EV battery companies use AI to identify defects | Automotive News

Technology companies are rushing to identify defects in electric vehicle battery cells that can cause fires and other problems. Artificial intelligence tools are helping them do it.

They are training AI models to quickly assess what's normal in a cell and what isn't. Automated tools significantly speed up quality checks.

"AI is all about that scalability: The more you can deploy it, the better off you are," Peter Kostka, director of battery solutions at PDF Solutions, said Oct. 10 at The Battery Show in Detroit.

PDF Solutions teaches its AI model to understand battery structures.

During the next two weeks, Automotive News is exploring the many ways that artificial intelligence is infiltrating and influencing the automotive industry.

"If our model learns, 'Hey, these are the kinds of things that I should generally see' ... then that same model can be applied across different manufacturing lines, and we get that scalability," he said.

UnitX also uses AI to improve the defect identification process. Its 3D technology can identify subtle abnormalities at high speeds. It also detects more depth than 2D vision, CEO Keven Wang said at The Battery Show.

Human operators scanned one cell every five minutes in a UnitX case study, while the AI tool scanned one every 3.5 seconds. Based on the case study, the factory could reassign three human inspectors by using the AI tool, Wang said.

"It needs to see the defects it has seen before, but you'll be surprised at how good it is and how few samples it takes to teach," he said.

Battery companies have used machine learning for years, but much of the battery industry hasn't yet embraced AI, said Richard Ahlfeld, CEO of Monolith, an AI software provider. The technology can cut battery testing time in half, he said.

"The EV race has gotten a lot hotter," he said. "People are now like, 'OK, what else can we do to go faster?' And this is a tool that has been proven that it can accelerate the development speed quite a lot."

Nio Europe said in September that it would use Monolith's technology to build a joint machine learning model for comparing current vehicle field data to bench-test data. It also would reduce time for battery data cleaning, resampling, analyzing and detecting abnormalities.

Companies are beginning to use AI to predict and optimize battery health through vehicles' battery management systems; clean, sort and restructure through ChatGPT; and map molecules to uncover next-generation materials.

Knowing the state of health can help drivers optimize charging and potentially extend the battery life by 10 to 20 percent, Ahlfeld said.

SES AI is working on AI models to map more molecules than humans can, said CEO Qichao Hu. The models can become as smart or smarter than top chemists, he said. SES AI believes the map will accelerate material discovery to solve any battery problem for EVs, electronics, grid storage and other applications.

But human scientists are key to making the database effective.

"The human scientists still need to synthesize the model, use and actually test the batteries. So it's almost like the idea creation that's done by the model, but the idea validation that's still done by humans," he said.

Advancing chemistries is the most exciting potential for AI in the battery space, said Patrick Hertzke, partner, automotive and assembly, at McKinsey & Co.'s Center for Future Mobility.

Numerous companies are doing incremental tests to improve batteries, he said.

"It's like making a vaccine or making a pharmaceutical drug. It's not easy, and it's not linear," Hertzke said. But based on breakthroughs in the pharmaceutical space, "you should equally be pretty excited about the potential for chemistry improvement in the battery space."

That potential could be years away, battery technology companies said.

"Battery manufacturing is more of an art than a science," Manan Pathak, CEO at BattGenie, said at The Battery Show. "It's extremely hard to have the end-to-end manufacturing process that can make really good, reproducible cells with very low error rates."

And even for catching defects, AI models require human training, Wang said.

"AI is another form of an algorithm," Wang said. "It's not a silver bullet. It's not magic. It's very good at predicting things."

Send us a letter

Have an opinion about this story? Click here to submit a Letter to the Editor, and we may publish it in print.

Please enter a valid email address.

Please enter your email address.

Please verify captcha.

Please select at least one newsletter to subscribe.

See more newsletter options at autonews.com/newsletters.You can unsubscribe at any time through links in these emails. For more information, see our Privacy Policy.