FROM A RULE-BASED TO A PREDICTIVE QUALITATIVE MODEL-BASED APPROACH USING AUTOMATED MODEL GENERATION - APPLICATION TO THE MONITORING AND DIAGNOSIS OF BIOLOGICAL PROCESSES

This paper deals with the monitoring and diagnosis of fedbatch biologi cal processes. It reports two aspects of experience conducted on the s ame pilot plant. The first part of the paper presents the rule-based a pproach which was used to build the expert system BIOTECH, and the sec ond part presents the dynamic model-based approach followed later to b uild the system CA-EN. Both systems detect abnormalities in the proces s behavior in advance, assess the state of unmeasured variables, advis e the operator about what to do to correct unwanted behaviors of the p rocess, and may be used as computer-aided process-monitoring systems. On the one hand, they share the same items of knowledge about the proc ess. However, as each system has its own knowledge-representation form alism, substantial differences exist in what can be implemented and ho w, and in their anticipatory capabilities. On the other hand, the two systems really diverge on the way in which they use the knowledge to p erform the monitoring and diagnostic tasks, BIOTECH having the task kn owledge and the process knowledge intimately linked in the rules, and CA-EN making use of the two types of knowledge in a strictly separate way. (C) 1998 Published by Elsevier Science Ltd. All rights reserved.