Multi-fidelity DEM database of triaxial compression responses for irregular granular assemblies

This dataset contains a multi-fidelity discrete-element-method (DEM) database generated for training multi-fidelity network models. The data were obtained from conventional triaxial compression simulations of granular specimens composed of irregularly shaped particles. The input parameters include particle roundness, confining pressure, interparticle friction coefficient, and initial void ratio. The output labels describe the macro–meso mechanical responses under different parameter combinations, including deviatoric stress, volumetric strain, coordination number, and contact anisotropy coefficient. The dataset includes high-fidelity training data, high-fidelity validation/test data, low-fidelity data, and a unified model-ready CSV file. The response labels are interval-averaged quantities evaluated over the axial-strain interval of 20% to 30% to reduce local fluctuations and improve label stability.