Technical University of Munich, Department of Computer Science
MQM Paper: [Struss 92a]

Peter Struss

An Amalgamation of Model-Based and Heuristic Reasoning for Diagnosis

In: Industrial Applications of knowledge-based Diagnosis. Eds.: G. Guida et al. Elsevier, Amsterdam, S. 311- 329, 1992.


In this paper, we characterize diagnostic systems of two generations and discuss their advantages and limitations. A widespread opinion is that first generation diagnostic systems work quite efficiently, but are unreliable and incomplete w.r.t. possible faults and symptoms, whereas model-based systems are complete and robust, but suffer from complexity of models and intractability of the incorporated algorithms. Further conclusions often drawn are a) that this is because the former use heuristics whilst the latter do not, and b) that combining the two types of systems might solve the problem.

We argue that the tension between heuristics and model-based reasoning is a non-problem, because model-based diagnosis involves certain heuristics and also requires them. It is shown that a "crossbreeding" approach, which uses one paradigm as a critic of the results of the other one, inherits the limitations of both instead of overcoming them. In contrast, we present a principled integration that is clearly and formally grounded on model-based diagnosis and does not require essential changes in the implementation. It is based on the introduction of retractable working hypotheses. Heuristics play their well-defined role in controlling and focusing inferences in the model-based diagnostic engine rather than replacing its rigorous inference schemes by plausible associations. Because focusing decisions are retractable, the completeness properties of the basic algorithm are preserved, while complexity of reasoning is reduced for all cases in which simplifying working hypotheses remain valid.


If you have comments or suggestions, email me at
created: Anton Seidl, Feb 12, 1997, last updated: Anton Seidl, Feb 12, 1997