NIST trains AI to hear the 'oh crap' moment before batteries explode
Machine learning used to separate opening a can of soda from something catastrophic
Researchers at the US National Institute of Standards and Technology (NIST) have come up with an early warning system for lithium-ion battery fires.
The nature of lithium-ion battery fires is that by the time smoke triggers a traditional smoke alarm, the fire might be well under way. They also tend to get considerably hotter far faster than ordinary flames. According to NIST: "A battery can emit a jet of flame up to 1,100°C (2,012°F), almost as hot as a blowtorch, and it reaches that temperature in about a second."
Battery fires pose serious risks. The devices have been known to combust anywhere from the holds of ships to passenger compartments of aircraft and can cause disaster if not dealt with promptly.
Hence, a few minutes' notice of a potentially catastrophic failure would be very helpful.
One hint that things might be about to go awry is when the safety valve breaks in a hard battery case to release the pressure caused by a chemical reaction within the unit. NIST described it as "a distinctive click-hiss, a little like the sound of cracking open a bottle of soda."
While the sound has been noted in previous studies, the issue with using it as an early warning system is that there are plenty of similar sounds, such as the dropping of a paper clip. The challenge facing researchers was removing the false positives.
- South Korean lithium battery plant blaze kills at least 23
- Samsung shows off battery tech it says will see you gone in nine minutes
- Toyota, Samsung accelerate toward better EV batteries
- A ship carrying 800 tonnes of Li-Ion batteries caught fire. What could possibly go wrong?
The solution came from machine learning. The NIST team took the audio from 38 exploding batteries and tweaked the pitch and speed to create more than 1,000 audio samples, which could then be used to teach the software what a breaking safety valve sounds like. Overheating batteries were subsequently detected 94 percent of the time.
One of the NIST team, Wai Cheong "Andy" Tam, said: "I tried to confuse the algorithm using all kinds of different noises, from recordings of people walking to closing doors, to opening Coke cans. Only a few of them confused the detector."
With lithium-ion batteries appearing in more settings, an early warning system – even if it provides just a minute or so – would be a welcome innovation. NIST reckoned the alarm, once fully developed, could find its way into homes, office buildings, and electric vehicle parking garages. The NIST team noted that the safety valve gave two minutes' notice in testing, but planned to continue the research with a wider variety of batteries.
"The key to a healthy relationship with lithium-ion batteries might just be careful listening," NIST said. ®