No abstract

No abstract

Air blast is considered to be one of the most hazardous environmental disturbances created by blasting operation. Prediction of air overpressure (AOP) generated owing to blasting is difficult due to the influence of several factors in the air wave transmission. Blast design parameters, wind direction and speed, atmospheric temperature, humidity and topography, etc. are all affecting AOP. In this paper, an attempt has been made to predict AOP using artificial neural network (ANN) by incorporating the most influential parameters like maximum charge weight per delay, depth of burial of charge, total charge fired in a round and distance of measurement. To investigate the effectiveness of this approach, the predicted values of AOP by ANN were compared with those predicted by generalised equation incorporating maximum charge weight per delay and distance of measurement. Air overpressure data sets obtained from four different mines in India were used for the neural network as well as to form generalised equation. The network was trained by 70 data sets and validated with 25 data sets. The network and generalised predictor equations were tested with 15 AOP data sets obtained from another two mines. The results obtained from neural network analysis showed that the depth of burial of the charges and maximum charge weight per delay were among the blast designed parameters that have most influence on AOP. Based on the ANN result, depth of burial of charge has more relative sensitivity and weight than the maximum charge weight per delay. The average percentage of prediction error for ANN was 2?05, whereas for generalised equation, it was 5?97. The relationship between measured and the predicted values of AOP was found to be more logical in the case of ANN (correlation coefficient: 0?931) than that of generalised equation (correlation coefficient: 0?867).

No abstract

Ground vibration induced by rock blasting is an unavoidable effect that may generate severe damages to structures and living communities. Peak particle velocity (PPV) is the key predictor for ground vibration. This study aims to develop a model to predict PPV in opencast mines. Two machine-learning techniques, including multivariate adaptive regression splines (MARS) and classification and regression tree (CART), which are easy to implement by field engineers, were investigated. The models were developed using a record of 1001 real blast-induced ground vibrations, with ten (10) corresponding blasting parameters from 34 opencast mines/quarries from India and Benin. The suitability of one technique over the other was tested by comparing the outcomes with the support vector regression (SVR) algorithm, multiple linear regression, and different empirical predictors using a Taylor diagram. The results showed that the MARS model outperformed other models in this study with lower error (RMSE = 0.227) and R2 of 0.951, followed by SVR (R2 = 0.87), CART (R2 = 0.74) and empirical predictors. Based on the large-scale cases and input variables involved, the developed models should lead to better representative models of high generalization ability. The proposed MARS model can easily be implemented by field engineers for the prediction of blasting vibration with reasonable accuracy.