Qualitative Model Abstraction for Diagnosis

Building qualitative models is a crucial task for modelbased diagnosis. This paper discusses the techniques to automatically transform a quantitative model in CAD environment into a qualitative model, under the cases that the real numbered landmarks are known and unknown. With known landmarks, the abstraction is through the discretization process where the simulation data is discretized according to the given landmarks. If landmarks are unknown, the landmark generation process, which is inspired by the discriminability analysis for multiple behaviour modes, is applicable. For dynamic systems, the pseudo-variables are introduced to describe the dynamic behaviour. The techniques developed are demonstrated by a simplified automotive subsystem.