Self-evolving experimental platform for 3D sand printing

Abstract 3D sand printing revolutionizes casting mold fabrication while confronting nonlinear high-dimensional optimization challenges stemming from variable combinations and objective conflicts. Here, we develop Ex3DSP, a self-evolving experimental platform combining robotic ‘can-do’ capabilities with artificial intelligence ‘can-think’ cognition to establish a ‘do-while-thinking’ self-evolving experimentation paradigm. Addressing a 3-objective, 3-variable high-dimensional optimization problem, our Ex3DSP discovers the Pareto front with a 1148-fold reduction in experimental workload. The platform provides two solution types (objective-optimal and balanced) with interpretability and reveals underlying mechanisms (variable-objective interaction and sand mold fracture behavior). Moreover, these solutions are applied to guide real-world casting parameter selection, yielding castings with up to 385% improved physicochemical and mechanical performances. This work demonstrates a self-evolving experimentation paradigm for 3D sand printing, offering both an intelligent additive manufacturing roadmap and pre-industrial screening potential.