Enhancing Prediction of Blast-Induced Ground Vibrations through Machine Learning

Abstract

This Master’s thesis investigates the application of Machine Learning (ML) in predicting blast-induced ground vibrations in mining, with the aim of sur- passing the precision of the current industry-standard model that utilizes an empirical, regression-based method. The study applied a Deep Neural Network (DNN) model, selected for its capability to consider a broader range of variables than the industry-standard model, leading to significantly enhanced predictive capabilities. The evaluation of these models was conducted using three statisti- cal criteria: coefficient of correlation (R2), mean square error (MSE), and mean absolute error (MAE). The key finding is the DNN model’s superior performance, achieving an R2 of 0.94, an MSE of 0.94, and an MAE of 0.60, which represent a significant im- provement and reduction over the industry-standard model’s predictive results. Specifically, there is an 84% improvement in the R2 value, an 87% decrease in MSE, and a 71% decrease in MAE compared to the industry-standard model’s R2 of 0.51, MSE of 7.41, and MAE of 2.04. This marked enhancement in predictive accuracy illustrates the model’s ability to analyze multiple variables concur- rently and highlights the potential of AI and ML to improve environmental safety and operational efficiency in the mining industry.