Simulation of blast-induced ground vibrations using a machine learning-assisted mechanical framework

Blasting, as a cost-effective method of rock breakage, assures a size distribution for efficient material handling in surface and underground mining. An immense level of energy released from explosive charge in blast holes induces plastic deformations in competent rocks, while the remaining part transforms into destructive ground vibrations. Seismic wave velocity is a major concern for blasting to mitigate possible human discomfort and structural damage. The common approach is to characterize the wave propagation within geological units by monitoring trial blasts depending on either conventional or advanced statistical methods. The scaled distance method and the novel soft computing techniques are commonly used for the prediction of peak particle velocity. This study explores the limitations of conventional statistics and the widely accepted parameters controlling blast-induced ground vibrations. The velocity vector direction was analyzed to reveal the effects of geological, geomechanical, and structural features. Discrepancies of the scaled distance method in predicting the peak particle velocity due to complex geology and the need for an extensive trial blast database motivate the research for an alternative approach. The proposed method takes advantage of machine learning for the calibration of the mechanical model and simulates ground vibrations by a mass–spring–damper system. Blast records from an aggregate quarry and a metal mine were used for validation purposes. The proposed method offers remarkable improvements in terms of accuracy by reducing prediction errors down to 0.04 mm/s. Regarding the structural condition of rock mass, the average error is around 4% for relatively massive rock mass and 20% for complex structural geological conditions.