A machine learning-based solution could help firefighters avoid deadly backdrafts
Lack of oxygen can turn even the hottest flame into smoldering ashes. But when fresh air comes in, let’s say a firefighter opens a window or door in a room, the fire can revive suddenly and violently. This explosive phenomenon, called a backdraft, can be fatal and is challenging for firefighters to anticipate.
Now, researchers at the National Institute of Standards and Technology (NIST) have come up with a plan to inform firefighters about the dangers lurking behind closed doors. The team obtained data from hundreds of backdrafts in the lab to use as the basis for models that predict backdrafts. The results of a new study, described at the 2022 Suppression, Detection and Signaling Research and Applications Conference, show that the model offers a viable solution for making predictions based on specific measurements. In the future, the team seeks to implement the technology in smaller-scale devices that firefighters could deploy in the field to avoid or adapt to dangerous conditions.
Read More : Top 10 Data Science Skills That Will Transform Your Career In 2022!
Currently, firefighters look for visual indicators of a possible backdraft, including soot-stained windows, smoke pouring through small openings, and the absence of flames. If signs are present, they can vent the room by making a hole in the ceiling to reduce their exposure. If not, they can come in right away. Ultimately, first responders must rely on their eyes in hazy environments to anticipate the correct course of action. And miscalculating can pay a heavy price.
“If you can take measurements on the scene and know reliably the potential for backdraft, you can open the door without risk. Or you can be more confident that you need to cool the compartment before entering.” is, either by moving up or down space through small openings,” said NIST engineer Ryan Falkenstein-Smith.
At NIST’s National Fire Research Laboratory, Falkenstein-Smith and her colleagues conducted experiments where they lit a stream of gaseous fuel that was piped into a small chamber and then closed its door. In each case, the door remained closed for several minutes as researchers continued to pump gas into the chamber and the fire burned itself out, reducing its available oxygen.
Then, from a safe distance, he remotely opened the door. Some experiments were relatively uneventful, with no indication of governance. In others, balls of fire, accompanied by waves of pressure, burst into the doorway, Falkenstein-Smith said.
Over the course of nearly 500 experiments, in which the researchers varied factors such as the type and amount of gas injected into the chamber, they took measurements that ran the gamut. He recorded the temperature, pressure, dimensions of the fireball, and much more. To specifically determine fuel abundance, they improved upon an instrument NIST developed decades earlier called a Phi meter.
The meter took a sample of the fuel and air gas mixture from the chamber, added a known amount of oxygen and then internally combusted the sample, measuring the difference in oxygen before and after. The less oxygen is consumed in the reaction, the higher the relative abundance of fuel in the mixture.
“We aimed to capture all these different components that create favorable conditions for a backdraft,” Falkenstein-Smith said.
Read More : The 10 Most Atrocious Python Mistakes Aspirants Often Make!
The team analyzed the measurements and picked up on some trends. For example, adding fuel to the chamber at high rates coincides with a high probability of backdraft. To get the most out of the data, the researchers also used a machine learning algorithm to set up a predictive backdraft model from their treasure trove of information.
As a preliminary test of the model, they fed it readings of gas concentration, fuel richness and temperature at the same spot in the chamber before opening the door during their experiments. Based on that information alone, the model had to estimate the likelihood of a backdraft occurring.
Taking predictions above 50% as a positive prediction and below 50% as negative, the model was correct in 70.8% of the experiments on which it was tested. And the accuracy increased to 82.4% with a measurement taken at a second location in the room.
Falkenstein-Smith said the team is confident in its technology and aims to keep the ball rolling, improving performance and improving the practicality of the technology.
The next step is to develop a portable device that combines the measurement techniques used in the lab with their computer models and then battle-testing the techniques in a more realistic building fire scenario. The team anticipates that firefighters using the handheld device will either probe the room’s air through existing openings, such as cracks around doorways, or create smaller openings.
