Bayesian Optimisation for Minimising Tritium Losses Within the Hydrogen Isotope Separation System of the Fusion Fuel Cycle

Tritium self-sufficiency is a fundamental design requirement of a fusion fuel cycle, necessitated by the limited global availability of tritium relative to the fuelling demands of a fusion reactor. Minimising tritium losses within a fuel cycle is therefore essential. The Hydrogen Isotope Separation System (HISS) employs cryogenic distillation technology to remove excess protium and deuterium while rebalancing the deuterium–tritium (DT) mixture required for reactor operation. However, the HISS design involves a trade-off between reduced tritium emissions and increasing internal tritium inventory, both contributing to the overall tritium losses. In this work, a multi-objective Bayesian Optimisation (BO) framework based on an ε-constraint formulation is developed to construct Pareto-optimal solutions to compare alternative HISS architectures. Gaussian Process surrogate models derived from physics-based Aspen Plus simulations are used to resolve the non-linear relationships between design variables and performance metrics, including tritium inventory, tritium emission losses, and bottom-product purity. Application of the framework to representative case studies demonstrates that tritium emission losses significantly exceed tritium decay losses associated with internal inventory hold-ups across the investigated operating conditions. The proposed framework enables quantitative comparison of equilibrator integration strategies to compare HISS architectures and assess their impact on tritium losses within the fusion fuel cycle.