license: mit
This work is under review. Please cite the published paper after it is accepted.
This repository contains original and scaled version of Deformed ABC.
We introduce Deformed ABC, a linear-elastic FEM dataset featuring diverse geometries from the ABC dataset, randomly selected materials, and boundary conditions. Geometries are represented as a mesh graph. FEM conditions and solutions are encoded as nodal features. Each node includes (1) three dimensional node coordinates, which represents position of the node, (2) Signed Distance Field (SDF), (3) normal vector, (4) Dirichlet boundary condition, where zero is assigned to fixed nodes and one to free nodes, (5) Nodal loads, represented as three-dimensional vectors, corresponding to the force components, (6) material data, which is Young's modulus and Poisson's ratio, and (7) displacement, which is the solution of FEM. FEM surrogate models predict nodal displacements from other node-wise input features. The resulting dataset is then preprocessed via filtering and scaling. Further details on the FEM formulation and data preprocessing are provided in the Supplementary Information, and Code snippets for data preprocessing are available in github.com/AiPEX-Lab/GDEA_and_Deformed_ABC.