State dependent monitoring of batch and multiproduct processes

The challenge in creating a useful process monitoring system is delivering the right information to the right user at the right level of detail. Batch, semicontinuous, and multiproduct processes present an additional challenge: the monitoring must be sensitive to changes of state of the plant. What constitutes an "abnormal" temperature, for example, may depend on the product being processed. We present a knowledge-based approach to process monitoring that meets the challenges of being both state-dependent and user-dependent. In addition, our approach overcomes two practical drawbacks of knowledge-based process monitoring: 1) process knowledge is context-specific, making it difficult to capture; and 2) knowledge-based systems are hard to maintain. The first problem is solved by state handling and event detection. The second problem is addressed through a modular architecture that emphasizes knowledge retention and direct maintenance of the knowledge by the users of the system. The approach has been successfully implemented at two chemical plants, one of which is a semicontinuous operation producing over 200 different products.