Probabilistic estimation of flexural loading capacity of existing RC structures based on observational corrosion-induced crack width distribution using machine learning

Corrosion-induced crack width can provide effective information on the deterioration level of in situ corroded reinforced concrete (RC) structures. However, the uncertainty associated with the relationship between corrosion-induced crack width and steel corrosion in RC structures is quite large. In this study, a probabilistic framework for estimating the structural capacity of corroded RC structures using the observational corrosion-induced crack width distribution and machine learning is proposed. Based on the experimental results of corroded RC beams, random field theory, and finite element (FE) analysis, artificial samples composed of corrosion-induced crack width and steel weight loss distributions over the RC beams are generated. Two machine learning-based models are then developed using these samples to estimate the steel weight loss distribution from the observational corrosion-induced crack distribution. Finally, a Monte Carlo-based FE analysis with the estimated steel weight loss distribution as the input data is conducted to obtain the probability density function (PDF) of the structural capacity of corroded RC beams. For illustrative purposes, the effect of observational corrosion-induced crack width distribution on the PDF of flexural capacity of an existing RC beam is investigated using the proposed framework. The results show that the proposed framework using a machine learning-based model is a reliable and computationally efficient approach for estimating the structural capacity of corroded RC members and demonstrates the potential for assessing the deterioration condition of existing RC structures based on the corrosion-induced crack width distribution.