Modeling full-lifetime swelling in U-Mo dispersion fuel: A physics-constrained data-driven calibration framework

Uranium-molybdenum (U-Mo) dispersion fuel is a promising candidate for high-performance research reactors, but its deployment is limited by complex breakaway swelling at high burn-up, which remains elusive. Here, a multi-physics, multi-stage modeling framework is developed to analyze the full-lifetime behavior of U-Mo fuel, from initial progressive swelling to catastrophic failure. The model integrates three interconnected processes, encompassing (i) fission gas kinetics within U-Mo particles, incorporating a recalibrated irradiation-induced recrystallization model; (ii) gas transport across the evolving interaction layer (IL) via a moving-boundary reaction-diffusion approach; and (iii) a semi-phenomenological model for interfacial evolution and a progressive failure module triggered by a composite criterion. Parameter uncertainty is systematically addressed through a data-driven calibration strategy, combining Latin Hypercube Sampling (LHS) for global exploration and Bayesian Optimization (BO) for refinement, guided by a physics-constrained composite objective function. The calibrated model reproduces experimental swelling data from the E-FUTURE program, demonstrating consistent performance across the progressive swelling regime. Furthermore, the framework provides a basis for identifying the onset of breakaway swelling and reproduces the subsequent accelerated evolution through its coupled failure module. Results reveal an evolutionary swelling sequence that progresses from an initial intragranular-dominated regime to an intergranular-dominated phase following recrystallization, and finally to breakaway swelling driven by IL and interfacial failure. This framework provides a systematically calibrated tool for analyzing U-Mo fuel failure within the present fuel-system scope, offers insight into the underlying physical mechanisms, and may also serve as a useful basis for analyzing other advanced nuclear fuels when supported by system-specific parameter identification.