Machine learning method to determine concentrations of structural defects in irradiated materials

Landon Johnson, Walter Malone, Jason Rizk, Renai Chen, Tammie Gibson, Michael W. D. Cooper, Galen T. Craven · 2025

Read Machine learning method to determine concentrations of structural defects in irradiated materials by Landon Johnson, Walter Malone, Jason Rizk, Renai Chen, Tammie Gibson, Michael W. D. Cooper, Galen T. Craven (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-17634.txt.

Introduction

Landon Johnson, with Walter Malone and five co-authors, speeds up defect forecasts for irradiated materials in a 2025 arXiv paper. Radiation creates structural defects, which are flaws in a material's crystal pattern, that grow and change how it performs. A standard method called cluster dynamics tracks each defect type with rate equations. That method grows costly when many temperatures and pressures must be explored.

The team trained a machine learning model to predict defect amounts directly. Their model skips the slow step by step tracking yet stays highly accurate. They demonstrate it on uranium nitride, a nuclear fuel material. The model captures complex links among material traits, temperature, radiation level, and defect amounts. It should transfer to other materials too.

Catalog record

CREATORLandon Johnson, Walter Malone, Jason Rizk, Renai Chen, Tammie Gibson, Michael W. D. Cooper, Galen T. Craven
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-17634

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Machine learning method to determine concentrations of structural defects in irradiated materials Landon Johnson, Walter Malone, Jason Rizk, Renai Chen, Tammie Gibson, Michael W. D. Cooper, Galen T. Craven Abstract The formation and subsequent growth of structural defects in an irradiated material can strongly influence the material’s performance in technological and industrial applications. Predicting how the growth of defects affects material performance is therefore a pressing problem in materials science. One common computational approach that is used to examine defect growth is cluster dynamics, a method which employs a system of mean-field rate equations to track the time evolution of concentrations of individual defect types. However, the computational complexity of performing cluster dynamics can limit its practical implementation, specifically in the context of exploring a broad set of physical conditions corresponding to, for example, different temperatures and pressures. Here, we present a machine learning approach to circumvent the computational challenges of performing cluster dynamics while maintaining high accuracy in the prediction of defect concentrations. Our method is illustrated on the nuclear material uranium nitride but is broadly applicable to other materials. The developed data-driven method is shown to accurately capture complex correlations between material properties, temperature, irradiation conditions, and the concentration of defects. 1 Introduction Structural defects alter the performance of materials in irradiated environments [ 1 ] . Predicting how defects form and then grow under a specific set of physical conditions corresponding to, for example, a specific temperature, pressure, and/or fission rate, is a problem that has generated significant interest due to its relevance in broad class of technological and industrial applications [ 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 ] . Defects types such as interstitials, bubbles, voids, anti-sites, among others, can form due to irradiation and alter performance properties [ 17 , 18 , 19 , 20 , 21 , 22 , 23 ] including thermal conductivity, electrical conductivity, and mechanical response. Performance alterations in irradiated and nuclear materials due to aging or irradiation-enhanced defect formation [ 24 ] can significantly reduce the viability of using a specific material due to safety and failure concerns that push an engineered system outside of the safe operating envelope [ 25 ] . Therefore, developing theoretical and computational tools to predict how defects form, the rate at which those defects grow, and how those defects affect the pertinent properties of a material under specific physical conditions are important questions with direct ramifications for technology advancement. Predicting defect growth is particularly important in multiscale modeling applications that use atomistic and/or microscale data to inform macroscale and engineering-level models. Predicting defect growth rates using computational methods is often a process that involves the implementation of complex computational procedures. For example, defect growth is often predicted by solving large systems of nonlinear differential equations [ 26 , 23 ] or performing kinetic Monte Carlo simulations [ 27 , 28 , 29 ] . Previous theoretical work on defect evolution of irradiated materials has generally focused on the development of systems of master equations that describe how clusters of point defects evolve in time, i.e., cluster dynamics simulations, [ 18 , 17 , 21 , 30 , 21 , 31 , 32 ] or on the application of lattice kinetic Monte Carlo methods [ 27 , 28 , 26 , 29 ] . The specific defect type, or combination of defect types, that give rise to structural changes generated by irradiation is typically specific to the material. In general, the propensity of each defect type to cluster with itself and with other defect types is material dependent [ 33 , 30 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 31 , 41 , 10 , 11 ] . This poses a problem from a modeling perspective because there is often limited transferability of a developed defect growth model from one material to another. One of the primary methods that is used to predict defect evolution in materials is cluster dynamics [ 17 , 18 , 19 , 26 , 20 , 21 , 22 , 23 , 42 , 43 , 44 ] . Cluster dynamics models can be used to predict the outcome of complex physical process that determine the time evolution of defects such as aggregation, recombination, and fragmentation. In a cluster dynamics simulation, a mean-field method is used to track the time evolution of concentrations or densities of point defects and defect clusters [ 23 , 10 , 11 ] . The mean-field nature of the method arises because the trajectories of each individual defect are not followed, only the evolution of the defect concentrations are followed. While implementing cluster dynamics is less computationally taxing than other approaches, for example kinetic Monte Carlo methods, it can still be computationally prohibitive to perform simulations over a broad set of physical conditions which may be needed to accurately inform macroscopic models in multiscale and multiphysics frameworks, for example, when a cluster dynamics code is used to compute a transport property that is then fed into a macroscale model. The speed of performing a multiscale simulation is therefore commonly limited by the model evaluation speed of the atomistic models at lower length scales in the hierarchically multiscale code structure. This problem is further pronounced in nonequilibrium atomistic environments, such as a material under irradiation, when the assumptions used to derive equilibrium kinetic theories breakdown [ 44 , 45 , 46 , 47 , 48 , 49 ] . One method that can be applied to microscale material models to increase evaluation speed and improve accuracy is to use data-driven approaches, such as machine learning (ML), to predict defect concentrations and reaction rates. There have been broad applications of ML methods [ 50 , 51 , 52 ] in physicals sciences and engineering [ 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 , 61 , 62 , 63 , 64 , 65 , 66 , 67 ] including in nuclear materials [ 68 , 69 , 70 , 71 , 72 , 73 ] . Some of the potential advantages of using data-driven methods are improved accuracy, fast model evaluation, and the ability to capture complex correlations in datasets. Some common disadvantages are that data-driven models typically have limited interpretability, limited transferability, and often generate unphysical predictions when extrapolation outside of the training data is performed. In this work, we utilize machine learning methods to determine concentrations of point defects and defect clusters in an irradiated material. Specifically, we present a data-driven approach that uses neural network models trained on limited cluster dynamics simulation data to predict the steady-state concentrations of defects in irradiated materials under a broad set of physical conditions. We illustrate the method on uranium nitride, a potential nuclear fuel. Our method consists of training a collection of deep neural networks, one network for each defect type in the material model, on data generated using cluster dynamics over a multidimensional space of state configurations. The concentrations can then be used to predict macroscopic quantities such

Read the complete text