An Interpretable Machine Learning Approach for Predicting Reinforcement Bond Performance in 3D Concrete Printing (2025-09)¶

3D concrete printing (3DCP) is revolutionizing the construction industry by offering automation, material efficiency, and design flexibility. However, reinforcement integration remains a critical challenge due to the quasi-brittle nature of printable cement-based materials and the complex bond between reinforcement and the 3D-printed matrix. This study develops an interpretable machine learning (ML) approach to predict bond performance and failure modes of reinforcements in 3DCP structures. A dataset comprising 550 pull-out test results from various studies was used to train decision trees (DT), and categorical boosting (CatBoost), and adaptive boosting (AdaBoost) models. The models were evaluated based on classification accuracy and regression performance using multiple metrics. Results show that boosting-based models (CatBoost and AdaBoost) significantly outperform DT, achieving an R2 of 0.985 in bond strength prediction and 98.1% accuracy in failure mode classification. SHAP (SHapley Additive Explanations) analysis was applied to enhance interpretability, revealing that bond index (fR), relative concrete cover (cs/ds), and embedded length (lb/ds) were the most influential parameters. A comparison with the fib Model Code 2010 demonstrated the superiority of ML models, with the CatBoost model achieving far lower prediction errors compared to the empirical approach. These findings highlight the potential of ML-driven approaches in optimizing reinforcement design and advancing the reliability of 3D-printed concrete structures.