Machine learning prediction of concrete compressive strength using rebound hammer test

Machine learning has become a key branch in artificial intelligence by providing unique pre-dictive modeling solutions. Predicting the compressive strength of concrete determined using non-destructive test techniques (NDT) includes high levels of uncertainty. This uncertainty directly depends on the repeatability of the measurement and the variability of concrete prop-erties. This study aims to evaluate the effect of mixture composition and age of concrete on the coefficient of variation (CV) of the rebound hammer index applied to various types of concrete. Several supervised machine learning models, including multivariate multiple regression (MMR), support vector machine (SVM), Gaussian process regression (GPR), and Regression tree (RT) were utilized to predict the compressive strength of concrete. A large dataset of 468 cubic concrete specimens was sorted into four categories and employed for simulation. Regardless of the selected dataset, it was concluded that GPR/SVM and RT yielded the most accurate model prediction metrics of compressive strength when using rebound hammer records over MMR model. The results of the adopted models were remarkably better when mixture proportion and age of concrete features (i.e., age and w/p) were considered in the simulation.

Automated summary

This study evaluates the effect of mixture composition and age of concrete on the coefficient of variation of the rebound hammer index. Several machine learning models are utilized to predict the compressive strength of concrete. The results show that GPR/SVM and RT models yield the most accurate predictions when considering both mixture proportion and age of concrete.