About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Modeling, AI Applications, and Method Development in Nuclear Materials
|
| Presentation Title |
Artificial Intelligence meets irradiation Physics: learning defects, entropy and free energies |
| Author(s) |
Mihai Cosmin Marinica, Thomas D Swinburne |
| On-Site Speaker (Planned) |
Mihai Cosmin Marinica |
| Abstract Scope |
Artificial intelligence is opening new routes to predict radiation damage in nuclear materials by exploiting not only machine-learning potentials, but also the high-dimensional descriptor spaces on which they are built. These internal representations encode structural, energetic and entropic information, and can therefore be used to identify defect populations, learn complex free-energy landscapes, and accelerate atomistic sampling.
We apply this viewpoint to the formation of dislocation loops in irradiated metals. Recent results show that self-interstitial atoms do not always aggregate directly into two-dimensional loops. Instead, they may first form compact three-dimensional Frank–Kasper nano-phases: C15 inclusions in bcc iron and A15 inclusions in fcc Al, Cu and Ni. These precursors influence loop density, morphology and irradiation-induced mechanical response.
By combining descriptor-space learning, entropy estimation and accelerated molecular dynamics, we propose an AI-driven framework to characterize, sample and predict defect evolution in irradiated metals. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Modeling and Simulation, Nuclear Materials |