Real—Time Process Monitoring for Changing Inputs

Abstract

A method is presented for statistical process control using real-time data in applications with finite processing times. This extends existing methods that monitor process behavior, assuming constant inputs and process conditions. Summary scores constructed from linear combinations of the realtime data are monitored both in real time and at the end of processing. The summary scores have a minimal loss of information due to the high dimensionality and correlated nature of real-time data. A multivariate regression model relates the inputs and initial conditions to the summary variables. This permits prediction of summary scores for different inputs and initial conditions. Large deviations from the predicted summary variables, measured by process monitor statistics, signal abnormal behavior. Diagnostic tools identify the patterns in the real-time variable(s) that cause the signals.