Home/Industry Projects

Smarter Simulations: New AI Model Advances Battery and Materials Research

Frontier AI, combined with materials science, has cracked the long-standing scientific challenge of simulating materials at quantum-level accuracy.

Challenge

From predicting how lithium batteries behave to modelling the corrosion of gas pipelines, simulating materials at quantum-level accuracy has long been a scientific challenge. We have developed DIEP, a new deep machine learning potential that outperforms existing models, offering faster, more accurate insights into fractures, battery interfaces, and other complex phenomena.

Impact

We have demonstrated how DIEP simulations could capture subtle details in the fracture of carbon nanotubes (tubes with diameters that are in the order of 1 billionth of a metre) and the diffusion of lithium between an anode and the electrolyte. We are now using DIEP to simulate more complex phenomena in much larger systems: the catalysis of chemical reaction on realistic surfaces in realistic environments, phase change materials, corrosion of pipeline surfaces, melting of high entropy alloys and other phenomena.

Changes in a material start at the level of the atoms. When you pull apart a piece of paper, the tension propagates through the paper until it reaches the weakest links, at which point tearing happens. The weakest links are where the attraction force between atoms is the weakest. Understanding the structure of the weakest links in the paper at the atomic level helps us understand tearing in detail. Likewise, the operation or lithium batteries, the corrosion of oil pipelines and the combustion of hydrocarbons are phenomena that involve the physical changes in materials that occur at the atomic level. To gain insight into these phenomena, one needs to simulate the material change at the level of atoms. However, there are usually millions of atoms involved in these phenomena, which makes it virtually impossible to simulate them, no matter how many CPUs or GPUs are used.

Solution

Machine learning potential models, such as DIEP, enable the simulation of complex phenomena that occur in materials with tens of thousands of atoms, such as our recent report on lithium battery interfaces (https://pubs.rsc.org/en/content/articlehtml/2025/ta/d4ta08189g). As part of the project of Ms Linh La, graduate researcher candidate at Deakin Applied AI, we are further improving DIEP to simulate hundreds of thousands of atoms, with the aim of reaching one million atoms. Given that DIEP machine learning potential models, including DIEP, exhibit quantum mechanical accuracy, our simulations should be able to enable reliable and robust material screening and discovery workflows.

Read more