Maschinelles Lernen von Maschinenfolgen für den simulationsbasierten Test von Verfahren der Produktionsplanung und -steuerung

Evaluating production planning and control (PPC) methods by means of simulation requires test data. Models of machine sequences of real productions can be created using machine learning. The paper describes how transformers and Bayesian networks are learned from real-life data of a manufacturing company. Both model types are well known for learning sequences with multiple conditional dependencies. These models are used to generate machine sequences which should be similar to the original machine sequences in terms of their statistical properties. From the generated machine sequences, work plans and production orders can be derived and used to test PPC methods by means of stochastic discrete-event simulation.