REAL-TIME ANALYSIS OF BOREHOLE INTEGRITY THROUGH MODEL BASED AND DATA DRIVEN METHODS

Abstract

Borehole stability remains a critical problem in drilling, specifically in challenging formations like shales found in the North Sea, where wellbore instabilities can lead to costly delays and safety risks. Traditional borehole integrity models are limited in real-time application, often due to data aggregation issues and sensor offsets that prevent direct measurement at the bit. Furthermore, many models claiming to be real-time lack the capability to incorporate time-dependent factors, such as pore pressure diffusion, thermal and chemical effects, all of which significantly impact borehole stability. This study develops a real-time borehole integrity advisory system that integrates a porothermochemoelastic model with the Mogi-Coulomb failure criterion. The system begins by employing machine learning (ML) algorithms to predict lithology and compressional wave velocity (DTCO) at the bit using drilling and petrophysical data. All predictions were validated using data from separate validation well. Among the evaluated models, random forest regression demonstrated the highest accuracy (R^2 = 0.92 for DTCO), while random forest classification achieved lithology prediction accuracies of 86% for the entire well and 95% in reservoir sections based on manually labeled data. Rock mechanical parameters, including Unconfined Compressive Strength (UCS) and flow factor, were calculated at the bit using rock strength correlations linked to the predicted lithology and DTCO values. To ensure the reliability of the input parameters for real-time borehole stability models, UCS and flow factor were also calculated using measured DTCO and actual lithology from the same validation well. Strong correlations were observed between the predicted and measured-derived parameters, with R^2= 0.82 for UCS and R^2= 0.92 for flow factor. The first model, a non-time-dependent approach, calculates minimum drilling fluid density (MDFD) and breakout depths to support immediate drilling decisions. The porothermochemoelastic model predicted an MDFD of 1.13 g/cc, closely matching the actual field value of 1.14 g/cc, and demonstrated that drilling could safely proceed at this lower density without risking borehole instability. The second model integrates time-dependent effects, analyzing variations in pore pressure, temperature, and chemical interactions over time. Time-dependent breakout analysis revealed that shear failures appeared between two to six hours, with failures remaining localized in sandstone-dominated reservoir sections after approximately 12 to 24 hours. In shale-dominated intermediate sections, breakout depths increased significantly after 16 hours, with failure progression being more apparent due to the combined influence of chemical and thermal effects. Field case analyses further validated the model, with pseudo-caliper logs aligning with actual caliper logs, while erratic hook loads confirmed the detection of problematic regions in shales. Time-dependent failed arc results identified 47° as a critical threshold, beyond which breakout depths became significant, indicating potential instability risks. A key contribution of this research is the real-time implementation of borehole stability analysis directly at the bit. Unlike traditional systems reliant on pre-drill inputs, this framework dynamically updates input parameters using real-time data, overcoming sensor offset challenges. The analytical model balances computational efficiency and accuracy, making it a practical alternative to finite element methods (FEM) for time-dependent scenarios.