Evaluation of Steels Susceptibility to Hydrogen Embrittlement: A Thermal Desorption Spectroscopy-Based Approach Coupled with Artificial Neural Network

Abstract

A novel approach has been developed for quantitative evaluation of the susceptibility of steels and alloys to hydrogen embrittlement. The approach uses a combination of hydrogen thermal desorption spectroscopy (TDS) analysis with recent advances in machine learning technology to develop a regression artificial neural network (ANN) model predicting hydrogen-induced degradation of mechanical properties of steels. We describe the thermal desorption data processing, artificial neural network architecture development, and the learning process beneficial for the accuracy of the developed artificial neural network model. A data augmentation procedure was proposed to increase the diversity of the input data and improve the generalization of the model. The study of the relationship between thermal desorption spectroscopy data and the mechanical properties of steel evidences a strong correlation of their corresponding parameters. A prototype software application based on the developed model is introduced and is openly available. The developed prototype based on TDS analysis coupled with ANN is shown to be a valuable engineering tool for steel characterization and quantitative prediction of the degradation of steel properties caused by hydrogen.

Hydrogen Embrittlement as a Conspicuous Material Challenge─Comprehensive Review and Future Directions

“…To further correlate H transport in the microstructure to the susceptibility to HE, Malitckii et al , built one conceptual method that is able to link TDS with HE using ANN model. Worthy of note is that TDS is considered an effective way to assess the microstructure-dependent H trapping.…”

“…For better quantifying the susceptibility to HE, the well-defined H sensitivity parameter (HSP), calculated as HSP = (ε – ε H )/ε × 100% where ε and ε H are the failure strains without and with H, was utilized for training. Based on the approach raised by Malitckii et al, , two ANN models were trained and validated for a series of materials including austenitic, ferritic, and ferritic-–artensitic steels. Both trained ANN models exhibited good accuracy in which a correlation of more than 90% is achieved between experimentally measured HSP and model predicted values, suggesting that the ANN models fed by TDS data can be a robust tool for predicting HE in metallic systems.…”