Abstract
Currently, no ASME BPV Section III code-qualified structural materials exist for nuclear fusion. Blanket modules endure extreme thermo-mechanical loads, heat fluxes, and irradiation. As part of the DOE FIRE IMPACT program, this work aims to accelerate the adoption of advanced reduced-activation high-temperature materials. ASME allows inelastic analysis for Class A components to optimize designs, requiring a high-temperature inelastic constitutive model that captures the material's microstructural response.
A major challenge is that the physics governing body-centered-cubic (BCC) dislocation mobility---specifically thermally activated kink-pair nucleation---makes the governing differential equations stiff and non-linear. Additionally, these models rely on up to 40 parameters. Because several phenomenological parameters cannot be derived from lower-scale modeling, robust non-linear fitting tools are essential.
This work details automatic multi-parameter fitting using deterministic and stochastic gradient-based optimization for an inelastic model of Reduced-Activation Ferritic-Martensitic (RAFM) steel. Utilizing the NEML2 framework, the model leverages PyTorch automatic differentiation to bypass analytical derivations, efficiently computing the gradient of the loss function with respect to the parameters of interest. While optimizing parameters against Eurofer97 tensile data successfully, the isolated BCC mobility formulation under-predicted athermal resistance and flow stress. This proves that capturing true yielding and strain hardening requires explicitly including hierarchical micro-structural length scales, like sub-grain boundaries and precipitates. Ultimately, this thesis establishes a robust optimization pipeline and clear trajectory for future fusion material qualification.