MODELLING AUTOMOTIVE ENGINES FOR AUTOMATED DIAGNOSIS
Charles David Hunter · Summit (Simon Fraser University) · 1991
As the complexity of vehicular electronic control systems increases, automobiles are becoming increasingly difficult to diagnose.Diagnostic expert systems based on production rules have been used with limited success for small, well understood domains.Model-based diagnostic systems offer advanced capabilities, as displayed for domains such as electronic circuits, where efficient models are easily developed.The success of model-based technology for automotive diagnosis depends on the availability of efficient automotive diagnostic models.Engine models are presently developed for two purposes, design and control.Models for design often require extensive computation, and deal with variables unrelated to diagnosis.Control models require empirical results from lengthy bench testing.Models specilkally for engine diagnosis have not been reported.All models are necessarily incomplete, and even the most detailed models will be unable to find all diagnoses.Quantitative models pursue excessively detailed calculations.Qualitative models are potentially more efficient while still providing the necessary detail for diagnosis.Our model represents physical components as primitives, and groups of components working together as composite components.We incorporate a specialization hierarchy, which uses inheritance to centralize, and reduce the storage of, information that is common to similar types of components.We also utilize a composition hierarchy ta derive the structure and behaviour of complex systems from that of its sub-components.W e present a prototype engine subsystem model to diagnose single, nonintemittmt faults, implemented with the Echidna constraint reasoning system, which incorporates constraint logic programming, truth maintenance, and dependency backtracking, all in an object-oriented framework.Performance of the prototype is reported, and is extrapolated to estimate the performance of a complete engine model.Limitations of the prototype model, and suggestions for further research, are discussed.I would like to thank the director of the Simon Fraser University Expert Systems Lab Dr. William S. Havens for leading the diagnosis group, acting as second supervisor, and for his determined generation of, and participation in, academic debate, the graduate students of the diagnosis research group, Afwarman Manaf, for his support a d encouragement, and Peter MacDona'fd, for forcing us to broaden our research horizons, McCarney Technologies fnc., for their financial support, and Stefan Joseph for cheerfully acting as McCarney liaison and external examiner.It was a pleasure to work with the staff of the Expert Systems Lab -Miron Cuperman, Rod Davison, and especially Sue Sidebottom, whose programming assistance and patience were invaluable.Finally,