Ultrasonic detection and deep learning for high-precision concrete strength prediction

Xuehong Gan, Wei Wang, Chenhui Jiang, Linhai Ye, Feng Chen, Tao Zhou, Youcai Zhao,
Ultrasonic detection and deep learning for high-precision concrete strength prediction,
Journal of Building Engineering,
Volume 104,
2025,
112372,
ISSN 2352-7102,
https://doi.org/10.1016/j.jobe.2025.112372.
(https://www.sciencedirect.com/science/article/pii/S2352710225006096)
Abstract: Periodic testing of the compressive strength of in-service concrete structures is crucial for ensuring the safety and reliability of building infrastructures. However, current non-destructive detection methods encounter challenges in terms of accuracy, reliability, and applicability. This study proposes a completely new technological approach, incorporating ultrasonic detection and deep learning techniques. First, standard concrete specimens were prepared, and both compressive strength and ultrasonic tests were performed, leading to the creation of an ultrasonic database with 623 metadata entries. Secondly, a hybrid deep learning model was designed, trained, and validated to predict concrete compressive strength. Finally, the performance of the hybrid model was further enhanced through the integration of hyperparameters, as identified by SHapley Additive exPlanations analysis. The optimal model showed a substantial improvement in precision, reducing the mean absolute error to 2.02 MPa and the mean absolute percentage error to 7.84 %. This research showcases the potential and contributions of ultrasonic detection and deep learning in assessing the strength of in-service concrete structures.
Keywords: Ultrasonic signal; Deep learning; Machine learning; Concrete compressive strength prediction; Non-destructive testing; Shapley additive explanations (SHAP)