GeoPINONet datasets, trained models, and source code for geometry-conditioned 3D elasticity

This archive contains the complete finite-element datasets, CAD/STL geometry files, trained model weights, source-code snapshot, and result files associated with the manuscript "GeoPINONet: Sparse-Supervised Physics-Informed Operator Learning for Full-Field Prediction in Variable 3D Elastic Domains". This version provides the corrected validation FEM CSV archives. Validation files contain complete nodal fields with coordinates, displacement components, and Cauchy stress components: x, y, z, ux, uy, uz, sxx, syy, szz, sxy, syz, sxz. The archive includes two geometry families: a 3D lug family and a plate-with-hole family. For each geometry, STL files defining the body, loaded boundary, and fixed boundary are provided together with ANSYS FEM CSV solutions for axial compression and lateral bending. Latin Hypercube Sampling parameter files for nested training subsets and validation splits are included. The archive also contains the final trained GeoPINONet model weights and result files used to reproduce the final-model, scaling-study, and ablation-study tables. The accompanying source-code repository is available at:https://github.com/v-urrutiav/GeoPINONet Version 3 note: This Zenodo version imports without modification all files published in version 2 and adds GeoPINONet_interactive_demo_weights_v1.zip, the dedicated checkpoint package for the reproducible interactive demo available in the GitHub repository. The v1 suffix identifies the first release of this standalone demo package and does not refer to the Zenodo record version. The imported dataset archives remain the corrected version-2 release; some legacy archive root directories retain their original internal v1 names.