Approximate Reasoning Techniques for Intelligent Diagnostic Instruction

Intelligent instruction of fault diagnosis, or troubleshooting, tasks requires the capability to automatically infer the significance of particular test outcomes observed by the learner in a practice environment. This single process is central to virtually every aspect of intelligent instruction in this domain, ranging from evaluating learner proficiency to recommending an effective testing strategy. In highly complex target systems this task becomes intractable, for the number of possible faults and system configurations is enormous. The issue is not that an artificial diagnostic expert must be highly precise, but that the internal expert must interpret symptoms observed by the learner in terms of the symptoms which might emerge from all possible sources, not just the faults that are defined in the pool of sample faults that are presented in the exercises. The approach presented here operates upon qualitative expressions of the possible symptoms produced by faults in various units of the target system. This greatly reduces the burden placed upon the author of the application, while capturing the essence of his or her symptom knowledge. This approach has been implemented in an instructional system that automatically computes a provisional bank of symptom possibilities by simulating the faults in the exercise pool, then acquires from the author qualitative refinements to those computed possibilities, and finally delivers instruction based upon that body of information.