Bellows expansion joints enhance the displacement performance of piping systems owing to their unique geometrical features. However, structural uncertainties such as wall thinning in convolutions, a byproduct of the manufacturing process, can impair their structural integrity. This study addresses such issues by conducting a global sensitivity analysis to assess the impact of these uncertainties on the performance of bellows expansion joints under monotonic loading. Global sensitivity analysis, which examines main and nth order interaction effects, is computationally expensive. To mitigate this, we employed a surrogate model-based approach using an artificial neural network. This model demonstrated robust prediction capabilities, as evidenced by metrics such as the coefficient of determination. The sensitivity indices of the main effect for the 2-ply and 3-ply bellows at the sixth convolution were 0.3340 and 0.3233, respectively. The sensitivity index of the sixth convolution was larger than that of other convolutions because the maximum deformation of the bellows expansion joint under monotonic bending load occurs around it. Interestingly, the sensitivity index for the interaction effect was negligible (0.01%) compared to the main effect, suggesting minimal activity between uncertainty factors across convolutions. Notably, bellows expansion joints under repetitive loading exhibit more complex behaviors, with the initial leakage typically occurring at the convolution. Therefore, future studies should focus on the structural uncertainties of bellows expansion joints under cyclic loading and employ a surrogate model for comprehensive global sensitivity analysis.

To obtain confidence in the safety of disposal facilities for radioactive waste, it is essential to quantitatively evaluate the performance of the waste disposal facilities by using safety assessment models. Thus, safety assessment models require uncertainty management as a key part of the confidencebuilding process. In application to the numerical modelling, the global sensitivity analysis is widely employed for dealing with parametric and conceptual uncertainties. In particular, the parametric uncertainty can be effectively reduced by minimizing the uncertainty of critical parameters in the safety assessment model. In this paper, the numerical model of each step disposal facility (Silo, Near surface, and Trench type) at Wolsong Low and Immediate Level Waste (LILW) Disposal Center is designed by using a two-dimensional finite element code (COMSOL Multiphysics). In order to determine the critical parameters for non-adsorbed nuclides such as H-3, C-14, Tc-99, we introduced the variance-based sensitivity analysis methodology of the global sensitivity analysis. In the case of Silo type, the density of waste is highly sensitive to the total leakage quantity of all nuclides. Additionally, the initial nuclide concentration of H-3 was identified as another important parameter of H-3. On the other hands, the mass transport coefficient showed a high contribution in C-14 and Tc-99. In other types of disposal facilities, the leaking properties of H-3 are significantly affected by the amount of infiltration water. However, C-14 and Tc-99 were found to be more sensitive to the density of waste.