Modeling and Compilation Aspects of Fault Diagnosis Complexity

Model-based diagnosis (MBD) is a promising approach for fast and accurate diagnosis of root cause of failure for complex systems. Critical success factors in MBD are accurate realism of the model and diagnostic inference speed. Both success factors are inherently at conflict. In this paper we demonstrate how to model system components that are typically found in technical systems in a qualitative approach, thereby limiting the need for more variables as well as increased domain resolution. Furthermore we outline the computational trajectory from model to diagnosis, in which we distinguish between (offline) model compilation to an intermediate representation, and the subsequent (on-line) diagnosis step, based on actual observations. We use the fuel system of an Airbus 330 and a scenario based on a real-life incident as leading example. The results show that our framework is well suited for qualitative models and that these models provide a realistic representation and accurate diagnosis. From our results we conclude that the selected knowledge representation can speed-up the diagnostic process to up to two orders of magnitude.