Researchers at the National Institute of Standards and Technology (NIST) have developed P-Flash: a tool that predicts and warns the deadly phenomenon of flashover in burning buildings. The tool can work even after heat detectors begin to fail.
Firefighting is a risky job, especially when fires in buildings can go from serious to deadly in moments. The phenomenon of a flashover occurs when combustible materials in a room ignite almost simultaneously, creating a flame that is limited only by the available oxygen.
Determining when this is going to happen — noticing rising temperatures or flames rolling across the ceiling — is extremely difficult amid the many time-critical responsibilities firefighters have in the chaos of a fire.
“I don’t think the fire service has a lot of technology tools to predict flashover on the spot,” said NIST researcher Christopher Brown, who is also a volunteer firefighter. “Our biggest tool is simply observation and that can be very deceiving.
From the outside it looks one way and when you come in it can be very different.” The NIST researchers developed P-Flash (the Prediction Model for Flashover) to predict and alert an incoming flashover; Incredibly, it’s designed to keep working after heat detectors start to fail, using residual devices to monitor the fire.
While models for predicting flashover based on temperature are not new, they have until now relied on uninterrupted streams of temperature data, which can be obtained under controlled laboratory conditions, but are by no means guaranteed during a real fire.
Heat detectors typically only operate at temperatures up to 150°C – well below 600°C where a flashover begins. To bridge the gap created by lost data, NIST researchers applied machine learning.
“You lose the data, but you have the trend to where the heat detector fails and you have other detectors. With machine learning, you could use that data as a starting point to extrapolate whether flashover is going to happen or has already happened,” says NIST chemical engineer Thomas Cleary:
The researchers used temperature data from heat detectors in an advanced digital twin of a burning house to train their model, because for obvious reasons it was not feasible to burn hundreds of houses to collect data.
They conducted more than 5,000 simulations with variations between each, such as the order in which furniture pieces ignited and which windows and doors were closed.
The researchers trained the model on one batch of data and then tested it on the remaining batches, refining it based on its performance in predicting when a flashover would occur.
They found that the model correctly predicted flashovers one minute in advance for about 86 percent of the simulated fires. Even if it missed the mark, it usually did so by producing false positives, which is preferable to giving firefighters a false sense of security.
“You always want to be on the safe side. While we can accept a small number of false positives, our model development is committed to minimizing or, better yet, eliminating false negatives,” said NIST mechanical engineer and co-author Wai Cheong Tam.
After this, they tried out their model using real-world data produced during a recent survey at a ranch-style home, similar to their digital version. They found that their model lacked a phenomenon called the envelope effect:
When fires burn in small, enclosed spaces, heat has little ability to dissipate, so the temperature rises quickly. However, many of the experiments that form the basis of P-Flash’s training materials were conducted in open lab spaces, so the temperatures of the real experiments rose almost twice as fast as the synthetic data.
Despite finding this flaw in the tool, the researchers are encouraged and plan to reflect the envelope effect in their simulations through more large-scale experiments.
They hope that eventually they will be able to provide a firefighters hand tool that communicates with detectors in burning buildings, giving firefighters an essential safety tool.