Accelerated Discovery of 3D Printing Calcium Sulphoaluminate Cement Composites Using Data-Driven Multi-Objective Optimization.

Additive manufacturing enables the fabrication of complex geometries and structures that are difficult to attain by conventional methods. However, many printable materials exhibit inherent trade-offs among their performance properties. Traditional material design, often relying on intuition-driven and inefficient trial-and-error approaches, frequently fails to identify optimal formulations. In this study, we propose a machine learning-assisted framework to efficiently refine the composition of 3D-printable calcium sulphoaluminate (CSA) cement composites with a balanced trade-off among thixotropy, mechanical strength, and shape stability. Our approach integrates a multi-objective optimization algorithm with a data-driven surrogate model to intelligently propose new mix proportions, thereby reducing the number of required experiments. Starting from seven primary formulations and 28 initial experimental samples, the method identified 23 improved mix proportions after only 20 algorithm iterations. The workflow demonstrates how optimization-assisted formulation refinement can accelerate the search for better-performing materials and is potentially adaptable to other material design challenges.