Automated Borehole Image Log Interpretation- Characterizing Geological Facies

Abstract The interpretation of geological facies from borehole resistivity images and well logs is foundational for robust reservoir characterization and decision-making in hydrocarbon exploration and production. However, traditional manual interpretation methods are inherently labor-intensive, subjective, and prone to inconsistencies, especially in complex geological settings. This study presents a novel, integrated machine learning (ML) workflow for the automated typing of geological facies, incorporating litho-unit prediction, bedding tracing, and a multi-level facies description system that closely mirrors the analytical approach of subject matter experts (SMEs). The methodology leverages conventional openhole logs (gamma ray, density, neutron and resistivity) to predict sand, silt, and clay fractions using the CatBoost algorithm, forming the basis for litho-unit classification. Borehole resistivity images are processed using a U-Net convolutional neural network (CNN) to segment and automate trace of geological planar features, with further refinement via sinewave matching. Geological facies typing is achieved through a hierarchical approach employing Vision Transformers (ViTs) of CNNs, which recognize and classify depositional structures in borehole images. This multi-step workflow ensures that each interpretative task is both automated and explainable, aligning with established SME practices. Results demonstrate that litho-unit predictions are highly consistent with SME assessments across a range of wells and formations, with mean deviations in most cases below 0.01. The electro-image facies classification system effectively distinguishes diverse facies, providing granular and repeatable subsurface interpretations. Notably, the workflow's modular design enables adaptability to new facies schemes and additional image sources, such as logging-while-drilling (LWD) and core images. By integrating advanced ML models with SME expertise, the workflow reduces manual workload, enhances the consistency and objectivity of subsurface geological description, and provides explainable predictions that foster user trust and adoption. This work marks a significant advancement in automated borehole image data interpretation and offers a scalable foundation for future development of digital tools supporting consistent, data-driven reservoir characterization.