Current CNERG Projects

Recent growth in computing now permits large-scale multiphysics modeling using Monte Carlo methods, considered state-of-the-art for particle transport. The Monte Carlo method is well-suited for benchmarking lower-fidelity tools, guiding multigroup cross section library development, and aspects of reactor design. However, Monte Carlo methods remain expensive, especially in multiphysics workflows where simulations are performed iteratively. Significant expertise is also needed to pre-divide phase space for scoring tallies, taking into account competing effects including spatial detail, runtime, memory usage, and propagation of truncation/statistical errors to coupled physics solvers. Adaptive mesh refinement (AMR) is a technique which adaptively refines (or coarsens) a mesh to preferentially add Degrees of Freedom (DOFs) where the error is highest, and remove DOFs where the solution is already well-captured. Despite the remarkable success of AMR in other fields, there has been very limited use of AMR for Monte Carlo tally meshes, especially in the context of multiphysics or in combination with hand p-refinement. This proposal will develop the fundamental methods and techniques for unstructured mesh AMR with Monte Carlo tallies to enable a transformative leap forward in speed, accuracy, and robustness of Monte Carlo methods for advanced simulation.

• Jordan R. Stomps, Paul P. H. Wilson, Kenneth J. Dayman, "Contrastive Machine Learning with Gamma Spectroscopy Data Augmentations for Detecting Shielded Radiological Material Transfers", (2024/1)