Methodology for Improving Accuracy of Nuclear Industry Codes based on VERA

Industry core simulators that are the cornerstone of reactor analysis for the operating fleet of nuclear plants are used in numerous application areas including core and fuel design, safety licensing, core monitoring, and operator training. Current licensed industry codes, while capable of reconstructing fuel rod detail for local power, suffer from a number of geometrical and physics simplifications that require additional margin in calculated operational and safety parameters. This excess conservatism results in more restrictive thermal, chemistry, and reactivity operating limits that ultimately result in higher fuel and operating costs for the reactor fleet. In addition, the introduction of new fuel products such as higher burnup, high enriched fuel, new cladding, and new fuel forms is hampered by assumptions and approximations built into existing industry simulation methods. Advances in modeling and simulation for nuclear reactors and the use of high-performance computing has been shown to provide unprecedented accuracy in predicting core and fuel rod behavior. These advances are well-positioned to identify and utilize the excess conservatism that exists today to improve the efficiency of the current and future light water reactor operating fleet, providing increased electrical output and decreased costs. Across the current national fleet of 93 reactors, eliminating conservatism in a handful of application areas through use of advanced modeling and simulation is estimated to have value of over $300 million per year. This includes providing industry solutions characterized by highly resolved multi-physics coupling and high accuracy in areas such fuel crud risk management, fuel performance during operational maneuvering, bounding licensing analyses such as reactivity insertion accident, and plant lifetime extension. This proposal seeks to address the two barriers to adoption of advanced modeling and simulation for the nuclear industry. First, is the high computational resource required, on the order of thousands of processors, which does not lend itself to production computing in many use cases. An example is reactor core design optimization that typically requires thousands of solution evaluations using the licensed industry simulator. Second, is the need to license an advanced simulator with the regulator which is costly in both resource and schedule from the viewpoint of the payback period. The proposal will focus on development of a methodology to characterize and correct, using machine learning, the model errors in industry codes such that robust agreement with high fidelity, advanced simulation is achieved. In phase 1, a methodology based on machine learning and the use of advanced modeling capabilities will be developed. This will focus on the error associated with spectral history effects inherent within the standard two-step approach used by industry codes for nuclear cross section generation. Phase 2 will focus on tightly integrating this methodology within the framework of a licensed industry core simulator. This would entail working closely with an industry partner to adapt the machine learning methodology to the industry simulator cross section generation workflow. Validation against historical operating reactors will then be performed to demonstrate the predictive accuracy of the new method.