Integration of multiple reasoning systems for process control

A general schema for integrating multiple reasoning systems is presented along with its implementation for emulsion quality control in the Sendzimir rolling mill. In this, probably the largest industrial AI application in Slovenia, benchmarks of different types were performed, including combinations, integrations and hypothesis boosting, with and without help from an expert. Experiments show that :he use of multiple reasoning systems is in general beneficial whenever art appropriate combination is found. Empirical measurements were also used to verify analytical models of multiple systems. On average, fitting errors were well below 1%. The schema enables integration of an arbitrary number of systems as black boxes into a transparent knowledge base. Integration is performed on the basis of training examples or hypothetical examples. (C) 1997 Elsevier Science Ltd. All rights reserved.