Researchers from the National Institute of Standards and Technology (NIST) announced the development of a method that uses artificial intelligence to determine when a lithium-ion battery could catch fire.
Popular in various products such as phones, laptops and electric vehicles, lithium-ion batteries can store a large amount of energy in a compact space. However, they pose a safety risk, as they can catch fire or explode if they overheat.
Most Read on IEN:
NIST reported that these fires can produce a jet of flame that reaches up to 2,012° Fahrenheit, nearly the heat of a blowtorch, in approximately one second. This hazard differs from traditional residential fires that start more slowly, which allows for smoke to reach a smoke alarm before the fire spreads.
During the research, NIST mechanical engineer Andy Tam said he noticed that a battery’s safety valve would break right before the fire started and make a small “click-hiss” sound that resembled opening a bottle of soda. Lithium-ion battery manufacturers design this safety valve to break when internal pressure builds up and can no longer expand due to the battery’s hard casing.
Tam and his colleague, Anthony Putorti, then found they could use AI to train a machine-learning algorithm to identify this specific sound. To accomplish this, they recorded audio from 38 exploding batteries and modified the pitch and speed of the recordings to generate over 1,000 unique audio samples, which they used to train the software.
Tam and Putorti reported that their algorithm accurately detected the sound of an overheating battery with 94% accuracy. Tam explained that he attempted to confuse the algorithm with other noises but said only a few tricked the detector.
Additional research estimated that a battery’s safety valve would break approximately two minutes before catastrophic failure. Tam and Putorti applied for a patent and hope to verify this warning time with further experiments on a range of batteries.
Once developed, the technology could serve as a new type of fire alarm in homes, offices, warehouses and garages.
Click here to subscribe to our daily newsletter featuring breaking manufacturing industry news.
WEBVTT
X-TIMESTAMP-MAP=LOCAL:00:00:00.000,MPEGTS:0
00:00.009 --> 00:03.849
Researchers from the National Institute of
Standards and Technology announced the
00:03.859 --> 00:08.149
development of a method that uses artificial
intelligence to determine when a lithium ion
00:08.159 --> 00:11.600
battery could catch fire.
Popular in various products such as phones,
00:11.609 --> 00:15.779
laptops and electric vehicles, lithium ion
batteries can store a large amount of energy in
00:15.789 --> 00:16.790
a compact space.
00:17.000 --> 00:21.069
However, they pose a safety risk as they can
catch fire or explode if they overheat.
00:21.200 --> 00:25.610
Nist reported that these fires can produce a
jet of flame that reaches up to 2002,
00:25.659 --> 00:29.829
12 °F nearly the heat of a blowtorch in
approximately one second.
00:29.889 --> 00:33.830
This hazard differs from traditional
residential fires that start more slowly which
00:33.840 --> 00:36.479
allows for smoke to reach a smoke alarm before
the fire spreads.
00:36.490 --> 00:38.900
During the research.
Nist mechanical engineer,
00:38.909 --> 00:43.580
Andy Tam said he noticed that a battery safety
valve would break right before the fire started
00:43.590 --> 00:47.520
and make a small click hiss sound that
resembled opening a bottle of soda,
00:47.529 --> 00:51.909
lithium ion battery manufacturers designed this
safety valve to break when in internal pressure
00:51.919 --> 00:55.220
builds up and can no longer expand due to the
battery's hard casing.
00:55.229 --> 00:59.979
Tam and his colleague, Anthony Por then found
they could use A I to train a machine learning
00:59.990 --> 01:03.000
algorithm to identify the specific sound to
accomplish this.
01:03.009 --> 01:07.410
They recorded audio from 38 exploding batteries
and modified the pitch and speed of the
01:07.419 --> 01:11.930
recordings to generate over 1000 unique audio
samples which they used to train the software.
01:11.940 --> 01:16.510
Tam and poor reported that their algorithm
accurately detected the sound of an overheating
01:16.519 --> 01:20.379
battery with 94% accuracy.
Tam explained that he attempted to confuse the
01:20.389 --> 01:23.379
algorithm with other noise but only said a few
tricked the detector.
01:23.389 --> 01:27.300
Additional research estimated that a battery
safety valve would break approximately two
01:27.309 --> 01:28.980
minutes before catastrophic failure.
01:29.220 --> 01:33.019
Tam and Pity applied for a patent and hoped to
verify this warning time with further
01:33.029 --> 01:37.000
experiments on a range of batteries once
developed, the technology could serve as a new
01:37.010 --> 01:40.379
type of fire alarm in homes, offices,
warehouses and garages.
01:40.389 --> 01:42.419
I'm Nolan Beein.
This is manufacturing now.
