Application of Machine Learning in rock mass characterization for Blasting

In recent years, implementation of Artificial Intelligence and Machine Learning model technology has gained ground in almost every aspect of human activities. The mining industry has not been foreign to these applications, with advantages including the creation of new optimization methods and the elimination of repetitive tasks. Those applications have been implemented in multiple operative areas, from preventive conveyor belt maintenance to energy optimization for minimizing environmental impact. The blasting process aims to use the energy liberated by the detonation of explosives to break the rock mass according to specifications that facilitate the comminution process. To optimize this process, the energy delivered must correspond to the necessary amount to exceed the tensional and compressional resistance of the rock mass. If the energy applied is insufficient, the rock will not reach the desired fragmentation level and the processing plant may incur additional costs. On the other hand, if the energy level required is exceeded, we incur increased blasting costs that correspond to excess explosive used. Optimization of this aspect of the mining process has major impact in the processing chain reflected in lower costs, efficient resources consumption and environmental impact. The Design for Outcome technology (DfO), developed by Orica, is used to characterize the rock mass in detail to classify, borehole by borehole, the rock mass relative hardness to deliver the adequate energy level to achieve optimal fragmentation. This technology uses information collected by drill sensors, producing data typically referred to as Measure While Drilling (MWD) data to classify the rock mass into domains that can later be used to apply optimal blast designs. In DfO, the data flow is fully automated, starting with the data produced by the drilling system and uploaded to the cloud, which is then domained using Machine Learning (ML) techniques, and then published for consumption using a web interface. Automated explosives loading rules are then generated for each rock mass domain, ensuring that for each borehole, the adequate explosive energy is delivered to achieve the desired fragmentation results with unprecedented accuracy. The application of machine learning techniques allows us to optimize the blasting process results by leveraging a deeper understanding of the rock mass. Applying these optimization techniques, we obtain considerable economic benefits, decrease resources consumption, and achieve a more sustainable mining operation. The application of DfO in a Chilean mining operation allowed the characterization of their rock mass into four domains and established specific explosive loading rules for each one. A 60kUSD reduction in explosives cost was achieved during the test period, while maintaining the desired fragmentation results and expected dig rate. These outcomes were the result of a comparison between the blasting costs used applying the technology mentioned above and a “traditional” classification of the rock mass with the corresponding blasting practices.