Simulation of hydrogen TDS peaks and ML-based prediction of desorption energies (SimuTDS) from experimental data

Abstract

Hydrogen is a promising energy carrier for achieving global decarbonization goals, yet its interaction with structural materials such as steels pose significant challenges for the safe application of hydrogen in steels used for storage and transportation systems due to hydrogen embrittlement. Thermal desorption spectroscopy (TDS) is widely recognized as an effective technique for characterizing hydrogen trapping behaviour in steels; however, its interpretation is complex and experimental implementation is costly and time-consuming. This thesis investigates hydrogen trapping behaviour in ferritic and martensitic steels containing titanium, vanadium, and niobium carbides through reviewing available experimental TDS data and computational modelling. Experimental data from literature were compiled to create a dataset including desorption energies, peak temperatures, and microstructural characteristics such as carbide size distribution. A MATLAB-based code was developed to simulate TDS spectra via convolution of each peak. Results indicate that incoherent carbides act as deep, irreversible traps with high activation energies, while coherent and semi-coherent precipitates provide shallower, reversible traps. Additionally, a machine learning approach was implemented to predict desorption energies from TDS profiles, demonstrating potential for reducing experimental effort. These findings contribute to improved interpretation of TDS spectra and provide guidelines for designing steels with enhanced resistance to hydrogen embrittlement.