Real‐time lithology identification while drilling based on drill cuttings image analysis with ensemble learning

Accurate lithology identification through geological exploration is crucial for hazard risk management during deep underground operations. Artificial intelligence has advanced in image recognition but using it to analyze underground drill cuttings for accurate lithology remains challenging. Issues include imprecise sampling control, harsh environments, and inconsistent image acquisition procedures, all leading to poor image quality. To address these issues, a lithology identification while drilling method was proposed. A cuttings sampling, testing, and transporting system was developed and deeply integrated with the drilling rig, achieving automation in cuttings sampling operations while standardizing the timing, procedures, and environment for sampling. A cuttings image preprocessing method was proposed, which meets the requirements of machine learning for image dimensions while enabling the automatic calculation of the proportions of different lithological particles. This is highly significant for accurately determining stratigraphic interfaces. An ensemble learning method was applied to enhance the identification accuracy. Underground trials were conducted at a coal mine in Huainan, China, involving the construction of four boreholes and the acquisition of more than a thousand cuttings images. During the trials, the system cooperated with the drilling rig to realize the accurate identification of lithology information during drilling, with an accuracy of 97.42% and an average processing time of less than 0.11 s per image. The results showed that the proposed lithology identification method can accurately obtain formation lithology in real time during drilling. This study guides drilling operations, ensuring target area coverage, effective hazard management, and supporting unmanned drilling technology development.