Harnessing machine learning for accurate estimation of concrete strength using non-destructive tests: a comparative study

Assessing the compressive strength of existing concrete structures is paramount for ensuring their safety and durability. Non-destructive testing (NDT) methods, while valuable, often have limitations in accurately predicting strength. The construction industry faces the challenge of evaluating the condition of existing concrete structures to determine the need for repairs or modifications. Accurate estimation of compressive strength is essential for informed decision-making. This study employed gene expression programming (GEP) to develop predictive models for estimating concrete compressive strength based on NDT measurements (Schmidt rebound hammer and ultrasonic pulse velocity). The dataset was divided into training and testing sets, and three GEP models were developed: GEP-I (rebound hammer), GEP-II (ultrasonic pulse velocity), and GEP-III (combined). The GEP models demonstrated superior performance compared to existing equations. GEP-I and GEP-III, which incorporated rebound hammer and/or ultrasonic pulse velocity, achieved higher accuracy, as evidenced by Pearson correlation coefficients (0.975 and 0.977), coefficients of determination (0.950 and 0.954), mean square errors (5.645 and 5.253), and mean absolute errors (1.772 and 1.802). The Taylor diagram further confirmed the superiority of the GEP models.