Predicting Drill Bit Remaining Useful Life and Identifying Drilling Disfunctions by Physics-Based Inversion and Data-Driven Model - Case Study of Utah Geothermal Wells

Abstract Drilling through hot rocks in geothermal wells often leads to multiple unplanned trips due to issues such as drilling dysfunctions, thermal degradation of cutters, inadequate hole cleaning, tool failure, and lost circulation. Accurately predicting bit conditions in real-time can help avoid poor decisions, such as continuing to drill with a damaged bit or pulling the bit out too early. This paper focuses on developing a hybrid model that combines physics-based inversion with data-driven techniques to accurately estimate the remaining useful life (RUL) of drill bits and detect drilling dysfunctions in geothermal wells in Utah. This approach combines inversion models for depth relaxation with machine learning algorithms to predict the drilling performance of each bit run in geothermal drilling operations. Data on drilling dynamics, total flow area (TFA), and bit dull grade were collected from 63 PDC bit runs in the lateral sections of 15 geothermal wells with an 8.75-inch hole size. The collected data was used in physics-based inversion to calculate relaxation depth, while the machine learning model identified patterns related to bit wear, drilling anomalies, and dysfunctions. Historical data and real-time drilling parameters were utilized to train and validate the model. The hybrid model showed substantial improvements in predicting the RUL of drill bits and detecting drilling anomalies and dysfunctions. Specifically, it was observed that short run depths could result from fast relaxation depth, which is linked to inefficient drilling, anomalies, or dysfunctions. Additionally, the study found that increasing TFA could mitigate the thermal degradation of bit cutters. These factors—relaxation depth, run depth, and TFA—were used as inputs for a machine learning model to predict RUL and bit dull grades for each bit run. Furthermore, the physics-informed machine learning model can guide the selection of drilling set points when anomalies or dysfunctions occur, enabling timely resolution. The integration of physics-based and machine learning approaches yielded high prediction accuracy, reducing non-productive time and optimizing drilling performance. This method has the potential to be applied across various geothermal fields, providing a valuable tool for enhancing drilling efficiency and lowering operational costs. This research presents a novel integration of physics-based and data-driven models for predicting drill bit RUL and identifying drilling dysfunctions. A key aspect is the exploration of the previously unexamined relationship between run depth and relaxation depth in the inverted domain. This analysis aims to enhance predictive accuracy and operational efficiency in geothermal drilling. The innovative approach marks a significant contribution to the advancement of drilling optimization techniques.