Blast-induced ground vibration and air overpressure prediction using artificial intelligence techniques

Abstract

Blast-induced ground vibration and air overpressure are considered as the most important environmental hazards of mining and can damage structures such as buildings, dams and pit slopes. Review of previous studies has shown that some empirical and Artificial Intelligence (AI) models have been proposed to estimate blast-induced ground vibrations and air overpressure. Notable AI techniques applied in prediction of blast-induced ground vibration include Backpropagation Neural Network (BPNN), Radial Basis Function Neural Network (RBFNN), Generalised Regression Neural Network (GRNN), Group Method of Data Handling (GMDH), Support Vector Machine (SVM) and Extreme Learning Machine (ELM). In this study, five techniques, namely Wavelet Neural Network (WNN), Multivariate Adaptive Regression Splines (MARS), Least Square Support Vector Machine (LS-SVM), Relevance Vector Machines (RVM) and Gaussian Process Regression (GPR) are proposed to predict blast-induced ground vibration using 210 blast data sets from Ghana Manganese Company Limited (GMC). Out of the data sets, 130 blast data sets were used to train the models and the remaining 80 to test the developed models. For comparison purpose and ascertaining the suitability of the proposed methods, four empirical techniques (United State Bureau of Mines, Langefors and Kilhstrom, Ambrasey-Hendron and Indian Standard) were also employed. With regards to the air overpressure, BPNN, GMDH, GPR and SVM are the only AI methods applied and captured in the literature. This study therefore tested the capability and applicability of some AI methods, namely RBFNN, GRNN, LS-SVM, RVM, ELM, WNN and MARS, that are yet to be explored in the prediction of air overpressure. To accomplish this task, air overpressure data sets from Newmont Golden Ridge Limited, Akyem Mine was used. In all, 98 data sets were used for the model construction and 73 data sets were used to independently assess the performance of the models formulated. Two empirical predictors, the general predictor model and ‘Newmont Model’, were utilised for the purpose of comparison. In the blast-induced ground vibration interpretations, the statistical results revealed that, four out of the five newly tested AI techniques (LS-SVM, WNN, MARS and GPR) could produce good ground vibration predictions comparing to the AI benchmark methods of BPNN, RNFNN and GRNN. Hence, LS-SVM, WNN, MARS and GPR have been proposed to be used as suitable alternative tools to predict blast-induced ground vibration. In comparing all methods applied, the proposed LS-SVM was the most accurate on the basis of the statistical analyses carried out in this study and thus the selected model for predicting blast-induced ground vibration at Ghana Manganese Company Limited. In the air overpressure prediction interpretations, it was found that four out of the seven methods (GRNN, RBFNN, RVM, and MARS) tested produced comparable and satisfactory results as the widely used BPNN, GPR and SVM and thus could serve as suitable alternatives to the prediction of air overpressure. However, it was found on the basis of the statistical analyses carried out that, the BPNN was the selected model for the prediction of air overpressure for Newmont Golden Ridge Limited, Akyem Mine. The overall analyses of the study showed that the AI techniques are superior in predicting both blast-induced ground vibration and air overpressure to the empirical predictors usually employed in most mining and civil engineering industries. To this end, a user-friendly AI-based software package was developed on the MATLAB platform and can be used in the industry for prior prediction of the blast-induced ground vibration and air overpressure based on the blast design in the mining industry.