November 2022 Volume 4

FORGING RESEARCH

Summary and Conclusions While a production forging environment presents significant challenges to implementing Industry 4.0 technology, the prototype set-up demonstrates that AI/ML systems have the capability to accurately predict open die forged part attributes using data collected during processing. The generation of the training data for forging environments can be augmented with simulated data to improve the training capability of these systems. The ANN input data and predicted outputs can be utilized for machine monitoring and to create the digital twin. Furthermore, examining the ANN weights can help determine those sensors that are most influential to the prediction and thus what sensors can be eliminated. For further information regarding the project or to learn more about ANN application in a forging environment, please contact Prof. JosephDomblesky via email at joe.domblesky@marquette.edu or by telephone at 414-288-7832. Acknowledgment This research is sponsored by the Defense Logistics Agency InformationOperations, J68, Research&Development, Ft. Belvoir, VA, and by DLA-Troop Support, Philadelphia, PA. References 1. Dae-Cheol Ko, Dong-Hwan Kim, Byung-Min Kim,” Application of artificial neural network and Taguchi method to preform design in metal forming considering workability”, International Journal of Machine Tools and Manufacture, Volume 39, Issue 5, 1999, Pages 771-785. 2. Sumit Kumar, Anish Karmakar, Sumeer K. Nath, “Construction of hot deformation processing maps for 9Cr-1Mo steel through conventional and ANN approach”, Materials Today Communications, Volume 26, 2021, p. 101903. 3. W.L. Chan, M.W. Fu, J. Lu, An integrated FEM and ANN methodology for metal-formed product design, Engineering Applications of Artificial Intelligence, Volume 21, Issue 8, 2008, Pages 1170-1181. 4. Kingma, D. and Ba, J. (2015) Adam: A Method for Stochastic Optimization. Proceedings of the 3rd International Conference on Learning Representations (ICLR 2015).

Figure 6. Comparison of predicted and actual upset workpiece diameter for AA6061. The ANN was trained using DEFORM simulation data and tested on actual forged data.

Figure 7. Comparison of predicted and actual upset workpiece diameter for 1018. The ANN was trained using DEFORM simulation data and tested on actual forged data. While the current data used lab scale workpieces, the ANN can easily be modified to handle larger billet sizes and more complex geometries using the same procedure outlined in the article. While the ANN could also be modified to incorporate material data, the team opted to limit model complexity to establish the model’s robustness and capabilities to predict attributes.

FIA MAGAZINE | NOVEMBER 2022 81