AI may potentially support future firefighting operations: research

December 19, 2022 5:40 am IS

Washington [US], December 19 (ANI): The worst flames in firefighting are the ones you can’t see coming. Amidst the chaos of a burning building, it’s difficult to spot the warning signs of an imminent flashover – a deadly fire phenomenon in which nearly all combustible objects in a room spontaneously ignite. Flashovers are a leading cause of firefighter deaths, but new research suggests that artificial intelligence (AI) could give first responders some much-needed advance warning.
Researchers from the National Institute of Standards and Technology (NIST), Hong Kong Polytechnic University and other institutions have developed a Flashover Prediction Neural Network (FlashNet) model to predict fatal events seconds before they happen.
In a recent study published in Engineering Applications of Artificial Intelligence, FlashNet outperformed existing AI-based flashover forecasting tools with an accuracy of up to 92.1% on more than a dozen popular US residential floor plans.
Lightning flashes usually flare up suddenly at around 600 degrees Celsius and can then cause temperatures to skyrocket further. To predict these events, existing research tools either rely on constant streams of temperature data from burning buildings, or use machine learning to supplement the missing data in the likely event that heat detectors succumb to high temperatures.
To date, most machine learning-based prediction tools, including one previously developed by the authors, have been trained to work in a single, familiar environment. In reality, firefighters are not afforded such a luxury. When charging into enemy territory, they may know little to nothing about the layout, the location of the fire, or whether doors are open or closed.

“Our previous model only had to account for four or five rooms in a layout, but when the layout switches and you have 13 or 14 rooms, it can be a nightmare for the model,” said NIST mechanical engineer Wai Cheong Tam, co-first author the new study. “For real-world application, we believe the key is to move to a generalized model that works for many different buildings.”
To deal with the variability of real fires, the researchers reinforced their approach with Graph Neural Networks (GNN), a type of machine learning algorithm good at making judgments based on graphs of nodes and lines representing different data points and depict their relationships with one Another.
“GNNs are commonly used for Estimated Time of Arrival or ETA in traffic where you can analyze 10 to 50 different roads. It’s very complicated to properly use this type of information at the same time, so we came up with the idea of ​​using GNNs,” said Eugene Yujun Fu, research assistant professor at Hong Kong Polytechnic University and co-first author of the study. “Except for In our application, we look at spaces instead of streets and predict flashover events instead of ETA in traffic.”
Researchers digitally simulated more than 41,000 fires in 17 types of buildings, representing much of the US housing stock. In addition to the arrangement, factors such as the seat of the fire, the type of furniture and whether doors and windows were open or closed varied. They provided the GNN model with a set of nearly 25,000 fire cases to use as study material, and then 16,000 for fine-tuning and final testing.
For the 17 types of homes, the accuracy of the new model depended on the amount of data it needed to process and the lead time it wanted to give firefighters. However, the accuracy of the model — 92.1% at best with a 30-second lead time — outperformed five other machine learning-based tools, including the authors’ previous model. Crucially, the tool produced the fewest false negatives, dangerous cases where the models failed to predict an imminent flashover.
The authors threw FlashNet into scenarios where it had no prior information about the specifics of a building and the fire burning within it, similar to the situation firefighters often find themselves in. Given those limitations, the tool’s performance has been quite promising, Tam said. However, the authors still have a long way to go before they can take FlashNet across the finish line. As a next step, they plan to test the model with real rather than simulated data.
“To fully test the performance of our model, we need to actually build and fire our own structures and put some real sensors in them,” Tam said. “Ultimately, if we’re going to use this model in real fire scenarios, that’s a must.” (ANI )