Preventing Wellbore Collapse with AI-Powered Cuttings Monitoring: A Field-Validated Approach to Enhance Drilling Decision-Making

Abstract This paper aims to introduce a novel technique for real-time monitoring of wellbore stability by automating the analysis of drilled cuttings from shale shakers. The objectives are to develop a vision-based system to classify cuttings by size distribution, also correlate cuttings data with logging parameters to detect wellbore collapse, and provide actionable insights to mitigate drilling risks. The scope covers field implementation, data validation, and comparative analysis with conventional methods. The proposed system employs high-resolution industrial cameras mounted at optimal angles on shale shakers to capture real-time images of cuttings. Advanced computer vision algorithms, including convolutional neural networks (CNN), segment and classify cuttings into three size categories: fines (10mm). Processed data is integrated with real-time drilling parameters (ROP, torque, ECD) through a cloud-based platform, where machine learning models analyze correlations between cuttings patterns and wellbore conditions. A dynamic thresholding system flags abnormal cuttings distributions, triggering automated alerts when large fragment concentrations exceed preset safety limits (typically >18% of total cuttings). Field validation involved comparative testing against manual sieve analysis across 6 wells, with continuous calibration to account for lithology variations. The entire process operates within a 5 second latency from image capture to diagnostic output, enabling near-instantaneous wellbore stability assessment.