ALROSA’s Vilyuisk geological exploration expedition has completed a pilot project to use neural networks in diamond mining. Experts have proven that machine learning algorithms are capable of effectively solving applied problems when searching for kimberlite pipes. The work was carried out on a site in the Mirninsky district, where the famous Aikhal, Yubileynaya and Komsomolskaya deposits have already been found.
The territory was not chosen by chance. The Alakit-Markhinskoye field hides the potential for new discoveries, but the geological structure here is complex. Overburden and numerous faults make searching using traditional methods difficult. The neural network was supposed to help sort out this chaos.
For training artificial intelligence used the company’s archival data for half a century. The algorithms were provided with the results of ground and airborne geophysical surveys, descriptions of well sections, geophysical studies, as well as satellite images and digital elevation models.
Engineers tested several types of neural networks. The best results were shown by deep convolutional networks, which are usually used for image recognition. It was they who helped identify linear structures – faults and cracks in the earth’s crust.
ALROSA chief geologist Roman Zhelonkin explained: the analysis confirmed that almost all known pipes are located near the intersections of such structures. Identifying this pattern provides the key to finding new deposits.
Besides, machine learning successfully coped with the prediction of the thickness of the overburden and the classification of geological sections. Algorithms have learned to estimate the probability of the presence of kimberlite bodies at specific points.
Based on the results of the study, geologists determined the optimal methods for interpreting data in areas with a wide distribution of trap rocks. These formations arose hundreds of millions of years ago during magma outpourings and for a long time prevented accurate exploration.
Neural networks also helped to outline the most promising areas for drilling. In the summer of 2026, these forecasts will be tested in practice; geologists will lay wells at the points indicated by artificial intelligence. If the predictions are confirmed, ALROSA will receive a new tool that can speed up the search and reduce costs.
Source: ALROSA
