Rapid on-site prediction of concrete compressive strength using integrated machine learning and non-destructive testing

Abstract

Accurate prediction of concrete compressive strength (f_c) is crucial for construction quality control and structural safety assessment. While non-destructive testing (NDT) methods offer practical evaluation, combining them with mix parameters can enhance prediction accuracy. This study aims to develop a robust machine learning (ML) framework integrating NDT methods rebound hammer (RH) and ultrasonic pulse velocity (UPV) with mix design parameters and curing age precise early strength prediction. An experimental dataset of 360 samples from 24 concrete mixtures (M20-M35) was collected, covering curing ages from 3 to 56 days. Six machine learning (ML) models were evaluated using comprehensive mix parameters and NDT measurements. The extra tree regressor (ETR) outperformed other models, achieving exceptional accuracy (R² =0.99, RMSE = 0.53). SHAP analysis revealed curing age as the most influential parameter, followed by water- cement ratio and NDT measurements. The proposed model demonstrates promising strength prediction capability as early as 3 days under laboratory conditions. The data-driven approach may contribute to improved quality assessment and construction planning and shows potential for future practical applications following validation using field data.