A data-based and physics-informed Bayesian updating method for deterioration models of RC structures in marine environment

Deterioration models of reinforced concrete (RC) in marine environment, to be applied to existing structures, usually need to be calibrated with long-term in-situ data. The Bayesian model updating provides a framework to incorporate measured data into existing models to make them more realistic. The measured chloride concentrations of concrete at different depths are related to each other through the Fick's second law, but they are treated as independent in existing updating methods, which affects the accuracy of model updating. To solve this issue, this paper proposes a data-based and physics-informed (DBPI) likelihood function to incorporate the physical law behind the measured data into Bayesian updating framework, whose validity is first confirmed through numerical examples, and then it is applied to the durability assessment of an existing wharf structure in marine environment. The parameters involved in the chloride ingress model and the critical chloride concentration model are updated using the data from durability inspections. The durability performance of the structure is then assessed using the updated models, which is consistent with the actual surface deterioration observed in the two inspections. Discussion of the updated results reveals that ignoring the physical law behind the measured data may result in incorrect inferences of the chloride ingress model and multi-mode distribution of the updated parameters, which is solved by using the proposed DBPI likelihood function, and the accuracy of Bayesian updating is significantly improved.