AI to guide future firefighting operations
As the saying goes, it’s always the worst cases that we don’t see coming; and it’s absolutely true in case of fires, more specifically something called flashover. Before you start wondering what it is, here’s a glimpse of it.
Flashover is a deadly fire phenomenon in which almost all combustible things within the vicinity ignite at random as well as together. Not only that, they tend to flare up all of a sudden at a temperature of nearly 600 degrees Celsius and rising. Yes, it is indeed as terrifying as it sounds. Now, in worst case scenarios, it’s often difficult to spot warning signs of impending flashover, especially when an entire area or a building is in flames. This not only results in the deaths of innocent civilians but also that of firefighters. In fact, flashover has been recognised as the leading cause behind their deaths.
However, brand-new research suggests that artificial intelligence (AI) can act as an essential forewarning to such first responders. This worthwhile and life-saving study has been carried out by researchers based at Hong Kong’s National Institute of Standards and Technology (NIST) and the Polytechnic University. They have created a model called FlashNet (Flashover Prediction Neural Network) that will help make predictions valuable seconds or minutes before a deadly flashover occurs. In fact, according to a recent experiment, FlashNet nearly outperformed all existing AI-based flashover forecasting tools and have even showcased a 92.1 per cent accuracy across a dozen distinguished residential floorplans in the USA. This latest study was published in the journal Engineering Applications of Artificial Intelligence.
Currently, the existing methods detect flashovers by using constant temperature data from burning infrastructures or employing machine learning (ML) to fill in the missing data, in case heat detectors fail to function. However, their primary problem was that they could only operate in single and known environments. But reality is often unknown and leads the firefighters into hostile territories with no prior knowledge of floorplans, emergency exits, exact location of the fire among others. Keeping this in mind, this new and generalised model has been developed that functions for all kinds of setups.
Alongside the AI-based FlashNet, researchers have also included graph neural networks (GNN) in the model, which is nothing but a kind of ML-based algorithm that helps in making judgements on nodes, lines and graphs that not only represent distinct data points but also their relationships with each other. So far, GNN has been effective in maintaining road traffic by estimating travel time and suggesting different routes.
As part of the study, researchers have virtually simulated around 41000 fires spread across 17 infrastructures in major US cities having varied factors such as distinct layouts, origin of fires, location of exits and so on. The study has even used 25000 fire cases as resources for their research, fine tuning and final testing. Apart from exhibiting massive accuracy within a 30 seconds lead time, the brand-new model was also able to detect false fires.