3D printing of machine learning-optimized continuous fiber-reinforced smart molds with a filament-withdrawal demolding strategy for extreme forming

The fabrication of complex hollow structures with variable cross-sections and curvatures remains challenging due to the limited deformability and poor demolding performance of conventional rigid molds. This study proposes a novel strategy that integrates 3D printing of continuous fiber-reinforced smart mold with a dimension-reduction wire-drawing demolding method. A photothermal dual-crosslinked polymer network was developed to enable reversible stiffness modulation, ensuring high rigidity during molding and flexibility during demolding. To optimize fabrication performance, a machine learning framework based on gradient boosting regression was employed to model and analyze the influence of key printing parameters on both mechanical strength and dimensional accuracy. Using this approach, smart mold with a bending strength of 620.71 MPa and a printing error of 2.46 % were successfully fabricated. The method was further validated through the forming and demolding of representative geometries, including dumbbell-shaped and variable-section components. Results confirm the feasibility and robustness of the approach under extreme forming conditions. This digitally driven, material–process–structure integrated solution offers broad application potential for the precision manufacturing of complex hollow composite structures in solid rocket motors, aerospace engineering, and architectural fabrication.