Experimental results show that the control charts perform better during the monitoring phase, frequently alarming in the presence of special causes.