Session: 09-01: Computational Methods in Micro/ Nanoscale Transport

Paper Number: 140668

140668 - Simulation of Nitrogen Atom Diffusion Using Machine-Learning-Based Interatomic Potential for Nitriding Model of Iron 

Abstract: 

Ammonia has been widely utilized in various industrial fields and attracted attention as a promising candidate for an energy carrier and green fuel. The emerging ammonia-fueled industrial furnaces, gas turbines, and pipelines made of iron-based materials are facing challenges due to undesired nitriding of iron. Our previous experiments [1] demonstrated that ammonia decomposition on the austeniticstainless steel wall from 700K to 873K causes nitriding, and the iron nitride in turn facilitates ammonia decomposition. Since excessive nitriding embrittles the material and causes mechanical failure of the reactors [2], the NH₃ decomposition on the surface and subsequent N-atom reaction and diffusion into the wall need to be understood. In this research, we propose an approach based on first-principle, non-phenomenological diffusion dynamics for predicting nitrogen depth profile in gas nitriding of iron. It provides first-principle understanding of the nitrogen diffusion process in iron as a function of temperature and nitrogen concentration.

In this work, a machine learning (ML) interatomic potential of Fe-N solid system is developed using DeePMD-kit [3] by training from a series of short ab-initio molecular dynamics (AIMD) and Meta-dynamics accelerated first-principle calculations up to 1000K. Density functional theory (DFT) calculations with climbing image-nudged elastic band (CI-NEB) are used to validate the trained ML potentials in terms of system energy and interatomic forces in the process of nitrogen reaction and diffusion in the Fe-N system. Then, the ML potentials are used to drive molecular dynamics (MD) simulations with more than 10 thousand atoms and nanoseconds timescale to obtain the reaction-diffusion coefficient of nitrogen atoms with concentration dependence in α, γ(γ’) and ε phases of Fe-N system up to 1000K. The temperature dependence of reaction-diffusion coefficients is resolved by using the Arrhenius form. To compare the calculated diffusion coefficient with our nitriding experimental data, a continuum mechanics model [3] is used and numerically solved to address the significance of concentration on the diffusion coefficient of nitrogen atom in iron. The ab-initio informed continuum simulations are compared against experiments in which NH₃/N₂ mixture is impinged on isothermal iron plate, the depth profile of nitrogen concentration in pure iron is measured through Wavelength-Dispersive X-ray Spectroscopy.

As a result, the trained ML potential is shown to outperforms the MEAM potential previously developed for the Fe-N system. It predicts the interatomic force and system energy more accurately than the MEAM potential in comparison with that from DFT. CI-NEB method depicts similar energetics in comparison with the Meta-dynamics accelerated first-principle calculation. This indicates that Meta-dynamics is an efficient approach with less human interference for sampling the potential energy surface of an interstitial solid diffusing system. Nitrogen reaction-diffusion in Fe-N system shows that when the nitriding degree (concentration of nitrogen in iron) affects the diffusion coefficients in α, γ(γ’) and ε phases, showing the same trend with the experimental data. Finally, the continuum mechanics model predicts the nitrogen depth profile and phase boundary, which is comparable to the nitriding experiment using pure iron.

This research demonstrates the first non-phenomenological, first-principle-based model to predict the nitrogen reaction-diffusion depth profile and phase boundary during gas nitriding. It improves our fundamental understanding of the nitriding phase boundary growth and solvent concentration profile with time at a given temperature.

Reference:

1.   P. Feng, M. Lee, D. Wang, Y. Suzuki, Int. J. Hydrog. Energ., Vol. 48, Issue 75, (2023), pp. 29209-29219.

2.   D. Wang, M. Lee, Y. Suzuki, J. Ammonia Energy, Vol. 1, (2023), pp. 1-10.

3.   H. Wang, L. Zhang, J. Han, W. E, Comput. Phys. Commun. Vol. 228 (2018) pp. 178–184 

4.    J. Wilhelm, Diffusion in Solids, Liquids, Gases (1952) pp. 71-77

Presenting Author: Peijie Feng The University of Tokyo

Presenting Author Biography: PhD second year student in The University of Tokyo, Department of Mechanical Engineering,
Micro Energy System Laboratory (MESL)
Current interest: multi-scale modeling, first-principle calculation, catalytics, ammonia

Authors:

Peijie Feng The University of Tokyo
Aditya Lele Princeton University
Minhyeok Lee The University of Tokyo
Yiguang Ju Princeton University
Yuji Suzuki The University of Tokyo