Efficient Computational Paradigm for Multiphysics-Driven Stochastic Corrosion in Nuclear Waste Disposal

The long-term safety of deep geological disposal systems relies on the structural integrity of metallic canisters that isolate high-level nuclear waste from the biosphere. These canisters are continuously exposed to complex subsurface thermo-hydro-chemical (THC) environments, and their interfacial corrosion behavior exhibits significant uncertainty due to spatially variable conditions, heterogeneous material properties and poorly constrained environmental parameters. Quantifying these uncertainties is essential for reliable prediction of corrosion progression over geological time scales. However, from a numerical perspective, current research still lacks efficient methods capable of handling large spatial domains, high-dimensional random inputs and long simulation durations. This study presents an efficient stochastic finite element method for performing the probabilistic evolution of corrosion depth at canister interfaces. The method solves a set of coupled stochastic THC PDEs and treats the corrosion interface as an evolving random field. Random initial conditions, uncertain material-environment coefficients and variability in corrosion mechanisms are represented systematically through (high-dimensional) random inputs. The proposed method approximates stochastic solutions by separating random, spatial and temporal components, enabling scalable computation on large-scale domains and extended time horizons. This leads to substantial reductions in computational cost while maintaining accuracy in evaluating probability distributions and ensemble statistics of corrosion evolution. The resulting full stochastic solutions further provide a robust basis for reliability assessment, parameter sensitivity analysis and inverse identification of key corrosion-controlling parameters, thereby supporting safety evaluations of deep geological disposal systems.