Automated Cutter Damage Classification for PDC Bits in Delaware Vertical Applications

Abstract The classification of cutter damage in polycrystalline diamond compact (PDC) drill bits has historically relied on manual dull grading, a process that is time-intensive, subjective, and inconsistent. This study introduces an automated classification approach leveraging machine learning and high-resolution imaging to improve the assessment of PDC cutter wear and failure in Delaware Basin vertical applications. A robotically controlled imaging system captures detailed cutter images before repair, feeding data into a convolutional neural network trained to recognize over ten distinct failure modes. The model employs multi-label classification to identify primary and secondary damage mechanisms, enabling a more comprehensive analysis of cutter degradation. By integrating these classifications with operational drilling parameters—including weight on bit, rate of penetration, torque, and cutter velocity—this study reveals key correlations between cutter damage and drilling conditions. The results indicate that lower weight on bit and higher cutter velocity contribute to secondary failure mechanisms, suggesting that improper engagement with the formation increases cutter instability. Additionally, a strong negative correlation between diamond removal percentage and secondary damage presence suggests that multi-modal failures often emerge before extreme cutter degradation occurs. Cutters exhibiting secondary damage tend to experience more rapid wear, reinforcing the need for early intervention strategies. Multi-modal failures were found to reduce total footage drilled per bit, indicating a direct operational impact that could be mitigated through improved parameter selection. The study demonstrates that automated classification not only enhances the accuracy and consistency of cutter damage assessment but also enables proactive decision-making in bit design and drilling operations. By leveraging machine learning-driven classification with real-time operational data, operators can reduce nonproductive time, improve bit longevity, and refine drilling practices based on objective, scalable damage assessments. The integration of machine learning with operational data provides an opportunity to develop predictive maintenance strategies, helping to prevent catastrophic cutter failure. Future advancements may incorporate real-time imaging systems for on-rig damage monitoring, allowing for dynamic drilling parameter adjustments that further improve efficiency, cost reduction, and bit performance longevity.