Cuttings Return Analysis by Visual Imaging: Novel Technology to Support Drilling Automation

Abstract

AbstractIn recent years, image acquisition of process variables has become a common practice as an essential monitoring tool for industrial systems. The rise of data analysis models and high performance computers enables the development of systems to support real time decision and automation initiatives. This work aimed to develop a comprehensive methodology to acquire, process and interpret shale shaker video images to support detection of problems associated with poor hole cleaning and wellbore instability. To achieve this goal, experiments in different operational conditions were performed in a pilot-scale vibrating screen using a sand suspension with xanthan gum at 0.1% to emulate the slurry flow in actual sieves. Such experiments aimed to evaluate how some of the main operational variables influence the area of the sieve filled with solids, the moisture content of the solids, and their flow velocity. The proposed methodology used a U-Net convolutional neural network to perform the semantic segmentation of the sieve’s images to estimate the percentual area filled with solids in the shale shaker. An image database with 11,140 pictures and their respective templates was created from 26 experiments. The templates were built using image processing techniques, and 75% of them were used for training, 10% for validation, and 15% for testing. The neural network evaluation metrics were accuracy for training, F1-Score, and MeanIoU for testing. In addition, the estimated values were compared to the experimental data. The estimated velocity, expressed in cm/s, was further compared to experimental data. The results obtained using the U-Net showed high segmentation ability, with an average accuracy of 97%, mean F1-Score of 92%, and mean IoU of 91%. According to these results, it can be concluded that the proposed image-based technique is a promising tool for monitoring important process variables in drilling operations. Additional experiments with different shape particles result in effective guidelines to detect wellbore instability events. In future work, the authors expect to improve the segmentation technique to obtain better estimations of the process variables and the implementation of the procedure in integrated real time diagnosis systems. This work shows an innovative approach to support the drilling process, allowing to evaluate, quantify and characterize the drilling cuttings that returns to the surface. This initiative is committed to helping to reduce the well construction costs, non-productive times (NPTs) and to guarantee drilling operations more efficient and safer. Besides, provides additional relevant information to support the future views of autonomous drilling and unmanned rigs.