Integration maschineller Lernverfahren in eine Materialflusssimulation zur Verhaltensabstraktion und -vorhersage komplexer Fertigungssysteme Integration of Machine Learning Techniques in a Material Flow Simulation for Behaviour Abstraction and Prediction of Complex Manufacturing Systems

Modeling complex interactions and dependencies in production systems proves effortful in traditional event-based simulation tools mainly relying on a whitebox approach. In this context, the integration of machine learning techniques offers new possibilities for establishing a more efficient modeling of the systems behavior by using historic data. Therefore, this paper presents a concept for linking a material flow simulation with machine learning models. The implementation of this novel, data-driven modeling approach is shown using the example of the hardening process for calcium silicate masonry units. By using and comparing different supervised learning models, a behavior model for the complex manufacturing control of the autoclaving process could be created based on historical data of the real process. As a result, the generated behavior modeling does not depend on a priori knowledge, which significantly reduces the modeling effort in many of these complex simulation cases.