Interactive SRU diagnosis using neural networks

The problem of repairing shop replaceable units (SRUs) is aggravated by the fact that most electrical components can be expected to exceed the service life requirements of the system in which they are embedded. It is not cost-effective to require testing time (and software development time) to diagnose failure modes which will probably never occur. In practice, the person repairing a circuit knows more about the circuit than anyone else, including the original developer of the diagnostic software. It is therefore desirable that meaningful diagnostic software be able to capture this expert knowledge as it becomes available to diagnose failure modes not anticipated in the original software development. To assist in this task, a user-friendly, menu-driven, artificially intelligent software program has been developed to assist the technician in diagnosing circuit card failures. As failures are diagnosed, the technician enters the corrective action as well as the essential failure and probing information. When similar failures are encountered in the future, the neural network diagnoses the most probable failure mode and directs the technician as to the information which would be useful in pinpointing the fault