Elaborating the problem-solving model of a fault diagnosis expert system by knowledge level prototyping

Abstract An important phase of the construction of an expert system is the specification of a problem-solving model, i.e. an abstract implementation-independent description of the modelling primitives that will allow capturing the expertise, the different high level actions to perform and the general strategy controlling these actions. This paper presents the construction of a problem-solving model for a fault diagnosis expert system on robots. This model has been elaborated in a particular context where (1) the problem-solving model must correspond to that of human experts and (2) an operational prototype should be available after a short delay. In order to tackle these two objectives we have used a knowledge level prototyping approach. This consists of the definition of a first version of the problem-solving model by data abstraction, the construction of a prototype using a high level language that avoids low level implementation problems and the refinement of the model based on the analysis of the prototype by human experts and reflective analysis tools. We summarise the major aspects of the elaboration of this problem-solving model and the corresponding prototype, and we discuss the advantages of this approach.