A Physics-Informed Decision Support System for Drill Bit Performance and Parameter Selection - Final Report

Abstract

The performance of drill bits significantly influences drilling efficiency and cost-effectiveness in oil and gas operations. Traditional selection and optimization methods, which often rely on historical data and expert judgment, are limited when facing complex and dynamic downhole environments. This study introduces an integrated machine learning (ML) framework designed to optimize drill bit performance by predicting Rate of Penetration (ROP) and evaluating energy efficiency using Mechanical Specific Energy (MSE). Using historical drilling data, physically meaningful features were engineered and incorporated into a LightGBM regression model. The model achieved high predictive accuracy (R² > 0.91) for both PDC and Tricone bits. A composite Performance Score was developed to rank scenarios by balancing speed and efficiency. Extensive simulation grids enabled robust recommendations across varying formation hazards and drilling conditions. Results showed that optimal parameter combinations vary by bit type and formation complexity, with Tricone bits generally performing more consistently in challenging formations. The framework bridges the gap between data science and engineering by enhancing model interpretability and offering actionable insights through quadrant-based performance analysis. This work demonstrates the potential for ML-based decision-support tools to significantly improve drilling operations through informed parameter optimization