Optimizing neural network technology for BIT applications

Increased fault detection capability in airborne systems is needed to reduce system life-cycle maintenance cost and improve mission readiness. Neural network techniques have been used successfully in applications requiring capabilities similar to those required to cope with the built in test (BIT) false alarm problem, and have demonstrated flexibility for application to fault detection and diagnosis. Many neural network techniques are applicable to optimizing BIT performance. Any implementations, however, must avoid drawbacks of neural networks such as processing requirements, real-time learning, and lack of effective verification means. An approach has been proposed which uses some of the techniques such as weighting and nodal connectivity, but suggests the need for simulators to verify and implement configuration controlled learning