Automatic generation of diagnostic expert systems from fault trees

When a fault tolerant computer-based system fails, diagnosis and repair must be performed to bring the system back to an operational state. The use of fault tolerance design implies that several components or subsystems may have failed, and that perhaps many of these faults have been tolerated before the system actually succumbed to failure. Diagnosis procedures are then needed to determine the most likely source of failure and to guide repair actions. Expert systems are often used to guide diagnostics, but the derivation of an expert system requires knowledge (i.e., a conceptual model) of failure symptoms. In this paper, we consider the problem of diagnosing a system for which there may be little experience, given that it might be a one-of-a-kind system or because access to the system may be limited. We conjecture that the same fault tree model used to help aid in the design and analysis of the system can provide the conceptual model of system component interactions needed in order to define a diagnostic process. We explore the use of a fault tree model (along with the probabilities of failure for the basic events) along with partial knowledge of the state of the system (i.e., the system has failed, and perhaps some components are known to be operational or failed) to produce a diagnostic aid.