Automated Computer Vision System for Real-Time Detection of Drilled Cuttings and Cavings

Abstract This paper describes the role of a computer vision system in assessing wellbore stability. It validates the application of computer vision to detect the presence of cavings at the shale shaker using a remotely monitored camera system. The main objective was the early detection of developing wellbore pressure and stability issues during drilling to avoid non-productive time (NPT), a potential boon for the industry. Two cameras with onboard computer vision software were installed in the hazardous zone above the shale shakers to monitor the returning drilling cuttings and quantify their sizes. Images from the cameras were analyzed onsite and cutting size and volume data were delivered to the drilling team in real-time. The term Unidentified Falling Object (UFO) was introduced to account for the current state of technology and the difficulty inherent in discriminating between cavings and cuttings at the shakers. Detecting an anomalously large object is considered a potential indication of a caving. Five wells in the Norwegian Offshore Sector were consecutively monitored over eight months across hole sizes ranging from 16½-inches to 8½-inches. The project was managed remotely via an interactive, multi- platform, open-source analytics and visualization application. We include a description of the project's implementation of new technology on an active drilling rig. The computer vision model was validated by dropping known test rock samples into the cuttings flow, measuring cutting size distribution (CSD) on cuttings samples, and weighing drill cuttings samples. The results were used to calibrate a computer vision model that accurately measured a range of cutting sizes from 10mm to 45mm. The data showed a credible correlation to changes in wellbore formation, and cavings were successfully detected. The model improved throughout the project, alerting the drilling team to suspected issues in near-real-time. Other valuable observations included cuttings overflow, vibration performance, and the potential for time savings during hole cleaning activities. This demonstrated the versatility of the system and its potential to provide a more comprehensive understanding of the drilling process. Wellbore instability is one of the largest contributors to drilling NPT. Early detection through the identification of cavings at the shakers is an established process but is currently manpower-intensive, intermittent, and subjective. Applying a computer vision model to monitor and analyze drill cuttings in real- time continuously—autonomously characterizing the drilled cuttings’ size, shape, and volume—will help optimize overall penetration rates and significantly improve hole-cleaning practices. This will reduce lengthy circulation times and pack-off events, enhancing the efficiency of drilling operations.