Machine learning researched as battery fire detection technique

Researchers at the Gaithersburg, Maryland-based National Institute of Standards and Technology (NIST) say they have developed a way to use sound to detect when lithium-ion batteries are about to catch fire.

The NIST team included Wai Cheong “Andy” Tam and Anthony Putorti.

Discarded lithium-ion batteries have been the cause of numerous fires at electronics and metals recycling plants, material recovery facilities (MRFs) and waste transfer stations.

According to NIST, a lithium-ion battery doesn’t produce much smoke to detect during the first stages of failure, making the fires dangerous because they get "blisteringly hot" almost instantly.

“While watching videos of exploding batteries, I noticed something interesting," Tam says. "Right before the fire started, the safety valve in the battery broke and it made this little noise. I thought we might be able to use that.”

Tam and NIST acknowledge they are not the first to notice the sound, but a barrier to following up on the finding is the existence of other sounds that are similar to a breaking safety valve, such as using a stapler or dropping a paper clip.

Tam and Putorti, though, have trained a machine-learning algorithm to recognize the distinct noise of a lithium-ion battery safety valve breaking.

The duo worked in collaboration with a laboratory at Xi’an University of Science and Technology in China, gaining access to recorded audio from 38 exploding batteries. They tweaked the speed and pitch of those recordings to expand them into more than 1,000 unique audio samples they could use to teach the software what a breaking safety valve sounds like.

Using a microphone mounted on a camera, the researchers reportedly detected the sound of an overheating battery correctly 94 percent of the time.

To what extent the sound detection technique can be tied to current fire suppression methods in the recycling and waste sectors likely will take additional research.