Physics-Informed Neural Network Modelling of Hydrogen Diffusion and Trapping in Microalloyed Steels: A Data-Driven Synthesis Across Multiple Alloy Systems

Hydrogen embrittlement is a critical degradation mechanism in microalloyed and pipeline steels used in hydrogen-economy infrastructure. We present a physics-informed neural network (PINN) framework that embeds Fick’s second law and the Arrhenius temperature dependence directly into the loss function, trained on 22 temperature-dependent data points spanning pure α-Fe and API X65 pipeline steels (modern and vintage microstructures). The PINN recovered the pure-iron activation energy (4.2 kJ mol−1 vs. literature 4.15 kJ mol−1, R2 = 1.00) and yielded Arrhenius activation energies of 28.5 and 45.2 kJ mol−1 for modern and vintage X65, respectively, indicating substantially stronger trapping in older microstructures. McNabb–Foster analysis of ten ternary Fe–Me–C,N alloys revealed flat-trap binding enthalpies of 19 ± 2 kJ mol−1 and deep-trap free energies of 57 ± 2 kJ mol−1, with effective diffusivities spanning three orders of magnitude governed primarily by flat-trap density. The framework provides a computationally efficient and physically consistent tool for hydrogen transport prediction, with a clear roadmap for multi-feature extension incorporating compositional and microstructural descriptors.