Citation

Barros-Daza MJ, Luxbacher KD, Lattimer BY, Hodges JL. Fire Technol. 2022; 58(3): 1545-1578.

Copyright

(Copyright © 2022, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s10694-022-01215-4

PMID

unavailable

Abstract

This paper presents a data-driven approach that can provide the most suitable decision to the mine firefighting personnel in real time during ongoing underground coal mine fires. The approach uses a feed-forward artificial neural network (ANN) to classify fires to provide the best decision considering only parameters measurable in underground coal mines. Additionally, the methodology along with the concepts that should be considered to elaborate a data-driven approach of this type are detailed. A total of 500 fire scenarios with different fire size, air velocity, fire growth rate, and entry dimensions were simulated in Fire Dynamics Simulator (FDS) and Fire and Smoke Simulator (FSSIM) for data generation to train and test the model.

RESULTS show that the ANN predicted fire classes with an accuracy and weighted-average F1-score equals to 97% and 96.7% for training and testing dataset, respectively.

RESULTS also show that 95% of ANN predictions of fire class change should not have a time gap greater than 18 s of the true fire class change for any fire position in the tunnel. Furthermore, the impact of fuel uncertainty during mine fires and how to address it is discussed in this paper. While the model presented in this work was designed to classify fires in a regular elongated coal mine entry, the same methodology could be applied to classify fires in other scenarios with similar geometry, such as road tunnels.

Language: en

Keywords

ANN neural network; Fire emergency response; Mine fire classification; Mine firefighters