Application Of Constraint Suspension Techniques To Diagnosis In Helicopter Caution/Warning Systems

This paper describes research done at McDonnell Douglas Helicopter Company in applying artificial intelligence techniques to the task of advising a helicopter crew of events taking place in their aircraft subsystems. As part of the U. S. Army Apache Enhanced Diagnostic System Contract (EDS), a demonstrator system was developed that is capable of simulating multiplex bus traffic of caution/warning related data, and invoking a diagnostic module upon detection of an abnormal condition. Problem indicators are prioritized based on severity and displayed on a simulated up-front cockpit display and multifunction display. The demonstrator attempts to provide a diagnosis, where possible, and also provides advice about actions to be performed in response to critical situations. Severe problems are announced using an aural annunciator with digitally sampled stored messages. The diagnostic module utilizes a technique called candidate generation via constraint suspension to prune the list of suspected components while diagnosing a failure. Constraint suspension is a form of model-based reasoning that allows faults to be diagnosed by modeling the intended correct behavior of the system and comparing the outputs of the model with the true behavior of the system. This technique, which was first investigated by Randall Davis at the Massachusetts Institute of Technology under OARPA funding, shows potential for use as a diagnostic tool in ground-based as well as airborne systems.