About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI-Enabled Materials Processing: Integrating Accelerated Experimental Workflows and Processing-Aware Machine Learning
|
| Presentation Title |
Real-time digital twins for open-die cold forging using machine learning surrogate models |
| Author(s) |
John S. Weeks, Aaron Stebner |
| On-Site Speaker (Planned) |
John S. Weeks |
| Abstract Scope |
Real-time digital twins for forging facilitate thermomechanical process control across large process parameter spaces, enabling enhanced material properties, reduced costs, and improved manufacturability. This work develops machine learning surrogate models that predict surface deformation and internal residual stresses of workpieces subject to pneumatic counterblow open-die cold forging. Neural network models are trained on finite element (FE) simulations that capture processing history and predict material states at 3000x lower computational cost than direct FE simulations. These models are integrated into an experimental system using time-history measurements of pneumatic piston positions. Surrogate models driven by experimental inputs serve as digital twins of the process and are evaluated in a model predictive control (MPC) framework for sequences up to 150 steps. The resulting closed-loop system is experimentally demonstrated on an aluminum alloy at rates of 10 hits/min using a 3-hit horizon MPC, enabling real-time control toward near-net shape and residual stress targets. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Process Technology, Shaping and Forming, Machine Learning |