Discriminatory Learning based Performance Monitoring of Batch Processes

This paper proposes a novel approach towards performance monitoring of batch processes that is oriented towards the requirements of real time assessment of batch health and online batch qualification. The proposed approach is based on the use of discriminant analysis and exploits class information that is generally known (but ignored) from the archive of historical batches. Wavelet approximations are shown to provide for a parsimonious representation of the batch profiles. A framework for batch classification that is based on the above discrimnatory learning is proposed to facilitate the task of performance monitoring. The developed methods are evaluated on a Penicillin fermentation process for their ability to monitor and to detect the faults both for real time batch qualification as well as for batch release procedures.