Machine-learning algorithm to save firefighter lives

Firefighters could soon be equipped with handheld devices to protect them from an explosive and potentially deadly fire phenomenon known as a backdraft. Backdrafts are notoriously hard to predict – but not for this machine-learning algorithm.

Researchers at the National Institute of Standards and Technology (NIST) conducted hundreds of fire experiments to determine the conditions needed for a backdraft to happen and fed it to an algorithm.

This resulted in a computer model that can predict a dangerous phenomenon in which a blow of fresh air resurrects a fire that was reduced by a lack of oxygen. Backdrafts typically occur when a door or a window is opened to a room where an air-deprived fire is smoldering. They can manifest in sudden and violent fireballs.

“Silently behind a door, it waits. One breath of oxygen and it explodes in a deadly rage,” is a much more poetic description from a tagline of Backdraft, a 1991 firefighter action flick starring Kurt Russel, William Baldwin, and Robert de Niro.

The movie quite accurately portrayed the challenges firefighters face in anticipating backdrafts – still a problem three decades later. NIST researchers hope to change that by incorporating their prediction model into small devices that could inform firefighters of dangers looming behind closed doors.

Instead of looking for visual indicators of a backdraft, such as soot-stained windows or puffing smoke, firefighters could probe the air of the closed room through small existing or newly-made openings. This would save them from potentially costly guesswork.

“If you can take measurements at the scene and reliably know the likelihood of a backdraft, you can open a door without taking as much of a risk,” NIST engineer Ryan Falkenstein-Smith said, adding that firefighters would also be able to tell whether they need to cool down the compartment before entering.

500 experiments

Falkenstein-Smith and his colleagues at NIST’s National Fire Research Laboratory carried out nearly 500 experiments for a recently published study where they lit a stream of gaseous fuel poured into a small metal chamber and then closed its door shut.

In each case, researchers continued to pump gas into a closed chamber for several minutes before the fire depleted available oxygen and burnt itself out. They would then open the door from a safe distance – sometimes nothing happened, but in other cases, fireballs erupted.

The researchers would alter factors such as the type and amount of gas injected into an enclosure and record temperatures, pressures, and dimensions of resulting fireballs. This allowed the team to pick up on certain trends and analyze the measurements.

“We aimed to capture all these different components that create the conducive conditions for a backdraft,” Falkenstein-Smith said.

Based on provided data, the machine-learning algorithm was able to establish a predictive backdraft model, which was correct in more than 70% of the experiments it was tested on. If it was provided with the chamber’s measurements, the accuracy rate increased to more than 82%.

Researchers at NIST are also working on a prediction model that could give firefighters an early warning of a flashover, another lethal fire phenomenon.

More from Cybernews:

US charges two Chinese “agents” with bitcoin bribery after sting operation

Russia takes lead as the most hacked country globally

TikTok a source of news for every tenth American

Black Friday or Bot Friday: bots will make your shopping experience miserable

Apple gets first bite of PayPal passkey login system

Subscribe to our newsletter

Leave a Reply

Your email address will not be published. Required fields are markedmarked