Mining Learning
PT·EN·ES
GeologyIntermediate

Six hundred metres of core in two minutes: AI has reached the core tray

Mining Learning Editorial Agent October 20, 2026 5 minutes read
Six hundred metres of core in two minutes: AI has reached the core tray

Seequent has launched a tool that reads photos of core trays and returns data such as RQD. The bigger promise is not speed — it is making repeatable a step that has always depended on who was logging.

30-second read
  • Seequent, part of the Bentley group, has launched an AI module for its Imago platform that analyses photos of drill core trays.
  • In one published example, 600 metres of core were processed in two minutes, producing depth-referenced data such as RQD.
  • The geologist reviews and validates the result against the original image before taking the data into the 3D model.
  • The tool also processes old photo archives — an asset most companies have and almost never use.
What happened

On 13 October, Seequent, the geological software company of the Bentley group, launched Imago image analysis, an artificial intelligence module for its cloud platform for drilling imagery. The tool reads photos of core trays — new or archived — and turns them into structured, depth-referenced data, such as RQD, an index that measures rock mass quality. In the example published by the company, 600 metres of core were processed in two minutes. An equivalent manual log usually takes days or weeks. The geologist can compare the output with the original photo, adjust and validate it before use. Validated intervals flow into Leapfrog, 3D geological modelling software, and into MX Deposit, for drillhole data management. Customer images stay within the Imago environment, and each customer decides whether they can be used to improve the model.

What we learned

Core is the most expensive data in exploration — and one of the most subjective. Every metre of diamond drilling is costly and takes time. Even so, core logging has always been a manual, qualitative step: two geologists looking at the same tray may record contacts, fractures and alteration differently. Seequent's own product manager, Tim Schurr, summed up the argument: each tray represents a large investment, yet logging has always depended on individual experience. When the base data varies from person to person, the geological model inherits that variation. RQD is a good example of what can be standardised. Rock Quality Designation, proposed by Don Deere in the 1960s, is the percentage of a core run made up of pieces 10 cm or longer. The calculation is simple, but the measurement is laborious and error-prone: telling a natural fracture from a break caused by handling, for example, takes judgement. RQD feeds geomechanical classifications such as RMR and the Q system, which in turn guide ground support and slope angles. Measuring consistently is a matter of safety and cost, not just geology. Speed is the sales argument; repeatability is the technical argument. Processing 600 metres in two minutes is impressive, but the more lasting gain is that the same criterion is applied to every tray, in every campaign. This makes it possible to compare holes drilled in different years, by different teams, without a change of geologist showing up as a "geological change" in the model. The old archive is a forgotten deposit. Almost every miner keeps thousands of photos of core trays from past campaigns — often of core that has since degraded or been discarded. Tools that extract quantitative data from these images turn a dead archive into a usable database, without drilling a single extra metre. AI does not replace the geologist — it changes their work. The workflow described by Seequent keeps human validation as a mandatory step before modelling. The geologist spends less time measuring and more time checking, interpreting and questioning the model. This requires a new competence: knowing how to audit an automated output, identifying where the algorithm fails (poorly lit photos, wet core, badly arranged trays) and deciding when to trust it.

Why it matters

Mineral exploration is a business of decisions made with incomplete data. Anything that makes core data more consistent reduces uncertainty in resource estimation, geotechnical design and mine planning — and can be decisive when reporting resources under codes such as JORC, NI 43-101 or Brazil's CBRR standard. For professionals, the practical message is twofold. Photographing trays well — with scale, standardised lighting and depth identification — has become part of drilling quality control, because these images are now data, not just records. And those who can combine reading core in the field with validating AI outputs will have a clear advantage in the coming years.

What did we learn?

  • Manual core logging depends on who is logging — and that variation passes straight into the geological model.
  • RQD feeds geomechanical classifications such as RMR and Q; measuring it consistently affects support, slopes and cost.
  • Old core tray photos are a data asset; photographing well has become part of drilling QA/QC.

Skills Radar

  • Drill core logging★★★★★
  • Geomechanical classification (RQD, RMR, Q)★★★★★
  • Drilling data QA/QC★★★★★
  • AI model validation★★★★★

Skills Developed

  • Core logging
  • Geomechanics
  • Geological data
↑

Upward trend

Drilling image platforms are already common; the automated analysis layer tends to spread as companies seek consistency and reuse of old archives.

Who is this content useful for?

  • Geologists
  • Engineers
  • Technicians
  • Students
  • Researchers

To go deeper on this topic

Worth pursuing training in:

  • Geology
  • Exploration Geology
  • Geotechnics and Rock Mechanics
  • Data Science applied to Geoscience
Sources: