Machine Learning-Based Drill Bit Wear Prediction for Enhanced Drilling Performance

Confronting the inefficiencies in current drilling industry practices, particularly the limitations of traditional scalar value-based assessments. This study introduces a novel machine learning-based approach to predict tool wear rate and assess drilling tool performance in real-time, By analyzing a comprehensive database, including formation properties, bit design, and usage, we aim to reduce non-productive time (NPT) and drilling costs. The methodology involved extracting data from end-of-well reports and using custom scanning tools to compile a dataset of 743 entries with 26 wear classes for 21 types of drilling tools from four major manufacturers. This dataset was refined to five key wear states and processed using K-means and SMOTE algorithms to balance class distribution. The study's central outcome is a tool condition assessment model, with the Random Forest classifier showing the highest efficacy, achieving 87% precision and 81% accuracy. Despite facing challenges due to data diversity, our findings underscore the correlation between tool wear and drilling parameters, the impact of equipment selection, and the potential for formation-specific model training. The study demonstrates the model's capability to enhance drilling operations by optimizing tool selection and usage, marking a significant advancement in predictive drilling technologies. 1. INTRODUCTION In the pursuit of enhancing drilling operations, the industry has traditionally relied on basic scalar values such as the rate of penetration (ROP) to estimate the condition of the drill bit. However, ROP is influenced by a myriad of factors, from drilling parameters and equipment to lithological properties, leading to decisions that significantly impact non-productive time (NPT) and thereby, the cost efficiency of the well. The lack of accurate bit state indication not only contributes to inefficient penetration but can also precipitate equipment failure, invoking costly mitigation techniques. Recent advancements in drilling technology, therefore, have underscored the importance of understanding and predicting drill bit wear more accurately and cost-effectively. Ivanicova et al. (2018) and Yang et al. (2020) delve into the indirect prediction of drill bit wear and real-time bit wear prediction using mud logger data, utilizing models based on penetration depth and mathematical approaches like PCA and wavelet analysis. Similarly, Wang et al. (2022) and Cheatham et al. (1985) focus on the wear of Polycrystalline Diamond Compact (PDC) bits, employing methodologies like finite-element numerical simulation and laboratory drilling tests to reveal the drastic effects of field wear on performance. In a dynamic shift towards a more sophisticated understanding, Samuel and Liu (2023) introduce adaptive data analytics for bit wear prediction, emphasizing the potential of real-time analysis and the dynamic nature of drilling environments using Empirical Mode Decomposition. Moreover, Capik et al. (2021) and Rashidi et al. (2008) explore the modeling of bit wear during drilling operations and real-time prediction by combining rock energy and drilling strength concepts.