Discriminating qualitative model generation from classified data

Modeling is quite critical and remains a bottleneck for modelbased diagnosis in many application domains. Quantitative models that are developed during the design stage are not applicable as so to model-based diagnosis engines. This paper proposes to take advantage of discretization algorithms used by the machine learning community to discretize the domain value of continuous variables and generate a behavioral qualitative model from the data clusters corresponding to classified data. The results of this approach are illustrated and discussed with the two tanks benchmark example.