Leveraging Machine Learning and Data Lakehouse to Enable UCS Prediction and Data-Driven Bit Design Selection

This study presents an integrated workflow leveraging Machine Learning (ML) and the Data Lakehouse architecture to automate Unconfined Compressive Strength (UCS) prediction and optimize bit design selection for drilling with Rotary Steerable Systems (RSS) through stringer formations. Data from multiple sources, including Logging While Drilling (LWD) and drilling performance records, were unified within a corporate Data Lakehouse, enabling automated preprocessing, model training, and visualization. Using AutoML functionality, the Gradient Boosting algorithm was identified as the optimal model, achieving significant accuracy improvement as the training dataset expanded from 1 to 40 wells. The inclusion of casing shoe and float valve parameters further enhanced model robustness, aligning with prior studies. Results revealed that drill bits with 4D cutter technology provided superior performance, achieving a 20% increase in Rate of Penetration (ROP) within horizontal sections compared to the previous year. The developed framework enables full automation of UCS estimation and bit performance evaluation, reducing manual intervention and ensuring consistent, data-driven decision-making. The proposed approach demonstrates the potential of the Data Lakehouse paradigm for scalable, real-time drilling optimization and predictive analytics, with future improvements expected through integration of downhole Weight on Bit (WOB) and Torque (TQ) measurements for enhanced model precision.