THE DISCRETE CHARM OF DIAGNOSIS BASED ON CONTINUOUS MODELS

We can characterize both methods of fault detection and isolation (FDI) in control engineering and model-based diagnosis (MBD) developed in Artificial Intelligence (de Kleer and Kurien, 2003) in a very general way as searching for system models that agree with the observations of the system’s behavior. MBD, or more precisely consistency-based diagnosis, is based on a rigorous logical formalization to define and identify “models that agree with the observations”. Model and observations are inconsistent if they entail different values for a variable. While there is no problem to distinguish discrete values, say 0 and 1 in a digital circuit, the absolute logical “either True or False ” may appear inappropriate when a measurement of 12.7 i