Answers are grounded in this paper and cite their sources.
Predicting the compressive response of concrete at high strain rates is central to protective design, yet common empirical relations are calibrated over narrow windows and behave unphysically outside them. Revisiting a split Hopkinson pressure bar programme on 40 and 63 MPa concrete (96 solid and annular specimens, 32–299 s−¹), this study develops a physics-informed framework: a monotone stretch-exponential dynamic increase factor (DIF) replacing the rational form, a Hill peak-strain function, and bounded pore-water and Pochhammer–Chree corrections. A physics-informed neural network (PINN) trained under five constraints is compared with six regressors. Across ten splits the closed-form models remain most accurate (R² = 0.82–0.84), with the PINN and Bernstein regression close behind; on this small, one-dimensional dataset added flexibility does not pay. The PINN alone extrapolates admissibly – monotone and bounded to 4000 s−¹ – though this is constraint-driven, not experimentally validated. SHAP attributes 79% of influence to logarithmic strain rate.
No takes yet. Share an insight, caveat, or question.
Ibrahim et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: