#### Withdraw

Loading…

# Developing machine learning interatomic potentials to probe the plasma-materials interactions of tungsten-based high-entropy alloys as plasma-facing components

#### Abdelghany, Muhammad A.

This item's files can only be accessed by the Administrator group.

## Permalink

https://hdl.handle.net/2142/120583

## Description

- Title
- Developing machine learning interatomic potentials to probe the plasma-materials interactions of tungsten-based high-entropy alloys as plasma-facing components
- Author(s)
- Abdelghany, Muhammad A.
- Issue Date
- 2023-05-03
- Director of Research (if dissertation) or Advisor (if thesis)
- Allain, Jean Paul
- Committee Member(s)
- Stubbins, James
- Department of Study
- Nuclear, Plasma, & Rad Engr
- Discipline
- Nuclear, Plasma, Radiolgc Engr
- Degree Granting Institution
- University of Illinois at Urbana-Champaign
- Degree Name
- M.S.
- Degree Level
- Thesis
- Keyword(s)
- Plasma-materials Interactions
- Interatomic Potentials
- Tungsten
- High-entropy Alloys
- Plasma-facing Components
- Nuclear Fusion Materials
- Atomistic Simulation
- Molecular Dynamics
- Density Functional Theory
- Ab Initio Molecular Dynamics
- Surface Science
- Machine Learning
- Condensed Matter Physics

- Abstract
- This thesis aims to extend the computational tools available in the field of computational plasma-materials interactions (PMI) by introducing a systematic approach to develop machine learning interatomic potentials for complex alloy systems under ion irradiation. These potentials will enable large-scale molecular dynamics (MD) simulations to efficiently capture PMI in complex multicomponent alloy systems, such as high-entropy alloys (HEA), under extreme conditions found in fusion reactor environments. Specifically, complex alloy systems such as HEAs have a highly complicated PES due to both their topological and compositional disorders. This complexity makes the development of representative interatomic potentials using traditional methods an extremely difficult task. Incorporating machine learning into the development of interatomic potentials represents a new paradigm in materials science, which provides a promising way to overcome these problems by representing the many-body PES in terms of deep neural networks (DNN). This approach can achieve the accuracy of quantum mechanics calculations for large systems made up of millions of particles with orders of magnitude less computational cost than quantum mechanics calculations. This technique is a step towards solving the fundamental problem of small-size and short-time scales associated with accurate first principle calculations of far-from-equilibrium processes, such as radiation damage and plasma-materials interactions. This study shows that splitting the surface irradiation process into a set of smaller independent physical scenarios modeled by first principle calculations of a few femtoseconds ab initio molecular dynamics (AIMD) trajectories can efficiently simulate long-timescale and far-from-equi far-from-equilibrium librium ion irradiation processes. We adopted the Deep Potential Molecular Dynamics (DeePMD) technique to develop ML many-body interatomic potentials for pure tungsten and tungsten-based high-entropy alloys (W-HEA). The potentials are trained using data from ab initio calculations of atomic forces and energies collected from relatively short AIMD trajectories. The obtained DeePMD potentials had almost the same accuracy as the DFT reference data, reproducing the same values of the DFT interatomic forces and energies within a small error. Furthermore, they could accurately capture both the structural and thermodynamic properties of the systems, even at elevated temperatures, as evidenced by their excellent agreement with the results of AIMD simulations. Using MD calculations, the developed W-H DeePMD potential was then used to simulate H ions irradiation of pure tungsten. The results showed that the developed potential effectively captures the same shape of the surface potential energy map of the classical Tersoff potentials and the pair-potential component of the DFT, Tersoff, and ReaxFF potentials. Furthermore, the H irradiation response of W, as predicted by the developed DeePMD potential, exhibits excellent agreement with the response predicted by the W-H Tersoff potential. In addition, the depth profile of the implemented hydrogen concentration in tungsten using the developed DeePMD potential resembled the characteristic spatial implantation regions of hydrogen isotopes in metals, which was observed experimentally and correlated with previous calculations. We also developed a novel deep neural network potential for W-HEA, a candidate material for fusion applications. HEAs are promising materials for use in extreme nuclear environments due to their high structural stability and excellent mechanical properties at high temperatures. However, we have limited knowledge of their irradiation responses and the underlying mechanisms controlling such responses, including their surface-plasma performance, fuel retention and recycling, and surface compositional and morphological changes. This work contributes towards filling this knowledge gap by studying the PMI properties of irradiated W-HEA surfaces using the developed DeePMD potential. The potential was used to simulate H ions irradiation of an equiatomic W-HEA surface using MD. The results indicate that H ions penetrate less deeply in W-HEA compared to pure W. Furthermore, the reflection yield and energy-angular distribution of the reflected H ions were evaluated to understand the plasma edge performance of W-HEA. Additionally, in this study, we extended the investigation of H ions irradiation of W-HEA to higher incident energies using Binary-collision approximation (BCA) calculations. DYNAMIX code was used to simulate H ions bombarding the W-HEA surface at energies ranging from 500 eV to 20 KeV, at a fluence of 1x10^17 cm^-2. The results revealed that the maximum sputtering yield of W-HEA is at approximately 2 KeV, with three times greater sputtering than the experimental value for pure tungsten irradiated with H ions of the same energy and angle of incidence. We further analyzed the energy-angular distribution of the reflected H ions and the sputtered target atoms at the peak sputtering case. We also studied the target’s compositional change induced by H-ion irradiation for the peak sputtering case. The evolution of the surface composition showed a slower reduction rate of the fractional composition of W and Ta on the surface than that of V and Cr due to the preferential sputtering of the latter. These results aligned with the results of N2 ions irradiation experiments performed at the RSSEL’s IGNIS-I facility. This work encourages the use of this approach to investigate more properties of W-HEA under higher energies irradiation, as well as the effects of grain boundaries and other materials imperfections on the irradiation response. Overall, this study contributes to filling the knowledge gap on the plasma-material interaction properties of W-HEA and provides a promising approach for simulating these properties in large-scale molecular dynamics simulations, which could help to inform the design and development of materials for use in nuclear fusion reactors.
- Graduation Semester
- 2023-05
- Type of Resource
- Thesis
- Copyright and License Information
- Copyright 2023 Muhammad Abdelghany

## Owning Collections

##### Graduate Dissertations and Theses at Illinois PRIMARY

Graduate Theses and Dissertations at Illinois#### Manage Files

Loading…

#### Edit Collection Membership

Loading…

#### Edit Metadata

Loading…

#### Edit Properties

Loading…

#### Embargoes

Loading…