Validation of a general-use high flux isotope reactor–specific metaheuristic optimization framework for isotope production target design

Currently, advanced optimization methods are limited for isotope production (IP) campaigns at the US Department of Energy's High Flux Isotope Reactor (HFIR) located at Oak Ridge National Laboratory (ORNL), leading to years of conservative and historical approaches with minimal innovation. Moreover, the growing demand for new and existing isotopes is beginning to challenge the capacity of HFIR. This work explores the development and integration of metaheuristic (MH) optimization techniques for more efficient target design and irradiation strategies. As a test case, the optimization framework was applied to a routinely produced isotope at HFIR, ¹⁸⁸W, with the objective of maximizing the specific activity (SA), a key production metric. The framework includes Gnowee, a Python-based MH optimization algorithm, coupled with the Monte Carlo N-Particle version 6 (MCNP6) and Oak Ridge Isotope Generation (ORIGEN) activation/depletion/decay codes to design, simulate, and evaluate thousands of potential target design and irradiation scheme candidates. The framework relies on mock input files, design and irradiation variables for the algorithm to select, as well as a user-defined objective function to score each candidate based on the returned SA. Given the inherent complexities and computational time required when modeling and simulating the full HFIR model, a novel simplified MCNP6 model is presented in this article to increase the computational efficiency of the framework. The variables explored include irradiation location, number of cycles, and the number of W samples. Over 1,000 simplified model candidates were simulated in the same amount of time as a single full HFIR model run. By comparing the simplified model optimization's top candidate(s) with the full HFIR model results, the framework was verified to accurately explore the design space and converge on the top performing candidates. Lastly, past experimental data was compared to the data generated by the framework/model and both show that fewer W rings return higher SA, as expected. The verified and validated techniques provide a standardized solution to increase IP efficiencies by exploring thousands of unique target designs and irradiation strategies in a similar time as that required to run a single case in the full HFIR MCNP6 model. Both the novel simplified model and the full HFIR model show a more than 30% increase in SA if all presented modifications are applied to the current design and strategy. Thus, the objective of building a general-use, computationally efficient optimization framework for HFIR IP was accomplished, and has the potential to be applied to other IP campaigns.