Researchers from the University of Surrey (UK) have taken the first step toward creating materials for sodium-ion batteries with the help of artificial intelligence. They assembled a library of 28 different chemical compositions and used the resulting data to train a machine learning model - an algorithm that finds patterns in large arrays of information. The work was published in the Journal of Materials Chemistry A.
Sodium-ion batteries are considered a promising replacement for the familiar lithium-ion ones: sodium is far more abundant in nature and is more widely distributed than lithium. However, the main obstacle to their mass use is improving capacity (how much energy a battery can store) and stability over repeated charging and discharging.
In the new study, the scientists varied the chemical composition of sodium manganese oxide - a material that serves as the cathode (positive electrode). The cathode largely determines how much energy a battery can hold and how long it will last. In the 28 samples, part of the manganese was replaced with different amounts of iron, titanium and vanadium, and then the researchers compared how each variant performed in sodium-ion cells.
The results showed a statistical association: adding iron was associated with higher capacity, vanadium improved resistance to repeated charge-discharge cycles, and small amounts of titanium increased performance at high charge rates. Some compositions with titanium retained more than 90% of their capacity after 100 cycles. We emphasize: this is a correlation, not proof of a causal relationship - the observations alone do not guarantee that these additives themselves cause the improvements.
To make sense of the large volume of data, the team developed an automated approach to analysis. The resulting dataset was used to train a machine learning model. The library is currently too small for reliable predictions, but expanding it could in future allow AI to propose new compositions even before they are synthesized and tested in the laboratory.
"We are heavily dependent on lithium-ion batteries, but lithium reserves are relatively scarce and unevenly distributed around the world. There is far more sodium, so sodium-ion batteries are an important technology to develop," says Dr Daniel Commander, a lecturer in the School of Chemistry and Chemical Engineering and one of the authors of the work. "The task is to find materials with the right characteristics. Our work is the first step toward creating large datasets that will allow machine learning to accelerate this search."
Co-author of the study, postgraduate student Eden Marks, noted: "What impressed me most was how relatively small changes in composition shift the balance between capacity, high-rate performance and durability. A map of these relationships helps us understand where to go next."
In the future, expanding the library of compositions could help scientists select cathodes for specific tasks - whether high capacity, fast charging or long service life. For now, however, this concerns laboratory samples, not a finished commercial product.