Mining companies are constantly collecting large amounts of data, but it can be challenging to turn this data into useful insights. That’s where we come in. Datarock was founded to help mining companies make better use of their data. Our productionised ML platform focuses on extracting valuable insights from a commonly underutilised geological dataset: core photographs.

Creating foundational geotechnical datasets, like geotechnical and lithological logs, usually involves geologists and engineers walking along the core and taking measurements and interpretations, which are then recorded on paper or entered into a database. Ideally, these individuals are well-trained, consistent, and working under a clear and consistent schema for the deposit. However, this ideal scenario is not always the case. Data is often collected by junior staff under pressure to reduce backlogs, leading to inconsistencies in the data. Additionally, personal bias, experience, and training can impact the logging style, which can further undermine the consistency and usefulness of the data. Other factors such as monotony, changes in management, changes in understanding of an orebody, and skilled labour shortages can also contribute to problems with data quality.

Datarock uses machine learning and computer vision to teach computers how to interpret core photographs. Expert annotations are used to train the models, so human expertise is still a critical part of the process. By using computer vision models instead of human observation, we can collect visual information from core in a consistent, repeatable way, reducing many of the issues mentioned above.

Our platform is easy to use. Simply upload your core photos to Datarock through our web portal or cloud repository, and our machine learning models will automatically analyse and detect different features within the image. These detected features are then transformed into logging style outputs that professionals can use.

Our methods are particularly well-suited for producing key geotechnical data, such as Rock Quality Designation (RQD) and Fracture Frequency/Spacing data, as well as data on rock condition and character. To date, Datarock has produced millions of  metres of traditional segment length RQD and fracture frequency.

Datarock geotechnical outputs based on a single drill hole of historic core photography. From left to right: Summary image made from cropped data, Median RGB value, Weighted Joint Density RQD, Segment length RQD, Measured RQD (by a geologist), Incoherent (the amount of broken rock) and FPM – fractures per metre.

Datarock is a tool to aid professionals to get more information out of their drill core, however please note we are not advocating that geologists stop looking at the core. Datarock outputs are numerical and quantitative and are ideal to be combined with other downhole datasets such as geochemistry, geophysics and detailed testwork to produce high value unified models of an ore body.

If you’re interested in learning more about how Datarock can help you build better foundational geotechnical datasets, please don’t hesitate to get in touch with us for a demonstration.