AI model developed to extend life of EV batteries

It is not uncommon for batteries in electric cars to be the first component of the vehicle to age. This is a waste of resources and is holding back the transformation of the transport sector. To address this issue, the automotive industry is developing software, often based on AI, to optimise battery management and control. Researchers at Uppsala University have now produced an AI model that can reportedly increase the reliability of battery health predictions by up to 70%.

“Being able to learn more about the life and aging of batteries will benefit future control systems in electric vehicles. It also shows how important it is to understand what happens inside the batteries. If we stop looking at them as black boxes that are simply expected to provide power, and instead acquire a detailed picture of the processes, we can manage them so that they stay in good condition longer,” said Professor Daniel Brandell, who led the study and is in charge of the Ångström Advanced Battery Centre at Uppsala University.

Several years of battery testing are behind the study, carried out in collaboration with Aalborg University in Denmark. A database was built up by collecting data from numerous very short charging segments. This was then combined with a detailed model of all the different chemical processes taking place inside the battery.

“Altogether, this gives us a very precise picture of the various chemical reactions that result in the battery generating power, but also of how it ages during use,” said Wendi Guo, who conducted the study.

The discovery could also affect the safety of electric vehicles. The safety problems that can occur in the battery are often due to design flaws and side reactions, which can also be predicted by studying data from the battery’s charging and discharging.

“The fact that we only use short charging segments is probably an added advantage. Battery data from electric vehicles is sensitive, both for the industry and from an anonymisation point of view for users. This research shows how far you can get without needing complete datasets,” Brandell said.

Image credit: iStock.com/PrathanChorruangsak