Ultrasonic strength evaluation of underwater heterogeneous concrete using random forest model constrained by physical laws

Abstract

Compressive strength evaluation of concrete is crucial for the safety of underwater structures. However, the concrete heterogeneity impedes accurate evaluation based on empirical formulas (EF) derived from linear regression. This study proposes a four-phase model to formulate physical laws (PL). The concrete heterogeneity including sand-aggregate ratio (S/A), water-cement ratio (W/C), and diameter of average aggregate (Da), is considered along with the Rayleigh (R) and pressure (P) wave velocities. The proposed PLs are used to constrain the fitness functions of Particle Swarm (PS) optimization and Genetic Algorithms (GA), and the Random Forest (RF) model is enhanced to PL-PS-RF and PL-GA-RF models. Ultrasonic and compressive tests are performed on 96 specimens with 32 different mix parameters to train the models. The maximum error significantly decreases from 20 MPa to 5 MPa with the PL-PS-RF model. Parameter analysis reveals the mechanisms behind the improvements. The proposed methodology improves the evaluation accuracy and the testing is extended from an aerial to an underwater environment.