ISSN: 2717-7203
Data-driven optimization of the surface integrity of cold forming punch depending on CNC turning parameters
1Nedu Bağlantı Elemanları San. Tic. A.Ş., Norm Tooling, R&D Center, İzmir, Türkiye; İzmir Kâtip Çelebi University, Department of Mechanical Engineering, İzmir, Türkiye
2Nedu Bağlantı Elemanları San. Tic. A.Ş., Norm Tooling, R&D Center, İzmir, Türkiye
J. Adv. Manuf. Eng. 57-69 DOI: 10.14744/ytu.jame.2026.00006

Abstract

Cold forming dies and tools, manufactured through various machining processes, have complex geometries and require high dimensional accuracy and surface integrity. Their surface quality highly depends on process parameters. Because the interplay among parameters is complex, optimizing machining processes is challenging. In that manner, data-driven approaches are significant for handling complex parameter interactions, thereby reducing time and cost. This study presents a data-driven framework for optimizing the surface roughness of a cold forming punch depending on the CNC turning process. A full factorial experimental design with three levels per parameter was used, yielding 27 combinations. The collected data were modeled using distinct functional types by a multiple nonlinear neuro-regression approach to achieve a robust model. The Ra values of different regions of the cold-forming punch were prioritized as the objective function. At the same time, the responses Rz and Rmax were used to define physical constraints during optimization. Modified versions of Simulated Annealing, Nelder-Mead, Differential Evolution, and Random Search algorithms were used to solve the optimization problem. The robustness of the models and the optimization results was validated experimentally by conducting CNC machining of a cold forming punch. Ra, Rz, and Rmax were found to be lower than 0.2 µm, 1 µm, and 1.30 µm, respectively, in all optimization scenarios across all regions of the forming punch tip. Results revealed that the modeling and experimental outcomes are in good agreement, with over 75% success rate. The suggested method shows great promise for adoption across various production domains and is especially beneficial for high-precision manufacturing, as this study includes an industrial case.