The hybrid two-level hierarchical knowledge structure based on the combined use of a first-principles ES and component-characteristic ANNs has been constructed for the diagnostic system. Implementation in the PRODIAG Code has been completed for the first-principles ES portion. From the initial results of the developmental testing at the early phase of "proof-of-concept", the proposed approach appears capable of alleviating most of the existing limitations in the use of artificial intelligence techniques to .diagnose processes in a realistic environment; being able to be comprehensively verified and validated, being flexible enough to diagnose an unforeseen event, and being capable of handling limited instrumentation. In addition, insight has been gained on techniques for the automated generation of rules and the optimization of minimum instrumentation for the plant-of-the-future. Work continues on the topology and training of the ANNs, and on refinements of the ES. Semi-blind and blind testing of the complete PRODIAG diagnostic system will be conducted.
展开▼
