­New publication in International Journal of Advanced Manufacturing Technology

Benedikt Zipfel, Oğuzhan Ahmet Arık and Udo Buscher have published the article “A Hopfield neural network-based approach for parallel additive machine scheduling problems with due dates” in The International Journal of Advanced Manufacturing Technology.

Over the past years, additive manufacturing has moved beyond pure prototyping into series production. This shift also raises questions about how to design production processes efficiently. What makes the setting special is that several parts are built simultaneously on one build platform, and the choice of which parts are printed together determines how long the entire print job takes. Planning therefore has to decide which parts are produced together, on which of several different powder bed machines this happens, and in what order.

The work centers on balancing the interests of production and customers. While customers expect deliveries to be as punctual as possible, production depends on consistently high utilization of its resources. The authors show that these goals can conflict noticeably, and they therefore take both into account at the same time. Methodologically, they combine a constructive heuristic with an improvement procedure based on a Hopfield network. This type of neural network had not previously been used for planning in additive manufacturing. A comparison with an exact optimization model and a genetic algorithm on 720 test instances shows that the new approach delivers better schedules, particularly for large orders and tight machine capacities. On smaller problems, it keeps pace with the genetic algorithm.