Diagnosis of Bottling Plants - First Success and Challenges

The paper describes an application of component-oriented consistency-based diagnosis to the domain of bottle-filling plants. The task is to localize the causes for stops of the central aggregate, the filler, based on recorded operation data of a plant. A model-based solution is challenging in several respects, especially due to high uncertainty in the transportation processes to be modeled, the nature of the available data, and the relevance of numerical temporal information. We give a short description of the application and its requirements and summarize essential characteristics of the solution. We focus on the evaluation of the first demonstrator and a discussion of some challenges for future work, which include questioning the "classical" notion of a fault in component-oriented diagnosis.