Time-Dependent Reliability Assessment of Corroded Steel Structures Using a Hybrid Monte Carlo–Random Survival Forest Framework

Reliability assessment plays a critical role in the design, evaluation, and life-cycle management of steel structures by providing a quantitative measure of safety that explicitly accounts for uncertainties in loads, material properties, geometry, and modeling assumptions. This research proposes a hybrid computational framework integrating Monte Carlo (MC) simulation with the random survival forest (RSF) model to evaluate the time-dependent reliability of corroded steel structures. The MC method is first employed to generate a physically consistent training dataset that captures the stochastic evolution of corrosion, sectional degradation, and failure times of steel beams under random loads and environmental effects. The resulting dataset, comprising physical random variables, failure times, and survival states, is then used to train the RSF model. The trained RSF learns the conditional survival function, representing the probability that a structural component remains safe at a given time for a specific combination of input parameters. The results demonstrate that the RSF model successfully reproduces the reliability distribution predicted by MC simulation with negligible errors while reducing computational cost by more than two orders of magnitude. This hybrid approach provides an efficient, data-driven pathway for time-dependent structural reliability modeling and can be extended to remaining service-life prediction and sensitivity analysis under complex corrosion environments.