Machine Learning-Based Optimization of Sustainable Mix Design for 3D-Printed Concrete Using Recycled Materials

3D printing of concrete is a potential way to reduce the environmental impact associated with construction and achieve rapid urbanization. There is potential to use waste-derived materials in 3D printed concrete, but there are significant gaps in predictive modeling of the mechanical properties. This study compiled a database of 477 experimental data points from 22 published papers, supplemented by 72 laboratory data sets. Fifteen input variables related to material composition and process parameters were used to train and evaluate three machine learning models. These were random forest, CatBoost, and TabPFN. The target variable was the compressive strength, and this was assessed in three loading directions (x, y, and z). All models were able to predict performance, with TabPFN consistently achieving the highest R2 values for molded concrete and 3D-printed concrete under the x loading direction. Curing age was the dominant predictor of compressive strength, followed by silica fume content, water-to-cementitious ratio, and cement content. Fly ash improves 3D printing because it enhances flowability and interlayer bonding. The clay and coarse aggregate content and pump speed had minimal impact. The optimized mix design incorporated 230 kg/m3 of recycled fine aggregate and this achieved compressive strengths of 25.5 MPa in the x loading direction. The work confirms the feasibility of using recycled materials in 3D printed ecoefficient construction, without compromising performance.