Professor Joshua Pulsipher is part of a team of researchers that have created a data-driven analysis system that could reveal how fires behave to better inform the building code, evacuation plans and advise first responders.
“We want to deepen our understanding so we can predict fire behavior and the gases it releases, enabling smart systems that support safer and more effective fire evacuations,” says Pulsipher a chemical engineering professor.
Fire behavior knowledge has not kept pace with today’s architecture and furnishing materials.Fabric and foam in modern furniture can produce toxic gases when on fire and the make-up and quantity of gases change during the evolution of a building fire.
Historically, older buildings were highly ventilated. However, modern buildings are sealed for energy efficiency with a vapour barrier that blocks airflow from the outside resulting in oxygen-starved conditions and fires leading to inefficient combustion, which create more toxic gases along with smoke.
Professors Beth Weckman and Vinny Gupta experts in Fire Safety Engineering and fire behaviour are part of the team that conducted a series of 15 residential fire experiments in a burn house near the Waterloo campus.
The building is equipped with sensors that measure temperature, airflow, humidity, heat flux and the concentration of many different types of gases generated by fire including CO2, carbon monoxide, oxygen, methane, volatile organic compounds and hydrogen cyanide.
The framework leverages principal component analysis (PCA), a math tool that reduces sensor signals into a smaller number of patterns, K-means clustering, a method to group similar sensor patterns and sparse-partial least squares (S-PLS) to link sensor data to outcomes, such as fire growth or toxicity, selecting the most important sensors.
“The input is the data from the experiments, the output is understanding what those experiments tell us and reveal fundamental insights about how the underlying fire evolves,” says Gupta.
The system uses statistical relationships from the data to determine when a fire begins to change its burning patterns and starts under‑ventilating, meaning it’s running out of oxygen.
This shift changes the combustion chemistry and produces more toxic smoke. By detecting this transition, researchers can understand how different fires will behave and which chemicals are most important to monitor to predict that change.
“Over the long term, we want to create smart systems that model and anticipate what a fire will do and then route people to get out of the building safely,” says Weckman a member of the Fire Research Group at Waterloo.
In future research, the framework will be extended to more complex fire scenarios, with greater spatial variation and ventilation.
The study, A framework for high-dimensional fire sensor data analysis was recently published in the Fire Safety Journal.