IMPROVING ON-BOARD DIAGNOSIS EFFICIENCY THROUGH AUTOMATIC SYSTEM MODEL ABSTRACTION

The paper presents an approach suitable for on-line diagnosis, which aims at automatically abstracting the domains of discrete variables in the model (i.e. behavioral modes of system components) in order to keep only those distinctions that are relevant given the available observations and their granularity. In particular the paper describes an algorithm which identifies indistinguishable behavioral modes by taking into account specific classes of available observations and derives an abstract model where such modes are merged and the domain model is revised accordingly. By considering increasingly restricted classes of available observations (and/or granularity of observations), a set of abstract models can be derived that can be exploited through model selection each time a new diagnostic problem has to be solved. The approach has been tested within the framework of a diagnostic agent for a space robotic arm, and experimental results showing the reduction in the number of diagnoses are reported.