Evaluating concrete properties: integrating ultrasonic pulse velocity and machine learning models for strain prediction

This study aims to evaluate the mechanical properties of M30 grade concrete using both destructive and nondestructive methods. Ultrasonic pulse velocity (UPV) testing assesses concrete integrity, whereas compressive strength tests and machine learning models are used to predict strain and assess structural health. Machine learning models such as Elastic net, Lasso, Ridge, and Support vector regression (SVR) were used on a dataset of 472 samples with corresponding values of UPV, stress, and strain. Among all the regression algorithms, the Elastic net had the highest accuracy in training, with an R2 of 0.967, and the testing accuracy of R2 was 0.923. The results of this study revealed that UPVs greater than 3000 m/s corresponded to high-quality concrete, whereas UPVs less than this value corresponded to microcracks and a reduction in the strength of the concrete. This value replicates the IS 13,311 (Part 1- NDT- Methods of Test) parameters, which describe the methodology used to calculate the nondestructive testing parameters. This study demonstrates the effectiveness of machine learning, particularly Elastic Net, in predicting structural strain and health. The integration of UPVs with ML provides a reliable assessment of concrete quality, although short curing times, limited data, and no environmental variation limit the use of UPVs.