Machine learning tool to minimise and predict airblast during blasting and to optimise the design of blasting operations

We present a method to minimise and predict airblast in blasting operations in an open-pit Debswana diamond mine. Blast engineers can use this tool to optimise their blast design to achieve desired blasting operation effect, i.e., airblast. The major novelty of this study is on the creation of a nine-dimensional solution space, optimisation of the blast design parameters, and minimisation of airblast using gradient descent method. We develop a solution surface using artificial neural network (ANN). This is our best-performing machine learning model compared to the three other models used, namely, support vector machine (SVM), k-nearest neighbour (k-NN), and random forest (RF). The computed nine-dimensional solution space has eight input parameters: stemming, distance from the blast face to the monitoring point, burden, powder factor, hole diameter, maximum charge per delay, spacing, and hole depth. Sensitivity analysis revealed that stemming is the most sensitive input parameter while spacing is the least sensitive. The minimum value of airblast computed in this study through unconstrained optimisation is around 40 dB, which is approximately equivalent to the sound of a whisper. This framework is adaptable to various geological and operational settings, highlighting its broader applicability in improving environmental compliance and blasting efficiency.