Neural Networks and Expert Systems have each shown important successes in diagnostic domains ranging from process control to vehicle power trains. Neural networks are particularly effective at lowering knowledge engineering costs by automatically learning patterns from training data, while expert systems are more effective at representing knowledge gained from experts, at providing explanations of its diagnoses, and at performing sequential actions such as test sequencing, repair, and repair validation. The proposed work investigates a hybrid system where neural networks perform an initial diagnosis via acoustic signal recognition, and an expert system performs follow-up tests leading to a specific diagnosis and repair strategy. The Phase I prototype showed that a standard low-cost platform could support a hybrid of neural network, expert system, and data acquisition software, that the hybrid system has a smooth user interface, and that the neural network can successfully classify acoustic signal data. The Phase I survey showed considerable interest from potential users in industry and the military. The Phase II study will address critical research issues: comparison of various neural network paradigms for effective signal classification, comparison of neural networks with traditional signal classification techniques, and effective techniques for explaining the hybrid systems's reasoning to the developer and end user. The hybrid system has potential applications in situations where signal data and symbolic data must be used together for a definitive and repair procedure. Potential applications include factory floor, military, aerospace, nuclear power plant, commercial vehicles, and materials certification.
