Non-destructive evaluation of bond strength between epoxy repair material and concrete under wet surface

The bond strength between epoxy repair materials and concrete is significantly influenced by surface moisture and roughness, leading to increased variability in bond strength measurements. Traditional pull-off tests, while commonly used, are destructive and inefficient. This study introduces a novel framework integrating three-dimensional morphology analysis, ultrasonic pulse velocity (UPV), and artificial neural networks (ANNs) to predict bond strength under wet conditions. Four concrete substrates (Surfaces I–IV) with increasing roughness were prepared, and 23 surface roughness parameters were obtained by three-dimensional scanning technology. The wave velocity and amplitude parameters of epoxy repair material-concrete were obtained through UPV method. The bonding strength was obtained by pull-off method. Based on Levenberg-Marquardt (LM), Broyden-Fletcher-Goldfarb-Shanno (BFGS) and Bayesian regularization (BR) ANN training algorithms, the nondestructive evaluation models of epoxy repair material-concrete bonding strength under wet surface considering roughness and acoustic parameters were constructed. The results revealed a 38.6% increase in bond strength from 2.16 MPa (Surface I) to 3.11 MPa (Surface IV), accompanied by 8.3% and 7.7% reductions in ultrasonic wave velocity (5.36 → 4.91 km/s) and amplitude (102.6 → 94.8 dB), respectively. Among 23 surface parameters, root mean square slope ( S dq ) and interface expansion area ratio ( S dr ) showed the strongest correlation with bond strength ( ρ =0.91 and ρ =0.90, Spearman rank). The BFGS algorithm achieved the highest prediction accuracy, with a correlation coefficient ( R ) of 0.89 and a mean absolute percentage error (MAPE) of 6.09%, while the BR model exhibited superior stability (MAPE = 6.08%, standard deviation = 0.12 MPa). This work provides the first systematic integration of 3D surface and acoustic properties for non-destructive bond strength evaluation in wet conditions, offering actionable solutions for optimizing epoxy repairs in moisture-prone environments. • 23 parameters of concrete surface roughness are got by three-dimensional scanner. • Ultrasonic velocity of epoxy-concrete negatively correlate with interface roughness. • Broyden-Fletcher-Goldfarb-Shanno algorithm predicts bond strength most accurately. • Bayesian regularization neural network provides stable bond strength predict results. • Chiseling concrete's surface boosts the wet bond strength of epoxy repair material.