Ultrasonic evaluation of compressive strength of underwater concrete using a support vector machine embedded with physical laws

Accurate evaluation of the compressive strength of concrete is crucial for ensuring the safety of infrastructure. However, the inherent heterogeneity of concrete restricts the predictive accuracy of the empirical formula (EF) derived from linear regression. The support vector machine (SVM) offers robust capabilities for nonlinear regression involving multiple factors associated with concrete heterogeneity. In this study, parameters such as the sand-aggregate ratio, water-cement ratio, aggregate diameter, pressure wave velocity, Rayleigh wave velocity, and compressive strength are integrated into the SVM framework. For the first time, underwater concrete specimens are examined to facilitate the evaluation of underwater infrastructure. Compared with EF, which exhibits a maximum error of 20 MPa, the SVM reduces this error to 7 MPa. Furthermore, through the integration of intelligent algorithms and the embedding of physical laws, the optimised SVM further reduces the maximum error to 3 MPa. A parameter analysis is conducted to clarify the mechanisms underlying these improvements. The proposed method improves the accuracy of evaluations, extends applicability from aerial to underwater environments, and fundamentally enhances the SVM framework by embedding physical laws.