Diagnosis As Planning: Two Case Studies

Diagnosis of discrete event systems amounts to finding good explanations, in the form of system trajectories consistent with a given set of partially ordered observations. This problem is closely related to planning, and in fact can be recast as a classical planning problem. We formulate a PDDL encoding of this diagnosis problem, and use it to evaluate planners representing a variety of planning paradigms on two realistic case studies. Results demonstrate that certain planning techniques have the potential to be very useful in diagnosis, but on the whole, current planners are far from a practical means of solving diagnosis problems.