Diagnosis by Algebraic Modeling and Fault-Tree Induction

We outline relevant characteristics of the vehicle diagnosis domain and requirements for diagnosis support. We argue that a combination of a model-based and a fault-tree approach will meet the requirements. We motivate and specify the task of automated modeling and develop a method to derive linear algebraic models of a given device. Models are derived by performing series-parallel analysis and by applying Cramer’s Rule in a tractable way in cases when spanalysis fails. The models are used to predict observations under arbitrary multiple-fault assumptions as a basis for fault-tree induction. The models can also be used for purely modelbased diagnosis. We sketch how to derive a fault dictionary from a model and how to induce a fault tree with an ID3-like algorithm.