Dataset with graph representations of FEM cure simulations for training Graph Neural Network surrogates

This dataset includes graph representations of cure simulations for training, validating and testing a Graph Neural Network (GNN) surrogate for cure modelling. The data consists of graph representations (in .pt format) of finite element meshes for rectangular and L-shaped domains, with each graph representing the state of the system every 100 seconds during the curing period. It encompasses thickness from 5-30 mm and a wide range of two-dwell cure cycles. The boundary conditions are representative of Vacuum Assisted Resin Transfer Moulding, with a prescribed temperature boundary on the mould side and a convective boundary condition on the vacuum bag side. The graphs are Torch Geometric Data objects, with their attributes describing the node features, edge features, connectivity, and labels. The labels are the rates of change of temperature and the degree of cure, respectively. The graph data was compiled by extracting results from finite element cure simulations in COMSOL and then postprocessing with a python script to create the graphs. The graphs are grouped by simulation name, which describes the domain shape (rectangular or L-shaped) and the simulation parameters: thickness, first dwell temperature, dwell duration, second dwell temeprature, and ramp rate. They can be accessed as key-value pairs in a Python dictionary, where the keys are the simulation names and the values are the graphs for that simulation.