Diagnosis and diagnosability based on a Symptom Propagation and Transformation model

The automated monitoring and diagnosis of very large, heterogeneous, distributed automation systems is a complex task for which both technical and process-related factors have to be optimized. Industrial applicability of a diagnostic procedure depends as much on the quality of diagnostic results as on the time and cost it takes to provide and maintain a suitable description of the target system's structure and behavior. In earlier work we proposed a new symptom-based, diagnostic model, which addresses complexity issues of traditional model based diagnosis. In a pragmatic approach, our component-based model concentrates on a causal, present-knowledge description of deviant behavior only, which significantly reduces computational complexity and also preceding efforts for initial system modeling. In this paper, we expand on our formal concept of Symptom Propagation and Transformation. Applying causal reasoning to this model, we provide definitions and algorithms for both online diagnosis and static analysis of diagnosability. We evaluate our results on a case study from the rail automation domain.