AI can hear when a lithium battery is about to catch fire

The rapid growth in lithium-ion battery usage has led to increasing safety concerns, particularly as battery fires pose unique challenges with their intense heat and rapid onset.

The innovation breakthrough: NIST researchers have developed an AI-powered system that can detect the distinctive sound made by lithium-ion batteries moments before they catch fire.

• The system recognizes a specific “click-hiss” sound, similar to opening a soda bottle, which occurs when a battery’s safety valve breaks due to internal pressure
• Researchers Wai Cheong “Andy” Tam and Anthony Putorti trained the AI using over 1,000 unique audio samples derived from 38 exploding battery recordings
• Initial testing shows a 94% success rate in detecting the warning sound, even in noisy environments

Understanding the urgency: Battery fires represent a growing safety threat, particularly in urban environments where lithium-ion powered devices are increasingly common.

• New York City reported 268 residential e-bike battery fires in 2023, resulting in 18 deaths and 150 injuries
• Battery fires can reach temperatures of 1100°C (2012°F) within seconds, comparable to a blowtorch
• Unlike traditional fires that begin slowly, battery fires provide minimal early warning signs and limited smoke for conventional detectors to sense

Technical implementation: The AI system addresses unique challenges in distinguishing battery failure sounds from similar everyday noises.

• Researchers collaborated with Xi’an University of Science and Technology to gather initial audio samples
• The system successfully filters out similar sounds like staplers or dropping paper clips
• Testing revealed approximately two minutes of warning time between the safety valve breaking and catastrophic battery failure

Future applications: The technology shows promise for widespread implementation in various settings where lithium-ion batteries are present.

• Potential applications include homes, office buildings, warehouses, and electric vehicle parking facilities
• The system could provide critical evacuation time currently lacking with traditional fire detection methods
• Researchers are expanding their testing to include more battery types and microphone configurations

Looking ahead – balancing innovation and safety: While this detection system represents a significant advancement in battery fire prevention, it highlights the ongoing challenge of managing the risks associated with energy-dense power sources in our increasingly electrified world. The success of this technology could influence future battery safety regulations and standards.