Data for predicting flexural toughness of steel fiber-reinforced concrete beams with limited data using a physics-informed neural network

This dataset supports the study “Predicting flexural toughness of steel fiber-reinforced concrete beams with limited data using a physics-informed neural network.” The archive contains material parameters, uniaxial tensile and compressive constitutive tables, load–deflection curves, and derived toughness indicators for 225 three-dimensional finite-element simulations of steel fiber-reinforced concrete beams under four-point bending. The indicators include peak flexural strength, peak deflection, equivalent flexural strengths at three prescribed deflection limits, and the corresponding flexural toughness ratios. The dataset also includes theoretical reference responses calculated using Timoshenko beam theory and nonlinear section analysis, fixed training–validation–test splits, model predictions, error statistics, and results for different training-set sizes, curve perturbations, loss weights, and material-parameter extrapolation scenarios. The experimental and finite-element curves used for the experimental comparison were obtained by image digitization and are distinguished from the exported curves of the 225-case numerical dataset. Digitized curves are not original instrument records. This release provides data only. Model source code, plotting scripts, trained model weights, original Abaqus ODB files, and experimental instrument records are not included. Please consult the accompanying README and data dictionary for file descriptions, units, and data provenance.