The mining industry is increasingly switching to equipment with lithium-ion batteries. Such batteries are efficient, but under certain conditions they can overheat and catch fire, releasing toxic substances. In the confined space of underground workings this is especially dangerous.
Researchers from Pennsylvania State University (Penn State) built virtual models of mines to predict fire development in advance and test safety measures. The work was published in the journal Process Safety and Environmental Protection.
Computational fluid dynamics (CFD) was used to create the digital copies - a modeling method that shows how liquids and gases move under complex conditions. In this case, the scientists reproduced a typical underground room-and-pillar mining system: rock is extracted on a grid, leaving untouched pillars that support the roof.
Deep learning models - algorithms that find patterns in large data sets - were then trained on these virtual mines. They predicted how temperature, smoke and carbon monoxide change under different ventilation operation.
Lithium-ion batteries are dangerous because of so-called thermal runaway - an uncontrolled reaction in which the battery's internal temperature exceeds the permissible threshold and a self-sustaining process starts. It simultaneously releases heat, fire and oxygen, which sustains combustion. In a mine with limited air exchange, smoke and heat accumulate faster.
"Once thermal runaway has started, it is impossible to stop it - you can only reduce the consequences," explained lead author Khadija Omar Said, a doctoral student in the Department of Mineral and Energy Engineering.
Causes can include mechanical damage, high ambient temperature, repeated recharging due to control errors or a manufacturing defect. Tightly packed battery cells contribute to a chain reaction: overheating of one triggers heating of neighboring ones.
Using simulations, the team tested how fan speed affects the spread of toxic gases. It turned out that increasing airflow improves heat removal and dilutes smoke, giving miners time to evacuate. However, beyond a certain limit the effect hardly grows: flows from the fire begin to compete with the ventilation.
Accurate CFD models require large computational resources, so the researchers developed simplified deep learning models that provide fast predictions. They were tested on data from real mines and incidents: if a model reproduces values that were not in the training set, this increases confidence in its predictions.
The authors note: the work shows the fundamental possibility of predicting the consequences of thermal runaway in a confined environment under different ventilation modes. The next step is the creation of automatic control algorithms that could turn on fans or fire suppression systems. There is no talk of commercial application yet.