...provide for safe, secure long term storage of nuclear waste in the United States, one solution is to properly sequester the waste in subsurface geological formations. The Department of Energy has embarked on producing a simulation suite designed to help manage the risk of this by simulating the complex physical processes that occur over very long periods in underground storage. The physics and chemistry needed to adequately capture these processes can require very large investments in computational resources, and methods for reducing this burden are being sought. How This Problem is Being Addressed: Illinois Rocstar proposes to leverage state-of-the-art machine learning models and physics-based reduced order modeling capabilities to produce representations of subsurface reactive transport that will run quickly and produce accurate results for nuclear waste repository simulation. The proposed tool will both provide a framework within which users can produce their own new reduced-order models using these techniques, and will also provide for integration with the government repository simulation system under development. What Will be Done in Phase I: During Phase I, subsurface simulations of radionuclide transport in the subsurface will be modeled as a series of 1D, 2D, and finally 3D scenarios, building up consistently more capable reduced-order models. Techniques to be used are based on proper orthogonal decomposition, with a parameterized closure based on advanced machine learning models. The 2D and 3D simulations will use the governments developing system for nuclear waste repository characterization so that the resulting models will be compatible and consistent with those advanced multiphysics modeling tools. A connection back to the government system will be made through production of a new reduced-order process model to be embedded in the reactive flow solver system. Commercial Applications and Other
Benefits: Application of the system to be constructed will first be in the nuclear waste repository modeling and simulation area. However, the reactive transport reduced-order-models that will be generated should be able to be used in other subsurface modeling and simulation areas as well, such as to model geologic carbon storage sites, environmental restoration or accident sites, or possibly for oil and gas exploration. Ultimately applications to site characterization, site monitoring, accident analysis, and permitting applications are possible.
