1
24-017
Characterizing Fire in Large Underground Ventilation Networks
Using Machine Learning
D. Bahrami
CDC NIOSH, Pittsburgh, PA
Y. Xue
CDC NIOSH, Pittsburgh, PA
L. Zhou
CDC NIOSH, Pittsburgh, PA
L. Yuan
CDC NIOSH, Pittsburgh, PA
ABSTRACT
Underground mine accidents, such as mine fires, remain
a health and safety risk to mine workers. Researchers at
the National Institute for Occupational Safety and Health
(NIOSH) are developing a data-driven, predictive model
for characterizing the location and size of unknown under-
ground fires. This study examines applying a machine
learning-based model to predict fire size and location in a
large underground metal mine based on hypothetical sce-
narios on the model performance. The results show that
the size and location of an unknown fire can be determined
with over 80% and 90% accuracy, respectively, and poten-
tially help to reduce the risk of hazardous conditions for
emergency response.
INTRODUCTION
Mine fires continue to occur, although at a very low occur-
rence rate however, they remain a health and safety risk to
both surface and underground mining operations. Smoke
and toxic gas are hazardous results of an underground
mine fire, which flows to other areas of the mine via the
ventilation network [1] posing a higher health and safety
risk compared to the surface mine fires. Equipment fires
have been recognized as being responsible for most min-
ing injuries during 2000–2013 [2]. To reduce fire-caused
injuries in underground mines, it is important to ensure
the safety of the underground mine environment during a
fire emergency.
In underground mines, an Atmospheric Monitoring
System (AMS) is employed to monitor air quality param-
eters such as air velocity and other gasses concentrations
such as methane, CO, and CO2. The use of an AMS for
early warning and fire detection has a significant potential
to enhance the safety and well-being of underground min-
ers [1]. The AMS has been used to develop methodologies
to characterize an unknown fire in the Safety Research Coal
Mine (SRCM) at the Pittsburgh Mining Research Division
site and to determine fire location using carbon monoxide
(CO) arrival time and concentration [3-5].
An underground mine fire could cause changes in the
mine ventilation system by increasing airway resistance
by 10%–20% [6], which could lead to changes in airflow
quantities and airflow directions. Such changes could unex-
pectedly contaminate fresh air escape routes. Large-scale
test results in the literature show that a typical burning
of a wheel loader can generate a peak heat release rate of
approximately 20 MW, a CO concentration of 900 ppm,
and a smoke rollback of over 50 m in the mine entry that
could pose a severe risk to underground mine workers [7].
The ventilation network in metal/nonmetal mines tends
to have many airways and is complex with many branching-off
zones. This complexity is compounded by using ventilation
controls such as doors, regulators, booster fans, and auxiliary
fans to deliver fresh air to the working faces. Therefore, there
is a need to predict fire size and location within a mine to
determine the potential impact of a fire emergency. Such a
tool may provide mine operators the ability to evaluate and
improve mine ventilation networks to mitigate risk of haz-
ardous exposures to miners, and, to develop improved mine
rescue plans for emergency responders.
Successful forward modeling of an underground mine
fire scenario using mine fire simulation software requires