ABSTRACT
The diagnosis of faults on a flight deck differs fundamentally from static systems in which a malfunctioning device can be taken off-line for troubleshooting. In civil aviation, at least, it is unlikely that artificial intelligence concepts will find their way into flight-critical automation systems until they have been thoroughly proven in less critical applications. The first general artificial intelligence proposal for aiding diagnosis in real-time systems was the use of rule-based expert systems. The degradation of joint human–machine performance in such a system has been well documented. The fundamental assumption is that if the model of the system is indeed correct, then all symptoms arise from actual malfunctions in the system. Diagnostic systems should bring pilots closer to what is going on within a subsystem, rather than distancing them from the process. In a dynamic system, the process must go on while the fault is handled; an airplane cannot be "parked at a waypoint" while the trouble is dealt with.
