Scientists design AI-based system to warn firefighters of flashovers


Flashover is a very dangerous phenomenon, in which all combustible materials exposed in an enclosed area ignite almost simultaneously. Thanks to a new AI-based system, however, firefighters may soon receive warnings when flashover is imminent.

Typically, bypasses occur when a structural fire causes the temperature in an enclosed space to reach at least 593 ºC (1,100 ºF) – this is the approximate self-ignition temperature of many organic materials. commonly used. As a result, they will emit flammable gases and ignite more or less at the same time, even if they have not been directly exposed to the flames.

Needless to say, the firefighters not want to be in a building when a flashover occurs. However, given the chaos and limited visibility inside burning structures, it can be difficult for them to spot visible warning signs such as flames rolling through the ceiling. And while many buildings are fitted with heat sensors, devices typically stop working at around 150 ºC (302 ºF), long before flashovers occur.

This is where P-Flash – the prediction model for Flashover – is intended to come in.

Designed by scientists at the US-based National Institute of Standards and Technology (NIST), it uses machine learning algorithms that have been trained on 4,033 different computer simulations of a three-way ranch-style burning house. bedrooms and on one floor. In most states in the United States, this is the most common type of house.

One of the house fire scenarios that was used in the P-Flash training


Although this house’s virtual heat sensors are shut off at 150 ° C, the algorithms have learned to roughly predict when the temperature inside the house will reach 593 ° C, based on variables such as the speed at which the temperature was rising. When P-Flash was then tested on additional simulations of burning houses, it was 86% accurate in predicting flashovers one minute before they happened. Also, when he was less precise, he usually predicted them early… which is obviously better than predicting them late.

The system was then tested against actual data from 13 actual house fires, which were deliberately started in a ranch-style test house under varying conditions. While P-Flash remained specific in cases where the fire started in an open space like the living room, it was much less so when the fire started in closed rooms. This could be due to the fact that when fires are initially contained in such spaces, temperatures rise unpredictably as these fires then break out in the rest of the interior of the building.

Before and after shots inside the test house, which was intentionally set on fire as part of the Underwriters Laboratories experiments

Before and after shots inside the test house, which was intentionally set on fire as part of the Underwriters Laboratories experiments

UL Firefighter Safety Research Institute

Despite this current loophole, scientists believe that once developed, P-Flash could still become a valuable tool. By combining data from a home’s real-time heat sensor with information on variables such as its building materials, the system could send warnings to firefighters who might otherwise not know that a flashover was about to happen. to occur.

The research is described in an article recently published in Proceedings of the AAAI conference on artificial intelligence.

Source: NIST


About Author

Leave A Reply