Presented on 20 May 2026: Session 15 A strong geological understanding is fundamental to most petroleum operations, including exploration, field development, completion selection, well design, and safe and efficient drilling. However, geological information from cuttings is often overlooked, described only in passing, and rarely digitised for use into today’s digital subsurface workflows. Historically, cuttings have been difficult to digitise, with descriptions that are typically qualitative and vary in detail. Little more than lithology information is captured in subsurface software. Artificial intelligence (AI) machine vision revolutionises this process by enabling the acquisition of quantitative visual data on a grain-by-grain basis. These data are statistically filtered or combined to capture details like colour, shape, and size, generating multiple new geological data curves. By using high-resolution images taken under consistent lighting and with a consistent white balance, the cuttings are analysed with purpose-trained, cutting-edge AI, cuttings can then be fully integrated and treated as an interactive digital-native dataset. The following workflow is applied: (1) Segment Anything Model 2: developed by Meta, this model splits a cuttings sample into thousands of individual grains and rock chips; (2) machine vision techniques: using polygon-based analysis, statistical methods, and other advanced machine vision techniques, cuttings can be sorted and categorised for plotting alongside logs. Multiple use cases are explored, including the ability to determine grain size for completion sand screen selection and the ability to trace borehole breakout to its source formation, leading to better and more efficient well designs. Geological cuttings commonly sit unused in archives. AI now offers a way to bring this data back from archives, integrate it into industry standard tools, and use it to improve operational outcomes. To access the Oral Presentation click ‘Supplementary data’ below. To read the full paper click here
Alex Fuerst (Thu,) studied this question.