Rheological behavior of 3D printed concrete: Influential factors and printability prediction scheme

Abstract

The rheological properties of cementitious materials play a crucial role in determining the printability for extrusion-based 3D concrete printing. This study develops data-driven machine learning (ML) models to predict two key rheological parameters - plastic viscosity (PV) and yield stress (YS) of 3D printable cementitious composites based on the mixture composition and time after water addition. A systematic experimental study is conducted by varying the contents of cement, fly ash, silica fume, sulfoaluminate cement, superplasticizer, and water-to-binder ratio, and time after water addition. The measured rheological data is used to construct a database for training predictive models including linear regression, support vector regression, random forest, extreme gradient boosting, and multi-layer perceptron neural network. The extreme gradient boosting model achieves the highest prediction accuracy with low root mean square error and all coefficients of determination exceeding 0.9 for both plastic viscosity and yield stress. Importance analysis identifies the most influential parameters affecting the rheological properties. A printability classification scheme is proposed using the model predictions by defining a printable zone of PV and YS. The data-driven framework is validated to effectively predict printability of new mixtures without trial-and-error. This study demonstrates the potential of ML models to accelerate the design and optimization of 3D printable cementitious materials.