According to the researchers, this safety valve breaks and makes a distinct clicking sound, a bit like the sound of opening a soda bottle.

The ML algorithm can recognize the sound of a safety valve breaking

Researchers at the National Institute of Standards and Technology (NIST) trained a machine learning algorithm that can recognize the sound of a breaking safety valve.

Researchers first needed many examples of sounds to make the algorithm work.

In collaboration with a laboratory at Xi’an University of Science and Technology, they recorded the sounds of 38 exploding batteries.

Then researchers adjusted the speed and pitch of these recordings to expand them into more than 1,000 unique sound samples that they could use to teach the software what a struggling safety valve sounds like as.

The algorithm detects the sound of an overheated battery 94% of the time

Researchers claimed that the algorithm works remarkably well as they discovered the sound of an overheated battery 94% of the time with a microphone mounted on a camera.

“I tried to confuse the algorithm using all sorts of different sounds, from recordings of people walking, to closing doors, to opening Coke cans.” explained Wai Cheong “Andy” Tam of NIST. “Only a few of them confused the detector.”

Researchers used 1,128 samples of acoustic data, including various human activities, to facilitate the development of a detection model that can be used in real-world environments.

“Using 10-s acoustic data as input and a neural network model structural structure such as the backbone, the detection model has an overall accuracy of 93.9% with a precision and recall score of 91.6% and 97.7%, respectively,” researchers explained in their study.

Parametric studies are performed to evaluate the robustness of the proposed model structure and the effectiveness of the data augmentation methods. In addition, the model’s performance is assessed against two full tests using leave-one-test cross-validation, according to researchers.

It is also claimed that the proposed work can help develop a robust detection device that can provide early warning of thermal fugitives and give users extra time to mitigate potential extreme fire hazards and/or safely evacuate.