House fires have changed over the years as modern building practices and synthetic furnishing change how things burn. Now, researchers are hoping artificial intelligence can help firefighters and residents anticipate how those fires behave. Researchers at the University of Waterloo have developed a new AI framework that analyzes fire progression. They hope the analysis will improve building codes, help emergency responders and power future home evacuation systems. The team conducted more than a dozen controlled tests inside a two-story burn building on campus to observe how modern furniture burns under different ventilation conditions. “With many of the types of additives that we put in modern building furnishings or materials like couches, they may slow the growth of a fire, but it may have a trade-off,” said Vinay Gupta, an assistant professor of fire safety engineering at the university. “It could produce more smoke, or it could produce more types of gases that are quite harmful for people.” Gupta said modern, energy-efficient building practices compound those risks. Because newer homes are built airtight. “What that essentially means is that the fire can reach conditions where you consume all the oxygen available to keep it burning,” he said. To monitor those changes, researchers outfitted the burn house with up to 175 sensors tracking temperature, airflow and gas concentrations, generating hundreds of millions of data points. “AI is looking at hundreds of millions of sensor measurements simultaneously to try to uncover patterns that go across the experiments and try to reveal insights that humans simply can’t divine by just staring at Excel spreadsheets,” said Joshua Pulsipher, an assistant professor of chemical engineering. Researchers said the framework could eventually integrate into commercial smart home networks to guide occupants safely out of burning structures. Beyond future home evacuation systems, the findings could provide valuable insight for future updates to building codes and help emergency services. “The idea is to try and provide people that are responding to the fire both during an active fire event with the most amount of information possible so that they can improve their decision making,” said Gupta. The research team plans to adapt the AI framework to analyze other fire safety issues like lithium-ion battery fires and wildfires in the future.
