You are here : Home > News > Visualizing defects in crystal solids revolutionized by Artificial Intelligence!

News | Scientific result | Physics | Materials | Safety of nuclear facilities | Nuclear physics | Artificial intelligence

Visualizing defects in crystal solids revolutionized by Artificial Intelligence!

​A study published by researchers at the CEA challenges the ways in which defects in crystal structures are described, proposing a novel approach to characterizing and modeling crystal defects at the atomic scale based on machine learning techniques. The precise characterization of defects is essential to better understand changes in materials in hostile environments, for example how they are affected by irradiation in the core of a nuclear reactor. This novel approach goes further than conventional methods, opening up many new prospects in fields extending beyond materials science. The results of the study were published in Nature Communications on September 17, 2020.

Published on 30 September 2020

We use crystals all the time in our daily lives - most metals, for example, are crystalline. Known for the almost perfect organization of their atoms, crystals nonetheless always contain some imperfections, which we call defects. The concentration and morphology of defects in a crystalline solid have a direct influence on the properties of the material. Improving our understanding of crystal defects and their evolution will therefore make it easier to predict changes in how materials behave over time. Understanding such changes is especially crucial for ensuring the optimal design of facilities subject to severe environmental conditions, such as irradiation.


In modern materials science, we can simulate the onset and evolution of defects in crystalline solids using very large-scale computer simulations. However, the immense stream of data generated makes analyzing numerical simulation experiments an extremely complex process. Researchers at the CEA, the results of whose work have just been published in Nature Communications, on September 17, propose a novel approach, which can be applied universally, to overcome this difficulty. This new approach is in fact the first method that can be applied to all materials with a crystalline structure. Providing a continuous visualization of a defect and its atomic environment, this facilitates the description of complex physical processes such as the migration of defects under irradiation.


The researchers, from the Nuclear Energy Division and the Military Applications Division of the CEA, have drawn on Artificial Intelligence methods to develop an algorithm that describes distortions in the local atomic environment caused by defects in the material. This distortion score facilitates automatic defect localization and enables a 'stratified' description of defects that can be used to distinguish zones with different levels of distortion within the crystalline structure.


defauts cristallins ia.jpg

This periodic representation (i.e. a 'pattern' repeated to infinity) shows the level of distortion reached within the body-centered cubic structure of iron due to the insertion of screw dislocations. The atoms are colored according to the distortion score: the zones impacted by the elastic field are shown in darker colors. © CEA


The results of this study open up many exciting possibilities for future development across the entire materials science community.  These simulation tools can be used to automate analysis of huge datasets, such as those generated as a result of experimental techniques like atom probe tomography, transmission electron microscopy and synchrotron radiation, methods already being used to probe the mysteries of matter. These developments may also be applied in other fields, including Chemistry, Biology and Medicine, for example, to detect cellular defects characteristic of cancer.


Top page