Abstract
Fabricating free-standing 3D surfaces using conventional 3D printing technology often requires much supporting material, which is later discarded and hence wasted. Moreover, using supporting material tends to rough the printed surfaces, and removing it can damage the printed parts. By contrast, 4D printing can create 3D surfaces without needing supporting material. By integrating active material into the 3D printing, 4D printing allows the printed structure to re-deform through external stimuli, such as heating. Because of this characteristic, 4D printing can achieve shape-morphing from the designed 2D grid into the target 3D gridshell. In this work, we utilize multi-material 4D printing technology to fabricate face-like masks. We first print a 2D grid consisting of rectangular arranged double-layered segments; each segment has four different material combinations, in which each layer is made of either shape memory polymer (SMP55) or PLA. Based on those different material combinations, 4D printing can generate corresponding deformation modes as SMP55 shrinks by heating to release the residual stress stored during 3D printing. Lastly, a highly complex human-face mask gridshell can be achieved by controlling the deformed grid’s global and local curvature through the material combination of each segment and their distribution. The inverse design from the target 3D gridshell to the initial 2D grid is challenging due to the high nonlinearity of the deformation mechanism. To solve this problem, we use a deep learning model: fully convolutional networks (FCNs) to automate the inverse design process. This study uses human-face and Noh (a traditional Japanese art) masks as examples. First, we divide the human-face mask 2D grids into several parameters to generate random 2D human-face mask grid designs, which can be converted to corresponding depth images as 60,000 datasets after simulating their heat deformation processes by finite element method (FEM). Next, the depth images are used for training the FCN model that executes image segmentation tasks for inverse design, and then we can use Noh mask depth images to verify the validity of trained FCN models. Finally, although the average similarity of human-face masks between using 4D printing and the target is 0.9, the trained FCN model does not perform well for Noh masks since human-face masks’ parameters cannot describe Noh masks adequately. In conclusion, our 4D printing technology successfully demonstrates the feasibility and potential to generate complex 3D curved structures.
