Rapid prediction of mine tunnel fire smoke movement with machine learning and supercomputing techniques

A method and a thorough demonstration combining supercomputing and machine learning (ML) were proposed to help quickly predict the spread of fire smoke in mine tunnels. 1000 cases of a three-dimensional mine tunnel fire problem under different ventilation, thermal and geometric conditions were numerically simulated using supercomputing. Then four ML models were trained and applied to predict fire dynamics in the tunnel. It is revealed that all ML models performed well in predicting the occurrence of backflow and the back-layering length of the smoke. In particular, the random forest (RF) and support vector machine (SVM) models have the best performance for predicting whether backflow of fire smoke will occur, while the artificial neural networks (ANN) model shows the best performance in predicting the back-layering length. In addition, the ML models were used to evaluate the factors that affect the fire dynamics in the tunnel. The results show that the ventilation velocity and tunnel inclination angle are the most critical factors under the investigated ranges of ventilation, thermal and geometric conditions. Owing to the high performance in numerical simulations and prediction, the proposed method combining supercomputing and ML may provide a novel and efficient way for rapid prediction of mine tunnel fire dynamics.