Engineering-informed machine learning for bond-strength prediction, mechanism discovery, and reliability assessment of FRP-reinforced seawater sea-sand concrete

Abstract The bond behaviour between fiber-reinforced polymer (FRP) bars and seawater sea-sand concrete (SWSSC) governs load transfer, anchorage performance, crack development, and the structural reliability of marine concrete members. Although recent machine-learning (ML) models have demonstrated high predictive accuracy, they provide limited insight into the governing bond-transfer mechanisms and rarely quantify the prediction uncertainty required for reliability-informed structural design. This study develops an uncertainty-aware and explainable ML framework to predict the bond strength of FRP–SWSSC systems while identifying the engineering variables and behavioural mechanisms governing bond transfer. A database comprising 286 experimental bond tests collected from 17 independent studies was analysed using five ML algorithms within a leakage-free validation framework. The optimized Random Forest model achieved the highest predictive accuracy on the independent testing dataset (R 2 = 0.820, RMSE = 2.52 MPa, and MAE = 1.77 MPa). Engineering interpretation revealed that the bonded-length-to-diameter ratio, FRP elastic modulus, concrete compressive strength, and rib-height-to-diameter ratio are the primary parameters governing bond performance. The analyses further demonstrated that bond behaviour is controlled predominantly by nonlinear threshold responses and progressive mechanical interlocking rather than simple linear additive effects, highlighting the coupled influence of anchorage geometry, reinforcement stiffness, and concrete confinement on load-transfer efficiency. Bootstrap resampling confirmed stable predictive performance (R 2 = 0.777 ± 0.038), while distribution-free split-conformal prediction achieved 91.23% empirical coverage for a nominal 95% confidence level, providing calibrated uncertainty estimates for unseen engineering configurations. The proposed framework bridges data-driven prediction and structural mechanics by integrating accurate bond-strength prediction, physically interpretable mechanism discovery, and quantified predictive reliability, providing a practical basis for reliability-informed assessment and design of FRP-reinforced seawater sea-sand concrete structures.