Extraction of event times in batch profiles for time synchronization and quality predictions

A technique for the extraction of the time of occurrence of consistent features in batch profiles is presented. The event times are needed for the synchronization of the batch trajectories for meaningful analysis of the historical data using multivariate statistical techniques such as principal component analysis and partial least squares and also for studying the correlation of the event times with the final batch product quality. The event detection technique is based on the matched filter approach. A normalization strategy that compensates for the measurement axis variability in the profile snapshots is enunciated for accurate extraction of event times. The methodology is applied to profiles generated from a simulated methyl methacrylate polymerization batch reactor and an industrial batch polymerization reactor. The use of the event times for time synchronization is illustrated. The importance of studying the correlation of the event times with the final batch quality is also emphasized.