Co-driven physics and machine learning for intelligent control in high-precision 3D concrete printing

Abstract

With the increasing demand for precise control in 3D concrete printing, coordinating material rheological properties and printing parameters has become a critical challenge. This paper addresses how to intelligently optimize printing parameters to adapt to varying concrete material attributes and improve printing quality. A dual-path framework co-driven by physical information equations (PIE) and machine learning (ML) is proposed. PIE is embedded into convolutional neural networks (CNN) to enhance rheological properties prediction, while also coupled with the random forest (RF) model to predict printing parameters. Results show this approach efficiently matches yield stress (YS), plastic viscosity (PV), extrusion speed (ES), and printing speed (PS), significantly enhancing printing performance. This research provides construction engineers and 3D printing operators with a physics-guided, interpretable intelligent tool that reduces trial-and-error and improves construction reliability. The integration strategy also opens promising directions for future studies on large-scale printing involving multi-scale material-process-structure optimization and time-dependent rheological modeling.