Machine Learning Methods for Prediction of Changes in Material Flow Networks

Abstract

New trends like cloud manufacturing, urban manufacturing, and additive manufacturing cause future manufacturing systems to be more decentralized and dynamic. These dynamics not only relate to the actual manufacturing process, but also to the infrastructure of networked machines, which is subject to constant change and reconfiguration. Forecasting this infrastructural change is vital to efficient mid-and long-term planning activities in such material flow networks. Therefore, we propose the application of machine learning methods for the prediction of changes in material flow networks based on readily available data about material flow, e.g., feedback data from manufacturing execution systems. State-of-the-art network embedding methods, such as Node2Vec, DeepWalk, matrix-and tensor factorization, are adapted for the manufacturing environment and subsequently evaluated using a real-world dataset.