Abstract The presented technology is part of a broader AI-driven program focused on automating geological and petrophysical interpretation from primary rock samples images. Our solution extends this paradigm to drill cuttings —an underutilized yet abundant data source. We introduce a modular system of three ML-powered components—Image QC, Annotation QC, and Multimodal Lithology Classification. This end-to-end pipeline transforms raw, heterogeneous cuttings imagery into standardized, high-confidence training-ready datasets. By combining image-based features with well data, the system enables consistent, scalable, and cost-effective prediction of lithology. Trained models are fully integrated into a user-friendly application, enabling geologists to interact with the system intuitively and perform real-time inference without coding. Ultimately, the approach increases data utilization, reduces expert time, and improves the accuracy and reproducibility of subsurface interpretation.
Mezghani et al. (Mon,) studied this question.