Abstract:
Multiple supervised and unsupervised artificial intelligence techniques have been adapted and applied for real-time drilling monitoring and optimization purposes. This chapter describes these methods and provides examples and case studies of how they are being applied to improve the efficiency and sustainability of drilling techniques. Data preprocessing, feature selection/importance, statistical and correlation methods, prediction reliability, multi-K-fold analysis, and annotated-confusion-matrix techniques all contribute to improved interpretation of these AI methods, particularly machine, and deep learning. Artificial intelligence (AI) is at the forefront of providing real-time monitoring and predictions required by digital-twin and automated drilling strategies. There are many supervised and unsupervised AI methods used with real-time drilling data inputs for various monitoring and prediction purposes. This chapter describes the various machine and deep learning supervised techniques and the unsupervised clustering and discriminant analysis methods providing examples of how they are applied to solve drilling-related issues. Data screening and performance assessment techniques are identified together with statistical and correlation analysis required to establish relationships among the variables involved. AI enhancement techniques, including multi-K-fold cross-validation, reliability modeling, feature selection, and importance analysis, refine the performance of the AI methods and establish their generalizability. Examples of how these techniques are being used in drilling research and development are provided throughout the review. Three case studies are presented to illustrate the advances being made and the challenges faced. These address prediction rheology and filtration parameters in water-based drilling fluids, rate of penetration predictions from mud-log and petrophysical data, and classification of loss of circulation from a large database of recorded drilling variables with imbalanced class distributions. The benefits are highlighted by recent AI prediction interpretation techniques involving multi-K-fold cross-validation, annotated confusion matrices, and optimizer-assisted feature selection. Β© 2024 Elsevier Inc. All rights reserved.
