Strength Characterisation of Fly Ash Blended 3D Printed Concrete Enhanced with Explainable Machine Learning

Abstract

This study investigates the performance of 3D printed concrete incorporating fly ash as a partial cement replacement and develops a machine learning model to predict its mechanical properties. A total of 28 mixtures were prepared with varying fly ash contents (5–15%), water-to-binder ratios, and superplasticiser dosages. Of these, seven mixes met the requirements for printability in terms of flowability, extrudability, and buildability. Experimental tests were conducted to evaluate compressive strength, flexural strength, water absorption, and sorptivity. Results showed that mixes with 5% and 7.5% fly ash achieved improved strength and durability, whereas higher fly ash levels reduced early-age performance due to clinker dilution and slower pozzolanic activity. Microstructural analyses confirmed the presence of C–S–H, portlandite, and ettringite, with fly ash contributing to pore refinement and matrix densification. To enhance predictive capability, a TPE-optimised Extreme Gradient Boosting (TPE-XGB) model was developed using data obtained from laboratory testing. The model achieved excellent accuracy (R² > 0.997) in predicting compressive and flexural strength. A graphical user interface integrating SHAP visualisation was created to provide transparent predictions, supporting practical implementation. The findings highlight the potential of fly ash to improve the sustainability of 3D printed concrete at optimised dosages and demonstrate the value of interpretable machine learning tools in mix design optimisation.