Determining reinforcement bond strength in 3D printed concrete using machine learning with bayesian optimisation and numerical modelling

Abstract The reliable prediction of reinforcement bond strength in three-dimensional concrete printing (3DCP) remains a critical challenge due to material anisotropy, geometric variability, and complex interfacial mechanisms governing stress transfer. Conventional experimental approaches are resource intensive, while existing predictive studies remain limited in optimisation rigour, interpretability, and mechanical validation. To address these limitations, this study presents an interpretable machine learning framework using Extreme Gradient Boosting (XGB) algorithm optimised using Bayesian Tree Parzen Estimator and Taguchi optimisation techniques. A comprehensive database comprising 607 experimental pull-out test results retrieved from published literature was employed for model training and validation. K-fold cross-validation confirmed stable generalisation performance with TPE-XGB achieving a testing MAE of 1.46 MPa and RMSE of 1.7 MPa, marginally outperforming the TO-XGB model. Also, mechanical validation on unseen data configurations using finite element modelling demonstrated superior accuracy of trained XGB models compared to nonlinear finite element simulations. The finite element analysis reproduced rib induced stress transfer mechanisms with peak concrete stresses reaching 38.58 MPa at pull-out, supporting mechanical consistency. Moreover, interpretable machine learning analyses identified reinforcement diameter and embedment length as the primary governing parameters. The integrated framework provides a computationally efficient and mechanically consistent tool for bond strength prediction in reinforced 3DCP to support practical implementation.