SUMMARY
Fault diagnosis of highly automated manufacturing systems presents considerable difficulty to the human operator. A decision support is highly desirable. Many neural networks have been developed for diagnostic decision support. Most of them rely on an identification of fault causes through a direct recognition of fault symptom patterns using a single back-propagation network. However, those neural networks have difficulty in distinguishing fault events that share a common symptom pattern. Incomplete knowledge training may result in performance deficiencies of those neural networks. A hybrid intelligent system is developed here that overcomes these problems. It integrates neural networks with a procedural decision making algorithm to implement hypothesis-test cycles of system fault diagnosis. The hybrid intelligent system demonstrates highly reliable performance of fault diagnosis on tested fault events.