A Geometry-Preserving Shape Optimization Tool Based on Deep Learning

In free-form architecture, computational design tools have made it easy to create geometric models. However, obtaining good structural performance is difficult and requires further steps, such as shape optimization, to enhance system efficiency and material savings. This paper provides a user interface for form-finding and shape optimization of triangular grid shells. Users can minimize structural compliance, while ensuring small changes in their original design. A graph neural network learns to update the nodal coordinates of the grid shell to reduce a loss function based on strain energy. The interface can manage complex shapes and irregular tessellations. A variety of examples prove the effectiveness of the tool.