Temporal Fusion Transformer Surrogate Modeling of Chloride-Induced Corrosion Initiation Probability in Reinforced Concrete Bridge Infrastructure

Abstract

Abstract

Chloride-induced corrosion initiation is a major durability concern in reinforced concrete bridge infrastructure and can affect long-term structural performance. Reliable estimation of corrosion initiation probability is therefore important for durability assessment and maintenance planning. This study investigates the time-dependent evolution of corrosion initiation probability under uncertain material, environmental, and geometric conditions, with particular attention to concrete cover depth. A diffusion-based stochastic reliability framework is used to generate simulator-derived corrosion initiation probability trajectories. Corrosion initiation is represented as a threshold-crossing event at the reinforcement depth, and population-level probability is estimated through stochastic simulation and ensemble aggregation. To reduce repeated simulation effort during inference, a Temporal Fusion Transformer surrogate model is trained to reproduce simulator-derived probability trajectories from scenario variables and time. The results show that the surrogate closely follows the simulator-derived evolution of corrosion initiation probability within the sampled scenario space. The predicted trajectories preserve the expected influence of cover depth, with shallower cover associated with earlier initiation and higher probability levels. The model also provides low trajectory-level error relative to the probability scale and offers an inference-time alternative to repeated simulation-based evaluation. These findings indicate that simulation-trained sequence learning can provide a useful surrogate representation of time-dependent corrosion initiation probability under the assumed diffusion–threshold reliability framework. Further validation with field or experimental corrosion data is needed before practical implementation in reinforced concrete bridge infrastructure systems.