AI Steps Up: Detecting Lithium Battery Fires in Real Time Through Sound

Lithium-ion batteries are integral to modern life, powering everything from smartphones to electric vehicles. However, their energy-dense design comes with a risk: overheating or damage can lead to catastrophic fires. Recognizing this growing concern, researchers at the National Institute of Standards and Technology (NIST) have developed an innovative method to detect early signs of battery failure using sound.

Image: Time-lapse footage of one of the experiments shows the safety valve breaking about two minutes before the battery explodes.

The Risk Behind Lithium-Ion Batteries

With their ability to store large amounts of energy in a compact form, lithium-ion batteries are ubiquitous. Yet, their potential to catch fire or even explode under stress is a pressing safety issue. In 2023 alone, New York City reported 268 residential fires caused by e-bike batteries, resulting in 150 injuries and 18 fatalities.

These fires are particularly dangerous because of how rapidly they escalate. Unlike traditional house fires, which often start small, lithium-ion battery fires can emit flames reaching 1,100°C (2,012°F) within seconds, leaving little time to respond. Compounding the problem, these batteries produce minimal smoke at the onset, making standard smoke alarms ineffective at providing early warnings.

 A Unique Sound as a Warning Signal

 NIST researcher Wai Cheong “Andy” Tam observed an intriguing phenomenon while watching videos of exploding batteries. Right before a fire, the safety valve on the battery casing breaks, releasing a distinctive "click-hiss" sound as gases escape. Inspired by this, Tam and his colleague, Anthony Putorti, explored whether this sound could serve as an early warning for battery failure.

Their hypothesis: before a lithium-ion battery reaches the point of ignition, pressure builds inside due to chemical reactions. This pressure causes the safety valve to release gases with a unique acoustic signature, signaling imminent danger.

Training AI to "Hear" Battery Failures

To create a reliable detection system, the NIST team trained an AI algorithm to recognize the "click-hiss" of a breaking safety valve. Collaborating with Xi’an University of Science and Technology, they recorded sounds from 38 exploding batteries. By modifying the speed and pitch of these recordings, the researchers generated over 1,000 unique samples for training the algorithm.

The result? A highly effective system capable of identifying the sound of a failing battery 94% of the time, even in noisy environments. From footsteps to door slams, the algorithm successfully filtered out irrelevant noises, demonstrating its potential for real-world applications.

A Vision for Safer Spaces

Preliminary tests revealed that the safety valve's sound could be detected up to two minutes before a battery failure. This advanced warning could be life-saving, offering critical time for evacuation or intervention. Future developments could see this technology integrated into fire alarms for homes, offices, warehouses, and electric vehicle parking areas.

Tam and Putorti's groundbreaking research, presented at the 13th Asia-Oceania Symposium on Fire Science and Technology, has laid the groundwork for a new era of safety. By listening carefully to the sounds of our batteries, we can mitigate risks and make our environments safer for everyone.

Lithium-ion batteries power our world, but their risks demand innovative solutions. By combining sound recognition with AI, NIST researchers have taken a major step toward preventing battery-related fires. This technology promises a future where listening closely could save lives and protect property.

Source of content including image: National Institute of Standards and Technology (NIST)