AI could help save firefighters’ lives by predicting fire flashovers before they occur, according to new research published this week.
Flashovers occur when combustible material in a room suddenly starts igniting all at once, leading to a huge surge of heat and flammable gases that can break walls and burst windows. Around 800 firefighters have been killed and more than 320,000 injured on the job in the US over a 10-year period, from 2008 to 2018, and it is estimated that 13 per cent of those accidents are the result of flashover events.
Firefighters have to rely on their experience to predict if a flashover is about to happen, such as judging from levels of smoke and heat, but it’s not easy considering how quickly they can creep up. Computer scientists have tried to develop methods capable of detecting flashovers in real time for the last two decades, but it’s a difficult task to model something so erratic.
Researchers from the US government’s National Institute of Standards and Technology (NIST), Google, as well as the Hong Kong Polytechnic University and the China University of Petroleum, built a system using graph neural networks (GNN) to learn relationships between different sources of data, represented as nodes and edges, from simulated fires.
“GNNs are frequently used for estimated time of arrival, or ETA, in traffic where you can be analyzing 10 to 50 different roads.” Eugene Yujun Fu, the study’s co-first author and a research assistant professor at the Hong Kong Polytechnic University, said in a statement.
“It’s very complicated to properly make use of that kind of information simultaneously, so that’s where we got the idea to use GNNs. Except for our application, we’re looking at rooms instead of roads and are predicting flashover events instead of ETA in traffic.”
The team simulated all sorts of data, from building layouts, surface materials, fire conditions, ventilation configurations, location of smoke detectors, and temperature profiles of rooms to model 41,000 fake fires in 17 different building types. A total of 25,000 fire cases were used to train the model, and the remaining 16,000 were used to finetune and test it.
The GNN’s performance was assessed by whether it was able to predict whether a flashover event would occur within the next 30 seconds. Initial results showed the model had an accuracy of 92.1 percent at best.
- Totaled Tesla goes up in flames three weeks after crash
- OVHcloud datacenter ‘lacked’ automatic fire extinguishers, electrical cutoff
- How hot is it right now? ‘Water park catching fire and burning down’ hot
- That’s a sticky Siemens situation: Former coder blows his logic bomb guilty plea deal in court
The system, dubbed FlashNet, is more advanced than the team’s previous machine learning model P-Flash.
“Our previous model only had to consider four or five rooms in one layout, but when the layout switches and you have 13 or 14 rooms, it can be a nightmare for the model,” said Wai Cheong Tam, co-first author of paper and a mechanical engineer at NIST. “For real-world application, we believe the key is to move to a generalized model that works for many different buildings.”
FlashNet may seem promising, but it is yet to be tested with data from real fire rescues. That would require the model to analyze data from thermostats, carbon monoxide and smoke detectors, in smart homes, Tam explained to The Register. How firefighters could then be alerted to the model’s predictions is unclear.
“The focus of the research was to rely on building data that is or could easily be provided from available building sensors. One way to translate the research into reality is to integrate the model into a smart fire alarm control panel that would gather the temperature data from installed heat detectors and includes a computer module that can process the data and make the real-time predictions.”
“From the fire alarm control panel or other suitable piece of equipment, the prediction would be sent to the incident commander, or individual firefighters if deemed suitable. The exact mechanism of providing such predictive analytics is not decided and would require input from the fire service to develop a consensus,” Tam concluded. ®