Testing for Discrimination of Diagnoses
In: Fifth International Workshop on Principles of Diagnosis (DX-94) New Paltz, NY, USA, 1994
To discriminate among the diagnostic candidates which are hypothesized by a diagnosis system at a certain stage may require the application of tests, i.e. causal inputs to the device to be diagnosed that promise to reveal distinctions between the candidates through observable features. We develop a theory and an algorithm for generating and applying tests in diagnosis based on a behavior representation by relations. The approach is general and handles also continuous systems (in theory and, through model abstraction, also in practice), thus going beyond the area of digital circuits. We also present a test selection strategy based on a minimum entropy criterion which, as a spin-off, comprises a generalization of the widely used probe selection strategy in consistency-based diagnosis.