Data-driven reduced-order modeling of zircaloy cladding deformation behavior under LOCA conditions

During a Loss-of-Coolant Accident (LOCA), fuel cladding may undergo high-temperature creep, oxidation, ballooning, and even rupture, potentially leading to the release of radioactive fission products into the primary circuit or beyond. High-fidelity fuel performance simulations are important for evaluating cladding behavior under LOCA conditions; however, their high computational cost limits their use for fast prediction, extensive parametric analysis, and full-core fuel safety assessment. This study develops a data-driven reduced-order modeling approach for rapid prediction of cladding deformation and thermal response under LOCA conditions. First, the OFFBEAT fuel performance code was used to simulate and verify both an out-of-pile LOCA cladding ballooning and failure test and an in-pile fuel rod behavior experiment. The verified modeling approach was then applied to a typical Pressurized Water Reactor (PWR) fuel rod under a hypothetical LOCA scenario. Full-order models (FOMs) were established for full-length fuel rod cladding under two representative power distribution profiles, focusing on cladding hoop strain and peak temperature. Based on the FOM results, a reduced-order model (ROM) was constructed using the Proper Orthogonal Decomposition with Interpolation (POD-I) method to predict unseen fuel-rod cases from limited data of representative rods. The ROM predictions show good agreement with the reference FOM solutions. The relative error of cladding hoop strain remains within 10% at all time steps, with an average relative error below 5%, while the predicted peak temperatures also agree closely with the reference values. Moreover, the computational cost is reduced from several hours for a single-rod FOM calculation to only a few minutes for ROM training and prediction. The results demonstrate the efficiency and reliability of the proposed data-driven ROM for rapid LOCA cladding behavior analysis.