5
number of used features in comparison with the fire size
prediction.
LIMITATIONS
The methodology presented in this paper can be applied
to any underground mine operation. However, the results
in this paper are only applicable to the example mine pre-
sented here. Furthermore, any update by the mine operator
in a ventilation network typically results in significantly dif-
ferent airflow distribution and will require the model to be
rebuilt. The model does not account for normal operating
airflow fluctuations. The results for fire size and location
accuracy are specific to the testing of the methodology from
hypothetical scenarios generated from machine learning
algorithms to evaluate model performance. Further inves-
tigation at a field site will be needed to test the predictive
model obtained from this study in collaboration with the
partner mine.
CONCLUSIONS
This research supports NIOSH’s efforts to develop work-
place solutions to improve the detection of hazardous con-
ditions in underground mines. A previously developed
technique by NISOH based on an ML-based modeling to
locate an unknown underground fire and determine its size
was successfully applied to a large ventilation network. We
achieved reasonably good model prediction performance to
locate the fire and its size with over 90% and 80% accuracy,
respectively.
The ML-based model only used a small fraction of air-
ways, 20 at most out of 766, to achieve over 90% fire loca-
tion accuracy.
The accuracy of the developed ML-based model
depends on the continuously calibrated ventilation net-
work. This method provides a useful tool for solving the
problem of unknown fire locations and reducing the risk of
hazardous conditions in underground mines.
DISCLAIMER
The findings and conclusions in this report are those of
the author(s) and do not necessarily represent the official
position of the National Institute for Occupational Safety
and Health, Centers for Disease Control and Prevention.
Mention of any company or product does not constitute an
endorsement by NIOSH.
REFERENCES
[1] Smith, A., &Litton, C. (2015), “The use of atmo-
spheric monitoring systems for fire detection in
underground coal mines,” 15th North American
Mine Ventilation Symposium, Blacksburg, VA,
20–25 June 2015.
[2] Slaughter, C., Tien, J., &Cao, Y. (2015), “Analysis
of all reported mine fire incidents: 2000–2013 in all
types of United States mines,” 15th North American
Mine Ventilation Symposium, Blacksburg, VA,
20–25 June 2015.
[3] Bahrami, D., Zhou, L., Xue, Y., &Yuan, L. (2021),
“Application of machine learning to determine
underground hazard location,” Proceedings of the
18th North American Mine Ventilation Symposium
(NAMVS 2021), Rapid City, SD, USA, 401–409.
doi.org/10.1201/9781003188476-41.
[4] Bahrami, D., Zhou, L., &Yuan, L. (2020), “Field
verification of an improved mine fire location model,”
Mining, Metallurgy &Exploration 38, 559–566. doi.
org/10.1007/s42461-020-00314-6.
[5] Zhou, L., Yuan, L., Thomas, R., Bahrami, D., &
Rowland, J. (2020), “An improved method to calcu-
late the heat release rate of a mine fire in underground
mines,” Mining, Metallurgy &Exploration, 37(6),
1941–1949. doi.org/10.1007/s42461-020-00276-9.
[6] Gillies, S., &Wu, H. W. (2004), “Case studies from
simulating mine fires in coal mines and their effects
on mine ventilation systems,” Proceedings of the Fifth
Australasian Coal Operators Conference. N. Aziz and B.
Kininmonth (Eds.), Australasian Institute of Mining
and Metallurgy, pp. 111–125.
Figure 3. Model performance for fire size on the test dataset
using seven important airways
number of used features in comparison with the fire size
prediction.
LIMITATIONS
The methodology presented in this paper can be applied
to any underground mine operation. However, the results
in this paper are only applicable to the example mine pre-
sented here. Furthermore, any update by the mine operator
in a ventilation network typically results in significantly dif-
ferent airflow distribution and will require the model to be
rebuilt. The model does not account for normal operating
airflow fluctuations. The results for fire size and location
accuracy are specific to the testing of the methodology from
hypothetical scenarios generated from machine learning
algorithms to evaluate model performance. Further inves-
tigation at a field site will be needed to test the predictive
model obtained from this study in collaboration with the
partner mine.
CONCLUSIONS
This research supports NIOSH’s efforts to develop work-
place solutions to improve the detection of hazardous con-
ditions in underground mines. A previously developed
technique by NISOH based on an ML-based modeling to
locate an unknown underground fire and determine its size
was successfully applied to a large ventilation network. We
achieved reasonably good model prediction performance to
locate the fire and its size with over 90% and 80% accuracy,
respectively.
The ML-based model only used a small fraction of air-
ways, 20 at most out of 766, to achieve over 90% fire loca-
tion accuracy.
The accuracy of the developed ML-based model
depends on the continuously calibrated ventilation net-
work. This method provides a useful tool for solving the
problem of unknown fire locations and reducing the risk of
hazardous conditions in underground mines.
DISCLAIMER
The findings and conclusions in this report are those of
the author(s) and do not necessarily represent the official
position of the National Institute for Occupational Safety
and Health, Centers for Disease Control and Prevention.
Mention of any company or product does not constitute an
endorsement by NIOSH.
REFERENCES
[1] Smith, A., &Litton, C. (2015), “The use of atmo-
spheric monitoring systems for fire detection in
underground coal mines,” 15th North American
Mine Ventilation Symposium, Blacksburg, VA,
20–25 June 2015.
[2] Slaughter, C., Tien, J., &Cao, Y. (2015), “Analysis
of all reported mine fire incidents: 2000–2013 in all
types of United States mines,” 15th North American
Mine Ventilation Symposium, Blacksburg, VA,
20–25 June 2015.
[3] Bahrami, D., Zhou, L., Xue, Y., &Yuan, L. (2021),
“Application of machine learning to determine
underground hazard location,” Proceedings of the
18th North American Mine Ventilation Symposium
(NAMVS 2021), Rapid City, SD, USA, 401–409.
doi.org/10.1201/9781003188476-41.
[4] Bahrami, D., Zhou, L., &Yuan, L. (2020), “Field
verification of an improved mine fire location model,”
Mining, Metallurgy &Exploration 38, 559–566. doi.
org/10.1007/s42461-020-00314-6.
[5] Zhou, L., Yuan, L., Thomas, R., Bahrami, D., &
Rowland, J. (2020), “An improved method to calcu-
late the heat release rate of a mine fire in underground
mines,” Mining, Metallurgy &Exploration, 37(6),
1941–1949. doi.org/10.1007/s42461-020-00276-9.
[6] Gillies, S., &Wu, H. W. (2004), “Case studies from
simulating mine fires in coal mines and their effects
on mine ventilation systems,” Proceedings of the Fifth
Australasian Coal Operators Conference. N. Aziz and B.
Kininmonth (Eds.), Australasian Institute of Mining
and Metallurgy, pp. 111–125.
Figure 3. Model performance for fire size on the test dataset
using seven important airways