Troubleshooting: When Modeling Is the Trouble

TROUBLESHOOTING WHEN MODELING IS THE TROUBLE Paper type: Full paper Track: Engineering P . Dague Mailing address : Paris IBM Scientific Center 36 Ave R . Poincare 75116 Paris France Telephone: 33 (1) 45 05 14 00 P. 27-40 Address: earn/ bitnet dague at frpoi l 1 P . Deves Electronique Serge Dassault 55 Quai M. Dassault S , Cloud 92214 France O. Raiman Paris IBM Scientific Center 36 Ave R . Poincare Paris 75116 France When troubleshooting, finding components where changes in behavior occur only guides the search . In a misbehaving device, a correct component can indeed change its behavior. Distinguishing faulty components from correct ones requires modeling all possible correct behaviors of components. For analog circuits, there is a lack ofnumerical models anda lack of information on essentialparameters . Therefore, modeling becomes quickly the real trouble. Under the assumption that "a defect leads to significant change in the behavior of a device", we show that Qualitative Reasoning is a solution . To exploit the above assumption, order of magnitude reasoning is used for modeling and defining a strategy . Experience with the expert system DEDAIE has .shown that this assumption applies to a wide class of cases, and that Qualitative Reasoning can be usedfor a real size application . Submitted to AAAI-87 (also submitted to IJCAI-87) Topics: Engineering Problem Solving Troubleshooting Commonsense Reasoning Qualitative Reasoning