Real-Time and Continuous Shale Shaker Monitoring Using Machine Learning, Computer Vision and Edge Computing

Abstract This paper introduces a machine learning-based system for continuous, real-time monitoring of shale shakers to support early detection of wellbore instability. By combining high-resolution camera feeds with edge computing, the system enables on-site image processing, significantly reducing bandwidth requirements and providing timely insights into shaker conditions. A Vision Transformer (ViT) architecture is used to classify images into three categories—Empty, Fluid, and Cuttings. Image preprocessing techniques, including masking and motion detection, enhance model performance by isolating relevant activity. The model was trained and validated on expert-labeled data from two wells and deployed on edge devices for real-time execution. It achieved 100% classification accuracy, with model predictions strongly correlating with rig activities, including detection of operational anomalies. This automated approach eliminates the limitations of manual inspections, offering a scalable and efficient tool to reduce non-productive time, improve safety, and optimize hole cleaning.