AID System: Abductive and Iterative Diagnosis

PROJECT:

Model Based Diagnosis

MAIN GOALS:

Integrated environment for:

• building causal and behavioral models
• performing abductive diagnosis based on such models
• explaining results of diagnosis

SPECIFICATIONS AND APPROACH

• graphical editing causal and behavioral models:
  defining name and a set of possible values for
  • states of (part of) the modeled system
  • observable parameters
  • initial causes
  defining relations:
  • initial causes, states, context --> state
    (a combination of initial causes, states and contexts causes, or may cause, a state)
  • states --> observations
• model translated in textual (logical) form: set of clauses
  a1 and ... and an -> b
  where some of a1 ... an can be assumable (conditions that can be assumed in order to explain observations)
• abductive diagnosis based on logical models:
  find set of assumptions that, given the model, implies the observations;
  a typical choice for abductive diagnosis: give explicitly only the fault model (abnormal states --> abnormal observations); in this case the set of assumptions must imply abnormal observations and be consistent with normality observations.
  E.g., knowledge: flu -> fever, flu -> headache etc etc
  observations: fever, not headache
  flu is not a consistent explanation (implies headache); look for other explanations of fever not implying headache, but don't look for assumptions implying "not headache"
• typically, given initial observations, there is more than one diagnosis
  --> acquisition of further observations to discriminate (iterative diagnosis)
• explaining results of diagnosis: for each diagnosis, show part of the model (network) from the assumptions to the observations to be implied
• improving efficiency of diagnostic cycle with the use of compiled conditions: when there are choices, perform local discrimination using easy-to-observe conditions associated with alternative explanations: ask such conditions if not available. The conditions can be precompiled from the model

IMPLEMENTATION TECHNIQUES

• Reasoning implemented in Prolog
• Graphical interface implemented in C in the OSF/Motif graphical environment