Characteristic States Based Supervision Of Penicllin Fermentation Batch Processes

Nowadays, supervision technologies of batch processes including penicillin fermentation mainly involve multivariate statistical process control approaches such as PCA. With strong graphical modeling and mathematical analysis abilities available, Petri Nets are particularly adapted to deal with batch process issues. Inspired by these observations, regarding penicillin fermentation processes, this thesis presents a characteristic state based supervision approach which combines RBF neural network and timed Petri nets techniques.Initially, in-depth investigations on penicillin fermentation are performed, resulting in a couple of characteristic states which could best descript the operating conditions of the process. On this basis, a RBF network approach to extract the characteristic states is introduced. Subsequently, using timed Petri net modeling and simulating techniques, we explicitly establish timed Petri.net models for characteristic states evolutions of the penicillin fermentation, including the models in normal conditions and those of process supervision oriented. Finally, a characteristic state based intelligent supervision system which combines RBF neural network and timed Petri net models is developed, effectively implementing characteristic state extraction as well as process supervision in a hierarchy. Experimental studies are carried out to testify the approaches.In concluding, the methodology of characteristic state supervision combining timed Petri nets and RBF networks provides an effective means for batch process monitoring, expecting promising applications potential.