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 |
A physics informed digital twin for powder bed fusion additive manufacturing via data assimilation and AI surrogates |
| Author(s) |
Daniel Wheeler, Dilip Banerjee, Shengyen Li |
| On-Site Speaker (Planned) |
Daniel Wheeler |
| Abstract Scope |
We present a digital twin (DT) for powder bed fusion, laser based additive manufacturing. Current approaches rely either on data driven AI models that overlook materials microstructure, or on high fidelity physics model lacking the uncertainty quantification necessary for credibility. Our approach bridges this gap by integrating physics based models (Adamantine, AdditiveFOAM, ExaCA) with Fourier Neural Operator surrogates to capture transient thermal behavior and microstructure evolution in IN718.
Due to inherent limitations in physics based surrogate models, periodic correction from experimental data is essential for DT alignment with its physical counterpart. The system state is realigned to AM Bench IN718 data using statistical data assimilation, and TorchDA enables latent space corrections of high dimensional state spaces. Process parameter estimates and uncertainties emerge as intrinsic by-products of this stochastic framework.
The combined framework, managed through automated workflow orchestration, provides a trusted foundation for PBF-LB AM design and certification. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, Computational Materials Science & Engineering, Machine Learning |