b'Accelerated NuclearCombinatorial-based predictions using machine learning can shorten the Materials and Fuelnuclear materials selection cycle by half.Qualification by Adopting aR esearchers developed and demonstrated combinatorial workflows for materials alloy design through combined experimental and modeling First-to-Failure Approach research. Modeling, high-throughput fabrication, and predictive data analytics shortened the development cycle for structural nuclear materials selection. The current development process for nuclear materials involves a series of time-intensive steps and testing cycles. These methods suffer from ineffective early down-selection, requiring additional testing and qualification. The combination of several established and emerging methodologies promises significant opportunities TOTAL APPROVED AMOUNT:for innovation that can shorten the cycle by as much as 50%. Using modeling and $1,540,442 over 3 years large-data analysis tools, researchers adopted a combinatorial methodology as a PROJECT NUMBER:proof-of-principle development of a new class of alloys. Employing this approach 19A39-012 hinged on the ability to organize, classify, and screen candidate materials and available data utilizing multi-scale modeling and large-data analysis, including PRINCIPAL INVESTIGATOR:deep, recursive, and transfer learning approaches. The approach began with Seongtae Kwon combinatorial multi-scale modeling and scaled into extensive data analysis to CO-INVESTIGATORS: identify candidate compositions for down-selection. The latter underwent a first-David Swank, INL to-failure testing regime based on defined attributes critical for success from an Jeffery Aguiar, INL early stage. Subsequent enhancements and comprehensive testing focused on a Yongfeng Zhang, INL limited number of alloys with less risk in less time, demonstrating a new approach Pratik Dholabhai, Rochester Institutefor material qualification. Each component of the method is individually novel. of Technology Combined, they promise significant opportunities for continued innovation and Taylor Sparks, University of Utah lasting impact in the broader materials community.Tolga Tasdizen, University of UtahCOLLABORATORS:CalNano, Inc.Sandia National LaboratoryUniversity of MichiganMachine learning offers excellent prediction accuracy. These materials properties have one non-linear term (cohesive energy) and three linear features (radius, valence electron concentration, shear modulus).22'