Machine learning approach for predicting early-age thermal cracking potential in concrete bridge piers

Abstract

In concrete construction, early-age thermal cracks in foundations, abutments, piers, and slabs can arise from non-uniform temperature distribution due to heat from cement hydration. These cracks negatively impact the integrity, load-bearing capacity, and service life of the concrete structures. This paper investigates the application of machine learning (ML) models to predict early-age thermal cracking in concrete bridge piers. The study aims to develop models to forecast thermal cracking potential (η_max) and estimate the timing of potential cracking (t) based on a dataset of various cross-sectional bridge piers and typical tropical temperatures. Four ML models—Support Vector Machine (SVM), Extreme Gradient Boosting (XGB), Artificial Neural Network (ANN), and Genetic Programming (GP)—were trained on 759 samples. The dataset, prepared using the EACTSA program, included parameters like cross-sectional dimensions, ambient temperature, and initial concrete temperature, with η_max and t as outputs. Results show that all the ML models achieved high prediction accuracy with R² scores over 0.96. The GP symbolic equations offer transparency and practical implementation. Compared to conventional methods, ML models provide a rapid, effective tool to optimize concrete member dimensions, formwork removal timing, and control concrete temperature, mitigating early-age thermal cracking risk.