Experimental Investigation and Physics-Informed Neural Network Modeling of Hydrogen Embrittlement in Annealed 0.2 wt.% Carbon Steel

This integrated experimental–computational framework provides both empirical evidence and mechanistic insights into hydrogen transport in steels, establishing a robust pathway for the predictive modelling of embrittlement in hydrogen-based energy systems.

When steels are exposed to hydrogen, a crucial degrading process known as hydrogen embrittlement (HE) causes early brittle failure. In this investigation, baseline, pre-strained, and hydrogen-charged conditions were used to assess the mechanical reaction of annealed 0.2 weight percent carbon steel. Significant ductility loss was found in tensile testing; total elongation dropped from 34.6% in the baseline condition to 12.0% following hydrogen charge and then to 11.3% when pre-strain and hydrogen exposure were combined. In embrittled samples, the reduction in area decreased by over 80%, indicating a significant loss of toughness. Fractography showed a distinct change from brittle, flat fracture in hydrogen-charged conditions to ductile cup-and-cone fracture in uncharged specimens.

A strain-controlled diffusion model was created using Physics-Informed Neural Networks (PINN) and compared to a traditional Forward-Time Centered-Space (FTCS) solver in order to supplement the experimental results. The strain-dependent diffusivity included in the governing equation creates a feedback loop in which the concentration of hydrogen modifies the elastic modulus, strain, and diffusivity. The results show that, in comparison to Fickian diffusion, strain localisation speeds up hydrogen infiltration, which explains the experimentally observed embrittlement in pre-strained specimens.
A strong foundation for the predictive modelling of embrittlement in hydrogen-based energy systems is established by this combined experimental–computational approach, which offers empirical data and mechanistic insights into hydrogen transport in steels.