Discrepancies – theories and applications of model-based diagnosis

For about a decade, research on model-based reasoning has been carried out with increasing interest and efforts. One major motivation for this work was the insight in the limitations of first generation expert system technology. In particular in technical diagnosis, it became obvious that encoding experiental knowledge in rules does not suffice and that an explicit representation of the structure and behavior of the system to be diagnosed is necessary. A variety of model-based diagnosis systems has been implemented as prototypes, and recently, the number of contributions to the development of a (logic-based) theory of diagnosis from first principles is increasing significantly. At the present time, the area is crossing the threshold towards real-world applications. So far, the success is limited, and only part of the original goals has really been achieved. The situation reveals discrepancies between the theories developed so far and the requirements of real-world applications. There are two major problems in any application of model-based diagnosis: constructing the domain model and finding an appropriate model of the diagnostic process. In both respects, current technology offers only limited solutions, and the link between them deserves particular interest: ”How to structure and use device models in the process of diagnosis”. Solutions to this conceptual problem also address a major technical problem most real-world systems are facing: inefficiency due to complexity of the device model. In this talk, we outline some motivations, goals, and foundations of model-based diagnosis and try to assess the state of application-oriented work and the value of contemporary elements of a theory of diagnosis. Based on this, the discrepancies between theories and applications and, hence, perspectives of future research and tasks for theoretical work can be identified.