Russian scientists patent AI solution for rock mechanics

Experts at Innopolis University have developed an artificial intelligence (AI)-based software solution that automatically identifies cracks, faults, veins, breccias (sedimentary rocks), and other geological structures in photographs of core samples. Core samples are rock specimens extracted during exploratory drilling. The new method will increase the accuracy of core analysis tenfold and speed up the creation of geological models of mineral deposits. This is reported on the website of Innopolis University, a partner of TV BRICS.

As noted by Arseny Pinigin, head of the artificial intelligence technologies department at the Centre for Oil and Gas Technology, core samples are an important source of information about the Earth’s subsurface.

"Traditional manual core documentation is extremely labour-intensive, time-consuming and often subjective. Existing software solutions are either not universal or require constant supervision by a specialist. Our method solves the problem using artificial intelligence," the specialist explained.

University staff have developed a two-stage system for processing core images based on a neural network. First, images of core boxes – metre-long sections – are analysed by a transformer-based neural network, which automatically identifies the metre-long sections and precisely maps them to depths. Each section is then segmented using an AI model trained on a large dataset of images.

"For each core section, the AI generates a digital fingerprint – a database of all identified structures with their characteristics and depth coordinates – comprising 2,780 numerical values per image. These include texture, colour, contrast, the presence of fractures and other features extracted by the neural network. The algorithm clusters multidimensional feature vectors, which is particularly effective for identifying complex faults, tectonic breccias and other anomalous structures that affect the stability of wells and quarries," explained Ilmir Nugmanov, Deputy Director of the Centre for Oil and Gas Technology at Innopolis University.

The creators of the AI solution note that in 7 out of 10 cases, the system classifies core photographs in the same way as an experienced geologist. In the future, the developers plan to improve the accuracy of the method.

The development is expected to be useful in the study of core material in the mining industry, in the search for solid minerals, and in construction, where rapid and objective analysis of the structural characteristics of rocks is required.