About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Surrogate modeling of creep in nuclear materials using mixed adaptive basis functions with spatially varying parameters |
| Author(s) |
Karl Garbrecht, Andre Ruybalid, Laurent Capolungo |
| On-Site Speaker (Planned) |
Karl Garbrecht |
| Abstract Scope |
Due to extreme in-service conditions, nuclear materials exhibit complex creep behavior. This behavior can be represented by polycrystalline scale mechanistic models; however, these models are computationally expensive in engineering scale finite element simulations. To upscale mechanistic simulations, a nonlinear regression framework based on piecewise adaptive basis functions is developed for surrogate model construction. The method accommodates mixed basis forms with spatially varying parameters (e.g., powers, frequency multipliers, and weights). The continuous parameter fields are defined by linearly interpolated nodal values over global or partitioned domains. This constraint on the fields establishes a tractable optimization problem while enabling relatively simple basis functions to capture nonlinear behaviors. The approach is applied to construct creep surrogate models for Grade 91 steel exposed to high temperature and irradiation using data from mechanistic models. The surrogate models provide sufficient accuracy for incremental prediction of stress-strain response and microstructure evolution within a time-marched framework. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Other, Nuclear Materials, Machine Learning |