Geometric deep learning for statics-aware 3D gridshells

In the context of architecture, gridshells are three-dimensional frame structures in which loads are entirely born by edges, or beams. Our contribution is to draw the way to a computational method that, given an input gridshell provided by a designer, slightly changes the input to ensure good static performance. The changing is induced by structure node repositioning. If the gridshell is represented as a surface mesh, the problem boils down to finding a proper vertex displacement. The vertex displacement should strike a happy medium between structure rigidity, with load deformation as low as possible, and structure resistance, preventing stress caused breaks. In this report, we inculde a solution to solve this mesh vertex displacement learning problem with a target goal of reducing a physically-based loss function, namely the mean strain energy of a gridshell, by means of a graph neural network. We adopt several geometric input features and discuss their effects on the results.