Post-mortem diagnosis of bottling plants based on recorded data

We present LineMod, a model-based system that performs post-mortem fault localization of food packaging plants and, more specifically, bottling plants based on recorded data. In this project, we need to cope with non-negligible transportation time of objects and capture phenomena like the tailback of units (if transportation is blocked) or the propagation of gaps in the flow of units. Because the application context requires compositionality of the model, i.e. local, context-free models of the individual transportation elements, we are also facing the problem that whether or not a single element produces an output flow (or accepts an input flow) cannot be determined solely by the model of this element, but only through modeling the interaction with the subsequent element, which may block the output (or the previous one not providing the input). The diagnostic system scans a data base with recorded data from the involved machines for fatal symptoms (such as the stop of the filler) and determines the respective causes as a basis for taking actions to improve the performance of the plant. We present results of the evaluation of the first prototype and discuss challenges to future work.