Real-Time Borehole Stability Analysis at the Bit Using Analytical and Machine Learning Approaches

Borehole stability is critical in drilling, particularly in shale formations, where failures can cause significant challenges. Traditional models for borehole stability rely on empirical rock strength correlations using logging-while-drilling (LWD) data, often unavailable in top-hole sections or recorded far above the bit. This paper introduces a real-time advisory system that integrates machine learning (ML) and porothermochemoelastic analysis to dynamically calculate key parameters like minimum drilling fluid density (MDFD) and breakout depth. The system predicts lithology and DTCO at the bit in real-time, using ML algorithms trained on drilling parameters and gamma-ray data from Norwegian continental shelf wells. Lithology prediction achieved accuracies of 86% for entire wells and 95% for reservoir sections, while DTCO predictions showed an R2 of 92%. These predictions enabled rock mechanical property estimates, such as UCS and flow factor, with accuracies of 90% and 92%, respectively. The porothermochemoelastic model predicted an MDFD of 1.13 g/cc, slightly lower than the field value of 1.14 g/cc, suggesting potential optimization of drilling fluid usage. Compared to conservative estimates from linear elastic (1.27 g/cc) and poroelastic (1.20 g/cc) models, the integrated approach accounts for thermal and chemical effects, offering enhanced accuracy and efficiency in borehole stability analysis.