Symptom propagation and transformation analysis: A pragmatic model for system-level diagnosis of large automation systems

In automation systems, the diagnostic task is to efficiently deduce candidates for replacement or repair from centrally available observations. Often, classic model-based diagnosis cannot provide a practical solution as it requires in-depth knowledge of system components and their functions and interactions. Such detailed data is seldom readily available. Our approach to system-level diagnosis adopts a purely non-functional view of the system, similar to field-tested safety assessment techniques, e.g. fault trees or FMEA, but goes beyond that to support core diagnostic analyses (runtime diagnosis and diagnosability). We propose a new component-based model that is solely focused on faults and causal propagation and the transformation of symptoms. Modeling is now an iterative process that supports refinement as new data becomes available. We evaluate our approach on a small-scale rail automation example.