Machine learning for assessing chloride resistance of concrete

Predicting the chloride resistance property of concrete accurately is critical in structural engineering. This thesis project adopts a state-of-the-art machine learning algorithm, XGBoost, to predict the chloride migration coefficient (Dnssm) of concrete. An extensive database of experimental data covering various concrete types has been compiled from research projects and previously published studies. Depending on the number and type of input features, four Dnssm prediction models are developed. All models are verified with unseen data using four statistical performance indicators and compared to other five tree-based algorithms, which are Decision Tree, Random Forest, AdaBoost, Gradient Boosting, and Bagging. The verification results confirm that the XGBoost model accurately predicts the Dnssm. The model is indispensable in practice as engineers around the world can use it to assess the performance of newly designed concrete against chloride resistance. It also has economic implications as it helps to design durable concrete mixes without the need for time-consuming and resource-intensive advanced laboratory testing. The model could also improve environmental performance by reducing precautionary overdesign of concrete properties and saving natural resources that would otherwise be wasted, and thus contributing to the achievement of the Sustainable Development Goal (SDG 13)