Burn Notice

Take a few moments and have a look around you. Go ahead, I’ll wait right here. Did you take it all in? If you are observant, you likely noticed a large number of devices powered by rechargeable batteries in your immediate vicinity — and maybe even in your pocket or on your wrist.

Today’s rechargeable battery technologies have incredibly high energy densities and can discharge that power very rapidly. These characteristics have made them invaluable in powering everything from smartphones and earbuds to drones and even vehicles. Batteries may not be the main topic discussed around the average water cooler, but the modern world we have grown accustomed to would not be possible without them.

But before heaping too much praise on existing technologies, I feel that I should point out one very minor, teensy weensy little flaw found in many types of rechargeable batteries. Namely, they have a tendency to burst into fire, shoot out jets of superheated flames, and explode. It could be worse, right? Right?!?

Nothing is perfect, but that does sound pretty bad when you say it out loud. These problems arise from the same characteristics that make these batteries so valuable — high energy densities and quick discharge rates. When those discharge rates run out of control, either due to physical damage or overheating, there is no stopping the reaction until it burns itself out.

Adding insult to injury, these situations arise very quickly, and they initially release very little smoke, making them difficult to detect via traditional means. As more and more batteries are brought into our homes and workplaces, the potential for disaster will continue to rise. But help may be on the horizon, thanks to the work of a team at the National Institute of Standards and Technology. They have taken a novel approach that involves listening for the sound of a failing battery to detect battery fires before they even start.

Before catching fire, lithium-ion batteries swell up as a result of the internal chemical reactions that are happening. Typically, these batteries have pressure relief valves that are meant to avoid explosions. When these valves are triggered, they make a distinctive popping sound. This realization was the key to the team’s invention.

An artificial intelligence algorithm was developed that is capable of recognizing the distinctive characteristics of different types of sounds. The algorithm needed to be trained on examples of failing battery pressure valves, but that is a very hard dataset to collect. So rather than collecting the hundreds or thousands of examples that might be needed, the researchers instead collected audio samples from 38 failing batteries.

That is not enough data to develop a well-generalized model, so they then artificially manipulated the audio clips — by modifying their speed and pitch — to give themselves a set of 1,000 data points. This may not be the ideal solution, but it did prove to work quite well. Testing revealed that the trained model could identify failing batteries correctly in about 94 percent of cases.

After further refining their methods, the team hopes that their technology might one day be incorporated into a new type of fire alarm. They envision these devices being installed in homes, office buildings, and warehouses where they might be able to save both lives and property. There is still a lot of work to be done before we know that this system will be effective in the real world, but it is certainly a step in the right direction.