Data-driven framework for printability and geometric quality prediction in 3D concrete printing

Three-dimensional concrete 3D printing (3DCP) faces persistent challenges in achieving consistent geometric quality and reproducible printability across varying process conditions, limiting its large-scale industrial adoption. This study presents a data-driven framework that integrates experimental characterization with machine learning-based prediction to evaluate and optimize geometric quality in 3DCP. Functional geometries (cubes, overhangs, and bridges) were fabricated using a robotic printing system at controlled nozzle speeds (75–150 mm/s) and flow rates (478–593 cm³/s), resulting in 46 cubes, 21 overhangs, and 66 bridges. High-resolution imaging enabled quantitative extraction of geometric indicators, including layer height variation, angle deviation, and bridge span stability, which were consolidated into a weighted geometric quality metric. Two predictive models were developed: the first estimated geometric deviations from process parameters, while the second inversely predicted optimal process parameters for a desired material response. Among several algorithms, CatBoost and DecisionTree regressors exhibited the strongest performance, with the best model achieving an R² of 0.74 and a mean absolute error of 1.5 mm. The derived printability map identified optimal operational regions (100–115 mm/s, 470–490 cm³/s) corresponding to stable, high-quality prints. This integrated experimental–computational approach establishes a quantitative foundation for real-time process optimization, adaptive control, and quality assurance in additive construction.