Machine learning-assisted intelligent identification of hydrogen trap information in temperature-programmed hydrogen desorption

Abstract

Green hydrogen is a clean energy with great development prospects. However, in hydrogen-rich environments, certain mechanical properties of metals can degrade significantly, potentially leading to serious structural failures in engineering applications. Among the many factors affecting the hydrogen embrittlement sensitivity of materials, the diffusion and capture behavior of hydrogen is particularly key. Thermal Desorption Spectroscopy (TDS) is one of the key technologies for studying these behaviors. However, identifying hydrogen trap information from TDS data remains a significant challenge. To address this issue, this study proposes a machine learning-based approach for the intelligent identification of hydrogen trap information, facilitating the rapid, accurate, and robust extraction of hydrogen trap density and binding energy from TDS, including both single-trap and distinct double-trap scenarios. In the proposed approach, an Oriani-based hydrogen transport equilibrium model was implemented in MATLAB to generate a large number of TDS data samples. A TDS fitting procedure based on the asymmetric double sigmoidal function is developed to represent the TDS data using five parameters. These five fitting parameters and the hydrogen trap characteristics are used as input features and output targets, respectively, to construct the machine learning models. Among four classical machine learning algorithms evaluated, the support vector machine exhibits the highest predictive accuracy. Furthermore, compared with widely used methods for hydrogen trap identification, the proposed approach demonstrates significant advantages in terms of accuracy, efficiency, and robustness. This research provides a powerful tool for gaining deeper insights into hydrogen behavior in metals.