The problem of finding the balance between the performance of blasting and the milling process is always inherent in the mining industry. More than 60 percent of the total amount of electricity used for comminution comprises the hard-rock mining process, while up to 40 percent of the power consumption at the site is accounted for by grinding circuits. The blasting process is the first and most efficient step of the comminution process. Conventional methods of mine-to-mill design take into consideration only the macro-size distribution of particles; however, recent investigations provide information about two important discoveries: the micro-fracturing during blasting and spatial fragment assessment using deep learning.
Fragmentation caused by explosives includes macroscopic fragment production as well as microscopic pre-conditioning of blasted rock. Analysis using optical methods and Bond Work Index tests proves that blasting results in formation of a dense system of micro-cracks inside ore particles, primarily between mineral grains such as plagioclase phenocrystals. It was found that smaller fragments of rock subjected to blasting contain more cracks, thus reducing their specific Work Index and making them softer for grinding. Field experience showed that 33%-42% increment in powder factor of the explosive charge allowed reducing energy consumption of SAG mills by 29% while increasing capacity of the mill by 10%-16%.
Real-world industrial applications confirm the economic effect of a combined approach to fragmentation management. For example, at the Newmont Ahafo mine, a re-engineering of blast design included an enlargement of boreholes’ diameter from 140 mm to 165 mm, powder factors increment from 0.36 kg/t to 0.53 kg/t, and use of high-velocity dense prill explosive compositions. This led to an increment in fine fraction of -14 mm from 17.7% to 23.9%. As a result, throughput of the primary SAG mill increased by 30%, throughput of the plant by 8.4%, and specific energy consumption of grinding process decreased by 20% (from 10.8 to 8.6 kWh/t). In more than 30 mining facilities all around the globe, domain-based blasting simulation models provide a 5%-20% increment in concentrator throughput without any capital expenses.
Muckpile fragmentation assessment should be accurately conducted to retain these benefits. Traditional digital photoanalyzing systems, such as FragScan, WipFrag, and Split, require manual correction from operators along with calibration based on sieving procedures, which require several hours after each blast. Automated workflows incorporate highly-trained deep learning instance segmentation models, for example, YOLO12l-seg, capable of separating irregular fragment boundaries with a frame rate of 15 frames per second and a mask mean average precision at IOU=0.5 equal to 0.800. Masks generated by this model are used as inputs for spatial statistics calculations. PCA estimates anisotropy of fragments ejecting, KDE localizes areas with high concentration of fragments, and Delaunay triangulation finds distances between neighbouring fragments.
The integration of mine to mill optimization makes the process of blast design not just a one-off cost component but rather a strategy for increasing the throughput capacity of the plant. The phenomenon of micro fracturing due to blasting is known to greatly reduce the hardness of the material in the subsequent stages, whereas the process of deep learning segmentation provides real-time statistical feedback about the spatial data on site.
