Handling qualitative aspects of human knowledge in diagnosis

Knowledge involved in diagnosis of real complex systems comes from human experts and requires appropriate discrete and qualitative representation. The large amount of information resulted is difficult to be managed and prepared to enter the diagnosis system without the help of an appropriate tool. The paper proposes a knowledge elicitation scheme for multifunctional conductive flow systems under fault diagnosis along with appropriate representation of normative and faulty models. Prototype and instance manifestations get a semi-qualitative representation and symptoms refer to means-end and bond-graph entities in a new approach, suited to human diagnostician's conceptual view. The Computer Aided Knowledge Elicitation (CAKE) tool proposed copes with knowledge involved in the diagnosis. The case study on knowledge elicitation for fault diagnosis of a hydraulic installation and the conclusions highlight advantages of the present approach.