Machine Learning for Manufacturing Processes

Machine learning and the application of intelligent systems is significantly changing the view on existing problems in many application areas, including manufacturing. As experts in the field of material characterization, modeling and simulation of manufacturing processes, we investigate the use of machine learning for solving common and future problems in industry. The main advantage of machine learning over numerical simulations is its ability to provide real-time predictions. We use this to solve typical optimization problems in a time-efficient manner, such as the calibration of material models. We also use machine learning to solve complex problems, such as the design of materials and processes. Moreover, machine learning enables us to digitally represent processes and components based on numerical simulations in real-time including complex material behavior. For the development of machine learning models we use experimental data as well as simulation data and rely on the integration of expert knowledge.

Our services

Fig. 1.1: Procedure to create a machine learning-based transfer function for material model parameter identification
© Fraunhofer IWM
Time-efficient identification of material model parameters using machine learning.
  • Development of machine learning models for time-efficient calibration of material models. The identification of parameters of complex material models is very time-consuming and can take from a few hours to several days. A trained machine learning model can greatly accelerate this process. This works either by learning the behavior of the material model and identifying the material model parameters within an optimization or by directly learning the inverse relationship between material model response and material model parameters.
  • Development of machine learning-based process models, so-called surrogate models, taking into account the typically complex material behavior. Machine learning models are used to learn the behavior of a component or process simulation, providing a real-time capable digital twin of the process or component. Based on these models, variations of process and material parameters can be efficiently analyzed and their effect on the process result can be optimized.
  • Proof-of-concepts to evaluate the applicability and benefit of machine learning in the respective context.

to top

Main research topics

Our research in the field of machine learning concentrates on

  • Time efficient parameter identification of material models
  • Solving material design problems, including microstructure optimization
  • Surrogate modeling of process and component simulations as well as solving process optimization problems
  • Development of soft sensors for the online availability of information on the state of a material within a process

to top



  • Shoghi, R.; Morand, L.; Helm, D.; Hartmaier, A., Optimizing machine learning yield functions using query-by-committee for support vector classification with a dynamic stopping criterion, Computational Mechanics,  Online First (2024) 20 pp. Link
  • Wessel, A.; Morand, L.; Helm, D.; Volk, W., Machine learning-based sampling of virtual experiments within the full stress state, International Journal of Mechanical Sciences 275 (2024) Art. 109307, 19 pp. Link
  • Dornheim, J.; Morand, L.; Janarthanam, H. N.; Helm, D., Neural networks for constitutive modeling: from universal function approximators to advanced modeling and the integration of physics, Archives of Comptational Methods in Engineering Online First (2023) 31 pp. Link
  • Tarek, I.; Morand, L.; Dornheim, J.; Link, N.; Helm, D., A multi-task learning-based optimization approach for finding diverse sets of microstructures with desired properties, Journal of Intelligent Manufacturing, Online first (2023) 17 pp. Link
  • Dornheim, J.; Morand, L.; Zeitvogel, S.; Iraki,T.; Link, N.; Helm, D., Deep reinforcement learning methods for structure-guided processing path optimization, Journal of Intelligent Manufacturing 33 (2022) 333-352 Link
  • Kurnatowski, M. von; Schmid, J.; Link, P.; Zache, R.; Morand, L.; Kraft, T.; Schmidt, I.; Schwientek, J.; Stoll, A., Compensating data shortages in manufacturing with monotonicity knowledge, Algorithms 14/12 (2021) Art. 345, 18 pp. Link
  • Wessel, A.; Morand, L.; Butz, A.; Helm, D.; Volk, W., A new machine learning based method for sampling virtual experiments and its effect on the parameter identification for anisotropic yield models, IOP Conference Series: Materials Science and Engineering Vol. 1157, 40th International Deep-Drawing Research Group Conference IDDRG 2021; Liewald, M.; Karadogan, C. (Eds.); IOP Publishing Ltd, Bristol, UK (2021) Art. 012026, 10 pp. Link
  • Morand, L.; Link, N.; Iraki, T.; Dornheim, J.; Helm, D., Efficient exploration of microstructure-property spaces via active learning, Frontiers in Materials, Computational Materials Science 8 (2022) Art. 824441, 12 pp. Link
  • Morand, L.; Helm, D., A mixture of experts approach to handle ambiguities in parameter identification problems in material modeling, Computational Materials Science 167 (2019) 85-91 Link
  • Morand, L.; Helm, D.; Iza-Teran, R.; Garcke, J., A knowledge-based surrogate modeling approach for cup drawing with limited data, IOP Conference Series: Materials Science and Engineering Vol. 651/1; 38th International Deep Drawing Research Group Annual Conference IDDRG 2019; van den Boogaard, T.; Langerak, N. (Eds.); IOP Publishing Ltd., Bristol, UK (2019) 012047 1-8 Link

to top