This article is more than 1 year old

This AI could save a firefighter's life

When it works well enough in the real world, that is

Computer scientists have built an algorithm to predict deadly explosions within blazes before they occur in the hope it can one day serve as a warning system for firefighters tackling burning buildings.

Furniture and other stuff that's made out of combustible materials can, in the heat of a fire, suddenly erupt into balls of flames that spread quickly across a closed space, such as a room or floor of a building. These incidents, known as flashovers, can break windows and knock down walls, and are one of the most dangerous risks for firefighters.

Researchers led by the US government's NIST are trying to develop technology capable of alerting firefighters to flashovers before they happen. At the heart of the tech is a neural network trained to predict these explosions from the building's temperature sensors. And yes, it's supposed to be a little more sophisticated than an algorithm that just raises the alarm when the temperature goes over a certain level as these situations are usually more complex than that.

“I don't think the fire service has many tools technology-wise that predict flashover at the scene,” said NIST researcher Christopher Brown, who was not directly involved with the research and is a volunteer firefighter.

“Our biggest tool is just observation, and that can be very deceiving. Things look one way on the outside, and when you get inside, it could be quite different.”

The team created computer simulations of fires to not only teach the neural net to predict an imminent flashover but also evaluate its abilities, generating 4,033 training examples, and 504 test cases. The system, nicknamed P-Flash, was at least 80 per cent accurate with these simulations, according to a paper detailing the research that was published in the Proceedings of the AAAI Conference on Artificial Intelligence. Most of the errors were false positives, the researchers said.

However, the accuracy dropped to as low as 25 per cent when the algorithm was given temperature sensor data taken from real fires in 13 experiments conducted by the UL Firefighter Safety Research Institute, depending on what room the fire started in.

P-Flash was better at predicting flashovers 30 seconds in advance in areas that were more open, such as kitchens and living rooms, and it struggled with smaller spaces, such as bedrooms. Temperature readings in flashovers in confined areas shoot up more quickly. Since P-Flash wasn’t trained on data that fluctuated as wildly, its performance was less precise in those situations.

Our ability to accurately simulate fire conditions in buildings places a limit on model accuracy in predicting real fires

“Our ability to accurately simulate fire conditions in buildings places a limit on model accuracy in predicting real fires,” Wai Cheong Tam, a co-author of the paper and a NIST mechanical engineer, told The Register. “Improvement in fire models would help [improve the model].”

To get their model tested and working in real-world conditions, the team has to figure out a way to gather data in real time from a building's heat sensors and feed that information into devices running the machine-learning software. Such a device would need to be able to make predictions on the fly, and alert firefighters on the scene.

“The exact mechanism is not decided,” Thomas Cleary, a chemical engineer at NIST and a co-author of the study, told us. “The fire service will ultimately decide how to use this information. One could imagine a head-up display with various data streams.”

“In the next year we are planning to conduct experiments to demonstrate and test the algorithm [in real-time] in building fire tests with heat detectors,” he added. ®

More about


Send us news

Other stories you might like