Prediction of Dust Dispersion from Drilling Operation in Surface Mines

Abstract

Dust pollution causes various problems within and outside the mine environment. Dust emanating from different activities directly affects the people working in the mines. Dust deposition on Heavy Earth Moving Machinery (HEMM) and other machinery can damage the machinery. The various activities involved in mining to extract ore from earth lead to dust pollution. Especially, Particulates Matters (PM) present in mines area lead to various human respiratory diseases. Aerodynamic diameter of particles having less than 10µm called as PM10. Among all activities involved in mining, drilling activity is more important and it produces PM particles. Dust prediction models are necessary to identify the quantity of dust expected from drilling so that dust control strategies can be taken up at mine site. In order to develop dust prediction models in surface mines, field investigations were carried out in eight opencast mines. Among them, three are opencast coal mines, two are limestone mines and the remaining are granite quarries. Two opencast mines, two granite quarries and one limestone mine data was used to develop mathematical models. One coal mine data, one granite quarry data and one limestone mine data was used to validate developed models. To develop dust prediction models, 169 sets of data for emission model and 184 sets of data for concentration model from different rock formations were considered. Field monitoring was carried out according to Central Pollution Control Board (CPCB) standards. Rock samples were collected from different locations of mines and brought to the laboratory for determining required physico-mechanical properties according to International Society for Rock Mechanics (ISRM) suggested methods. Various rock properties considered are Moisture content, Density, Compressive strength and Schmidt rebound hardness number. Artificial Neural Network (ANN) analysis was carried out for different combinations of hidden layers. Feed Forward Neural Network with back–propagation algorithm was used to train the network. Four types of algorithms were used for development of models andii their performances were evaluated using Values Account For (VAF), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Network was trained using different types of Back-propagation algorithms such as Trainrp, Trainscg, Traincgp, Trainlm. The algorithm ‘Trainlm’ has high MAPE and less RMSE. Value of RMSE is 6.68, MAPE value is 33 and VAF value is 79.90. Trainlm algorithm was found to be the best method for prediction of PM10 from drilling operation and was used for comparison. The predicted values from ANN method and field measured values were compared. The R2 value for emission model is 0.81 and for concentration model it is 0.80, which shows very good correlation and gave better forecasting results using ANN method. Analysis showed that the field data is error free. But, ANN cannot give mathematical equations, so multi regression analysis was used for the development of models. Multiple regression analysis method was used to determine the relation between multiple independent variables (input) and single dependent variable (output). Mathematical equations were developed using statistical software, namely Statistical Package for the Social Sciences (SPSS). In order to assess the influence of input parameters on output, stepwise regression was used. Assessment of SPSS software based predicted values were evaluated by statistical parameters like coefficient of determination (R2), ANOVA, parameters coefficients and Variable Influencing Factor (VIF). The parameters chosen were found to be statistically more significant. The predicted values from multiple regression method and field measured values were compared. The R2 value for emission model is 0.82 and for concentration model it is 0.81, for 95% level of confidence, which shows very good correlation. A comparison was made between Multiple Regression Analysis Model and ANN model results. ‘Trainlm’ algorithm revealed that, MRA model gave better performance than ANN with lower RMSE and high MAPE values and higher prediction accuracy (VAF) value for all the predicted variables. The VAF values obtained for MRA is 87.1 per cent,iii RMSE is 3.22 and MAPE is 33.7 per cent. Finally, to validate developed models, field measured values were compared with SPSS model predicted values and USEPA predicted values. Analysis revealed that USEPA was giving around 99 per cent error and SPSS model was giving error of within 20 per cent. Therefore, SPSS models developed as part of this research work may be used for dust prediction from drilling activity under Indian Geo-Mining and weather conditions.