About this Abstract |
| Meeting |
2026 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2026)
|
| Symposium
|
2026 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2026)
|
| Presentation Title |
Physics-Preserving Graph Neural Network Surrogates for Thermomechanical Field Prediction: Toward Real-Time Digital Twins in Metal AM |
| Author(s) |
Usman Tariq, Frank Liou |
| On-Site Speaker (Planned) |
Usman Tariq |
| Abstract Scope |
Metal additive manufacturing holds promise for aerospace and defense, yet qualification remains slow due to the high cost of physics-based simulation. Finite element analysis of a directed energy deposition build can take hours to days, making it incompatible with real-time process control. This work develops a two-stage physics-aware graph neural network surrogate that inherits the accuracy of validated multiphysics finite element models at a fraction of their cost. The mesh is converted into a graph with physics-informed node features including laser proximity, boundary distance, and element birth timing. A DeeperGCN thermal model predicts full-field nodal temperature across deposition time steps. A recurrent graph neural network then maps the predicted thermal history to full-field residual stress and displacement, preserving the causal link between thermal and mechanical evolution. The framework generalizes to unseen built geometries without retraining. Ongoing work integrates in-situ sensor data toward real-time digital twin deployment for additive manufacturing qualification. |
| Proceedings Inclusion? |
Undecided |