Analyzing chloride diffusion for durability predictions of concrete using contemporary machine learning strategies

Abstract

Analyzing the migration of chloride ions in concrete can enhance its durability and decrease the probability of corrosion. Furthermore, machine learning (ML) offers a promising, cost-effective, and less intricate alternative to traditional experimental methods for accurately forecasting cementitious composites' chloride migration coefficient (Dnssm). This study utilized advanced ML methods like support vector machine (SVM), artificial neural network (ANN), and k-nearest neighbor (KNN) to forecast the Dnssm of concrete accurately. A comprehensive database was established considering significant input features that can impact Dnssm. The models undergo verification with unseen data, utilizing various statistical performance measures. The verification findings validate that all three models (SVM, KNN, ANN) accurately predicted the Dnssm with high precision. However, the SVM model outperformed with a higher coefficient of determination (R2) score of 0.90, whereas ANN and KNN yielded R2 scores of 0.88 and 0.83, respectively. The model has the capacity to substitute laborious, resource-intensive and time-consuming lab testing. Moreover, the RReliefF analysis indicated that water/binder and cement type are the prominent input factors affecting the target variable Dnssm.