Operational Insights with Predictive Gamma Analysis to Enhance Drilling Performance

Abstract The drilling process in oil and gas extraction is a multifaceted operation requiring the integration of various systems and technologies. These systems include fluid systems, mechanical rig systems, the drill string, and the bottom hole assembly (BHA), all of which play critical roles in achieving efficient and safe drilling operations. This study focuses on optimizing drilling operations by predicting gamma changes and trends based on drilling parameters measured at the surface, aiming to enhance drilling efficiency, reduce non-productive time (NPT), and minimize the frequency of unplanned trips due to measurement-while-drilling (M/LWD) tool failures. The study underscores the importance of "knowing the rock" by reducing uncertainty associated with gamma signal lag. This approach empowers drilling teams to manage formation transitions better and improve geosteering, particularly as the industry moves towards longer laterals. In vertical applications, accurate formation transition management and avoidance of unplanned trips due to M/LWD failures are critical for optimizing performance. In lateral applications, predicting gamma trends at the bit enables more precise geosteering and allows drilling to continue despite tool failures. This paper outlines the method used to generate usable machine learning algorithms for gamma trend prediction using only surface drilling parameters, including: data selection, data preparation, feature engineering, model selection and the challenges experienced. The data preparation process is discussed in detail, including initial data selection, smoothing, filtering, gap handling. Feature engineering involved selecting and transforming relevant drilling parameters, such as weight on bit (WOB), surface rotations per minute (SRPM), differential pressure (DIFF), rate of penetration (ROP), and surface torque (TQ). Pre-processing was conducted to identify steady and non-steady state drilling regimes, with unique models trained for each hole section. Two types of supervised classification models, Random Forest and XGBoost, were tested for gamma change trend prediction across three different hole sections. The results demonstrated the models' ability to identify and correlate changes in gamma trends with drilling parameters commonly measured on every rig, making the model widely applicable. Discrete models were trained per field and per hole section to account for the differences in drilling techniques across hole sizes and trajectories and proved successful with a low volume of offset data. The model results showed promise for enhancing drilling efficiency and reducing NPT. In conclusion, this study shows that machine learning techniques can effectively predict changes in gamma trends at the bit by correlating these changes with drilling parameters to optimize drilling operations.