Evaluation of Residual Strength of Corroded Reinforced Concrete Beams Using Machine Learning Models

One of the main causes of structural durability degradation of reinforced concrete structures is corrosion of reinforcing bars. Predicting the bearing capacity of corroded reinforced concrete beams has been examined from experimental and theoretical perspectives. Most of the research works have been done on using individual predicting models instead of exploring the capacity of ensemble models. This study employs various machine-learning models, including support vector regression, artificial neural network, generalized linear regression, classification and regression-based techniques, exhaustive Chi-squared automatic interaction detection, and ensemble inference models to predict the residual capacity of corroded reinforced concrete beams based on actual data. A dataset of 120 samples collected in Ho Chi Minh City, Vietnam, is used for constructing, validating, testing the proposed models. The experimental results and statistical tests show that the generalized linear regression is the best model among all considered single predictive models and the ensemble model of generalized linear regression and artificial neural network obtained the highest prediction performance in estimating residual strength. The contribution to the body of knowledge is the development of ensemble models and various individual models that can predict the residual capacity of corroded reinforced concrete beams in a short time. This study demonstrates an effective prediction application for early structural durability estimation in the planning of building maintenance.