A Differentiable Material Point Method Solver for the Modeling, Simulation and Analysis of Extrusion-Flow Process for 3D Concrete Printing (2026-02)¶

Recent years have witnessed rapid development in 3D Concrete Printing (3DCP), which offers a highly automated construction process with significant reductions in material, labor, and time costs. However, deviations from the original design often occur as a result of the complex extrusion-flow behavior of fresh concrete and the associated layer deformations, which are primarily caused by insufficient understanding of key process parameters. In this study, a differentiable material point method (MPM) solver is developed for the modeling and analysis of 3DCP processes, enhanced with reverse-mode automatic differentiation techniques built upon the discrete adjoint method to enable end-to-end derivative computations. The extrusion-flow process of 3DCP is modeled as a weakly compressible Bingham fluid. For single-layer deposition under various printing conditions, the maximum deviations in predicted width and height remain within 20% of experimental measurements. Despite the increased deviations for multi-layer structures due to cumulative effects, the framework provides a robust foundation for simulating the 3DCP process. Beyond its predictive capabilities, the differentiable MPM framework exhibits high efficiency in sensitivity analysis. As demonstrated in a case study of sensitivity with respect to eight process parameters, the framework reduced the computational time to just 60% of that required by the central difference method. This efficient gradient computation provides quantitative insight into 3DCP mechanics and offers a pathway for optimizing process parameters to improve final product quality.