Deterministic neutronics calculations in Griffin rely on multigroup neutron cross section libraries, which consist of databases of tabulated values, used to calculate the neutron cross sections through multivariate linear interpolation. However, the size of microscopic cross section libraries increases rapidly with the number of tabulations, requiring substantial memory and drastically slowing down the Griffin calculations when evaluating cross sections using the multivariate linear interpolation approach. Rising to these challenges, this work investigates constructing reduced-order models (ROMs) for multigroup microscopic cross sections to accelerate the cross-section evaluation in Griffin by collecting a database of multigroup cross sections considering all possible parameters designers could change during optimization studies. This work first presents a downselection of various ROM techniques jointly considering memory efficiency, predictive accuracy, computational cost, scalability, flexibility, and ease of implementation benchmarked against the multidimensional interpolation approach. Among all the ROM techniques, deep neural networks (DNNs) show exceptional memory efficiency, scalability, and flexibility while maintaining sufficient accuracy and robustness. Then we train DNNs for all the 273 possible isotopes in the depletion analysis for use in Griffin and developed a specific LibTorch interface in Griffin to enable cross-section evaluations by loading pretrained DNN models. Preliminary Griffin testing shows that DNNs exhibit exceptional predictive accuracy and provide orders of magnitude of improvement in memory efficiency compared to conventional interpolation techniques. These ROM techniques hold great promise to further increase the fidelity of the pebble-bed reactor (PBR) simulation by increasing the number of tabulations and state variables during cross-section evaluation, while keeping the computational cost affordable in Griffin. The current work lays the foundation for the online generation of cross-section data and concurrent update of the DNN ROM as new cross sections are generated. The constant generation of cross sections is necessary to account for self-shielding effects as the nuclide number densities in the system change over time (i.e., burnup).
