Lab study maps 28 sodium-ion cathode recipes to train AI

Twenty-eight. That is how many cathode recipes a team at the University of Surrey created to find out which mix stores the most energy in a sodium-ion battery, and which one lasts longest.

The cathode is the battery component that helps determine how much energy the battery can store and how well it performs over time. The researchers, led by Dr. Daniel Commandeur, lecturer at the School of Chemistry and Chemical Engineering, started from sodium manganese oxide. They replaced part of the manganese with iron, titanium and vanadium, in varying amounts, then tested each composition in sodium-ion batteries and compared the results.

The pattern is useful because each element pushes in a different direction. Iron was linked to higher energy-storage capacity. Vanadium improved stability over repeated charging and discharging. Small amounts of titanium improved performance at higher rates, and some titanium-containing compositions retained more than 90% of their capacity after 100 cycles. Eden Marks, PhD student and co-lead author, said that relatively small changes in composition could shift the balance between energy stored, charging speed and lifespan.

The team also built an automated method to analyse the large volume of experimental data, and used it to train a machine-learning model. The authors are clear about its limits: the current library is too small for reliable predictions. Expanding it to a much wider range of cathode materials is what could eventually let AI identify promising compositions before they are made and tested in the lab.

For battery researchers, the gain would be a shorter search: instead of testing recipes one after another, they could ask a model which mixes deserve attention first. Commandeur's argument rests on supply as much as performance, since lithium is scarce and unevenly distributed while sodium is far more abundant. But the study gives no date and no industrial partner, and these results come from laboratory tests that have not yet been reproduced at production scale.