Vladislav Korostelev1,Konstantin Klyukin1
Auburn University1
Vladislav Korostelev1,Konstantin Klyukin1
Auburn University1
The emergence of hydrogen as a promising clean energy source has underscored the critical need for the development of advanced alloys that can efficiently store this valuable resource. However, the presence of hydrogen also poses challenges, particularly in the form of hydrogen embrittlement, which can compromise the structural integrity of alloys. As a result, there is a pressing demand for the exploration and design of new alloys that not only enable efficient hydrogen storage but also exhibit enhanced resistance to the detrimental effects of hydrogen. In order to simultaneously address the challenges of efficient hydrogen storage and resistance to hydrogen embrittlement, we have embarked on the development of a comprehensive model with the capability to predict hydrogen solubility and diffusivity.<br/>The focus of our research lies in unraveling the atomistic factors that dictate hydrogen's interactions in high entropy alloys (HEAs) and investigating the underlying descriptors that influence its absorption and diffusion processes. Our ultimate objective is to construct an interpretable machine learning model that can accurately predict hydrogen absorption energy and diffusion rates, utilizing local physical descriptors associated with each interstitial site.<br/>To achieve this, we used Density Functional Theory (DFT) to calculate hydrogen absorption energies for over 1000 unique octahedral, tetrahedral and triangular interstitial sites within 12 HEAs. Subsequently, we performed calculations to determine the relevant local environment descriptors, which encompassed a comprehensive analysis of electronic structure features such as the d-band center. Additionally, we investigated structural descriptors, including interstitial pore volume, and examined the dynamical structure of the lattice atoms through an analysis of the phonon structure.<br/>To build an actual model, we used the sure-independence-screening-and-sparsifying-operator (SISSO) machine learning algorithm. This allowed us to develop physics-based model that accurately predict hydrogen absorption energy across various metallic systems. By identifying the key descriptors governing hydrogen absorption in HEAs, our models enable accelerated screening of potential compositions with optimal hydrogen solubility and diffusivity properties.<br/>During the presentation, we will discuss the key descriptors, which have a profound impact on hydrogen absorption and diffusion. Additionally, we will present the performance of our model on previously unseen HEAs and other metallic systems, including intermetallic compounds.