PhyGAT: 5,000-Sample Multi-Physics Graph Dataset for 5-Component Hybrid Nanofluids

Dataset Overview This dataset contains 5,000 steady-state multi-physics fluid flow and thermal simulations of 5-component hybrid nanofluids ($\text{Al}_2\text{O}_3$, $\text{Cu}$, $\text{TiO}_2$, $\text{SiO}_2$, and MWCNT dispersed in water) through a 2D channel containing a cylindrical obstacle at $\text{Re} \approx 80$. Ground-truth solutions were generated using the FEniCS finite element solver. File Contents graph_nanofluid_dataset_5000.pt: A serialized PyTorch list containing 5,000 graph dictionary objects. Graph Structure per Sample x_coords $[N \times 2]$: Spatial coordinates $(x, y)$ of the unstructured finite element mesh nodes ($N \approx 1,000\text{--}1,200$). edge_index $[2 \times E]$: Directed connectivity tensor defining adjacent node pairs across triangulated cells. edge_attr $[E \times 3]$: Edge features $[\Delta x, \Delta y, \Vert{}\Delta \mathbf{x}\Vert{}]$. phi_vector $[5]$: Nanoparticle volume fraction vector $[\phi_1, \phi_2, \dots, \phi_5]$ sampled from a Dirichlet distribution ($\sum \phi_i \le 0.04$). node_flags $[N \times 4]$: One-hot boundary condition indicators (interior fluid, obstacle boundary, inlet, outlet). sdf $[N \times 1]$: Signed Distance Function to the nearest solid physical boundary. y_velocity $[N \times 2]$: Solved velocity vector components $[u, v]$ in m/s. y_temperature $[N \times 1]$: Solved non-dimensional temperature field $T \in [0, 1]$.