DeepMines: A fog Enabled Prediction Platform for Underground Coal Mines

Sunny Sanyal, Animesh Chattopadhyay

Abstract: The underground mining industry observes enormous losses in terms of human lives and infrastructure every year due to fatal fire hazards and blasts caused due to methane accumulation. When the methane levels are high, the methane monitoring systems deployed inside the mines don't provide sufficient time to remove the accumulated methane gas and thus remains only one option of halting the work and evacuation of the workers. Therefore, the mining industry suffers a loss of productivity due to frequent evacuations, power terminations, and false alarms caused by conventional monitoring systems. This paper advocates an alternate paradigm of forecasting rather than detection, which ensures that the system gets sufficient time to take necessary measures to remove the accumulated methane gas without completely halting the work. The paper presents a fog computing enabled ioT data aggregation and accident prediction framework for high-stress underground mining scenarios that predict fatal accidents due to high methane accumulation using Deep LSTM encoder-decoder architecture. The experimental results show that the proposed solution can classify accidental scenarios at an accuracy of 94.23 percent along with a satisfactory long-duration future time series prediction.