Integrating Machine Learning with 3D Printing Concrete for Marine Structures: Mix Design, Strength Assessment, and Carbon Footprint Prediction

Abstract

A well-designed mix ratio for 3D printing concrete (3DPC) is essential for successful application of this technology in marine structures. Adjusting the material mixes through physical experiments poses challenges for sustainability regarding waste production, energy consumption and greenhouse gas emissions. To develop better 3DPC mixes, this study employs the XGBoost machine learning algorithm to construct the predicted model. A dataset of 126 unique mix design datasets was collected from the literature and utilized in model training and development to forecast the 3DPC strength and carbon footprint. To demonstrate the effectiveness of the predicted model, visualization and assessment metrics such as scatter plots, Shapley Additive Explanations (SHAP), R-squared values, and mean absolute errors were reported. Based on the outcome of this study, the XGBoost model displayed accuracy with a high coefficient of determination. It was found that the cement content, printing speed, and admixtures were critical components that directly impact the 3DPC mixes. Future studies can consider other advanced ML models, hyperparameter adjustments, and a larger dataset to achieve better forecasts and interpretability.