Scientists combine artificial intelligence and atomic-scale images to produce better batteries

Rechargeable batteries are a great and magical invention, but they age, making them expensive to replace or recycle. To address this problem, William Chueh, an associate professor of materials science and engineering at Stanford University, has invented a first-of-its-kind analytical method that can produce better performing batteries, thereby helping to produce the ideal "indestructible" battery.

(Image source: Ella Maru Studio)

Chueh, lead author Dr. Haitao "Dean" Deng, and collaborators at Lawrence Berkeley National Laboratory, MIT, and other research institutions used artificial intelligence to analyze new atomic-scale microscopic images to understand exactly why batteries wear out. The researchers said discovering these causes could significantly extend battery life. Specifically, they studied a specific type of lithium-ion battery based on LFP material. This battery does not use chemicals with limited supply chains, so it will accelerate the pace of electric vehicles entering the mass market.

Nanocracks

"Imagine a battery as a ceramic coffee cup," Chueh explained. "When you heat or cool it, it expands and contracts, which causes defects in the ceramic. The same process happens to the material in a rechargeable battery every time you charge it, which then uses up the charge and causes failure."

Chueh noted that temperature is not what causes the cracks in the battery, but rather the mechanical strain that the materials put on each other during each charging cycle.

"Unfortunately, we know very little about what's happening at the nanoscale where atoms are bonded together," Chueh said. "But with new high-resolution microscopy techniques, we can see the changes, and artificial intelligence helps us understand them. This is the first time we can visualize and measure these forces at the single nanometer level."

"The properties of any given material are a function of its chemistry and the physical interactions within the material at the atomic level, which I call 'chemomechanics,'" Chueh said. "What's more, the smaller the object, and the more diverse the atoms that make up the material, the harder it is to predict how the material will behave."

Transformative tools

Using AI for image analysis is not new, but using the technology to study atomic interactions at the smallest scales is rare. In medicine, AI has become a transformative tool for analyzing images of everything from knee defects to cancer. Meanwhile, in materials science, new methods of high-resolution X-ray, electron, and neutron microscopy allow for direct visualization at the nanoscale.

More and more electric vehicle manufacturers and battery-intensive companies are adopting LFP materials. In this project, the research team chose to use LFP and lithium iron phosphate, materials commonly used in positive electrodes. The electrode does not contain cobalt and nickel, which are mostly used in commercial batteries. LFP batteries are also very safe, although they contain less energy per pound.

Although LFP has been studied for 20 years, only two key outstanding technical issues have been understood so far. The first is to understand the elasticity and deformation of the material during charging and discharging. The second is how the material expands and contracts in a specific mechanism in which LFP is partially stable, or "metastable".

Deng explained these two questions for the first time by using image learning techniques. He applied it to a series of two-dimensional images produced by a scanning transmission electron microscope, as well as advanced (spectral-panel imaging) X-ray images. Deng said these findings are important for the capacity, energy retention and rate of batteries. More importantly, Deng believes that this technology can be generalized to most crystalline materials, or can make good electrodes.

“Artificial intelligence can help us understand these physical relationships, which are critical for predicting new battery performance, reliability for practical applications, and degradation over time,” Deng said.

The researchers say they have used the technology to enable a variety of new battery designs at the atomic level. One is new battery-control software that manages charging and discharging in a way that prolongs the battery's life. Another is the development of more precise computational models that will allow battery engineers to explore alternative electrode materials on a computer rather than in a lab. "AI can help us look at old materials in new ways, potentially leading to better alternatives," Chueh said.