Code and Data: Machine learning surrogate model-based optimization of multi-hit sequences in open-die cold forging

Code and data for "Machine Learning Surrogate Model-Based Optimization of Multi-Hit Sequences in Open-Die Cold Forging" by John S. Weeks, Ahmet B. Coksaygili, and Aaron P. Stebner at Georgia Institute of Technology. Submitted for publication in Integrating Materials and Manufacturing Innovation. Finite element simulation-based optimization frameworks are well established to design short sequence forging processes to achieve near-net shapes. However, multi-objective optimization of long sequences to meet shape and material property metrics remains an open challenge. Toward meeting this challenge, a machine learning surrogate modeling framework is developed for optimization of long counterblow cold forging sequences of Al6063-T52 to meet final shape and principal stress objectives. More specifically, neural network surrogate models are trained to approximate finite element forging simulations using a strain-rate dependent elastic-plastic constitutive model. Surrogate models exhibit mean absolute errors of 4 $\mu$m for surface displacements and 6 MPa for maximum principal stress at $3000\times$ reduced computational cost of the finite element simulations. A genetic algorithm is used as the optimization engine to call the surrogate models to optimize forging sequences comprised of $100-150$ hits to meet both cross-section shape and minimum residual stress metrics. When compared with short-horizon model predictive control sequences, the optimized sequences realize a $17\%$ to $32\%$ reduction in peak stress for a similar final geometry. These results demonstrate that surrogate model-based optimization enables predictive planning of long sequence cold forging to meet both stress and shape objectives. The fast computation times and increased multi-objective design space provide feasibility for real-time control and digital twin integration in future forging applications, including complex shapes and graded mechanical properties.