A Machine Learning Toolkit for Predicting Optimal Numerical Methods in NEAMS Tools

An important objective of the NEAMS program is to enable widespread use of the software tools among the industry, academia, and regulatory communities. For solving the problems occurring at various stages of NEAMS simulations, typically there are several possible choices for numerical sub- routines. Furthermore, the best method for a numerical problem may also evolve over the course of the simulation. The choice of the method can significantly enhance the portability of the NEAMS tools acroos a wide range of user base and computing platforms. RNET and UO will develop a machine learning plugin that automatically selects numerical methods) based on run-time dependent features of the problem and the hosting compute architecture with the goal of minimizing execution time or other objectives) over the course of a NEAMS simulation. The key features of this tool are automatically detecting change in problem characteristics in a long running nuclear simulation and the need for changing the numerical method, a minimal set of features to reduce the runtime overhead, runtime numerical method selection, and integration into the NEAMS toolkit to be released in 2018. The Phase I project will leverage the existing research on the applicability of machine learning to the selection of numerical solvers and demonstrate the applicability to a broad range of reactor simulations through various levels of solver selection with increasing data and computational complexity. The project will enable widespread use of the NEAMS tools among the industry as well as academia and government agencies. The end users of the tool are the non-expert users in the nuclear industry. The targeted customers include power companies, DoE agencies, and NASA divisions, DoD and its Prime Contractors, CFD software providers, oil and gas companies, and semiconductor design companies.