Optimizing durability assessment: Machine learning models for depth of wear of environmentally-friendly concrete

Abstract

The use of fly ash in cementitious composites has gained popularity. However, assessing the depth of wear (DW) of concrete requires expensive and destructive laboratory tests utilizing specialized equipment like the rotating-cutter method. Therefore, there is a need for alternative methods to predict the depth of wear of such composites more efficiently and cost-effectively. Accordingly, the objective of this research is to utilize machine learning (ML) approaches, including one individual algorithm (Decision Tree) and two ensemble algorithms (AdaBoost Regressor and Bagging Regressor) to estimate the depth of wear of fly-ash-based concrete. A collection of 216 experimental records was obtained from the existing literature. The efficiency of the models was examined with multiple statistical indexes. The bagging regressor (BR) model provided superior estimation performance with a correlation coefficient (R) of 0.999 compared to AdaBoost regressor (R = 0.965) and decision tree (R = 0.962). The ensemble models, notably BR, provided more accurate predictions with an 87.8 % lower mean absolute error (MAE) and an 85 % lower root mean square error (RMSE) compared to the decision tree model. In addition, the BR model exhibited the lowest performance index (ρ) values of 0.016 for training and 0.012 for validation. The SHapley Additive exPlanation (SHAP) revealed that the time of testing and age are the most dominant controlling features that significantly contribute to the estimation of the depth of wear. In conclusion, utilizing ML techniques and SHAP interpretation to estimate the DW of fly ash concrete significantly reduces reliance on expensive lab tests, making durability assessment more practical and cost-effective.