Diagnosis as a variable assignment problem: a case study in a space robot fault diagnosis

Abstract

In the present paper we introduce the notion of Variable Assignment Problem (VAP) as an ab-stract framework for characterizing diagnosis. Components of the system to be diagnosed are put in correspondence with variables, behav-ioral modes of the components are the values of the variables and a diagnosis is a variable assignment which explains the observations of the diagnostic problem, by considering the con-straints put by the domain theory. In order to have a concise representation of diagnoses and to reduce the search space, we introduce the notion of scenario for representing a set of diagnoses. The paper discusses the definition of preference criteria for ranking solutions and their use for guiding the heuristic search for di-agnoses. Experimental data are reported for the evaluation of such a heuristic search on a real-world diagnostic problem, concerning the identification of faults in a space robot arm; in this domain, where a high number of diagnoses may be possible, our approach allows one to get a concise representation of the large number of solutions and to define effective diagnostic strategies able to provide relevant information about fault localization and identification. 1