Real-Time Monitoring Model of PDC Bit Wear Based on GRU Neural Network

Abstract

ABSTRACT Improving drilling efficiency is crucial in reducing drilling costs, and monitoring drill bit wear provides an intuitive reflection of drilling efficiency. However, collecting on-site parameters that directly reflect drill bit wear is challenging, and few monitoring methods are available for assessing the degree of wear. Thus, quantitatively evaluating the wear of PDC drill bits has been a persistent research challenge. Evaluating drill bit wear status mainly relies on rock breaking efficiency and mechanical specific energy. In this study, a physical model was used to calculate mechanical specific energy, while wavelet analysis and clustering algorithms were used to characterize the wear process of drill bits. A GRU neural network was established to map drilling parameters to drill bit wear levels, achieving an accuracy of 95%. The model was tested using data from well A in China's Xinjiang oilfield, demonstrating its ability to predict the current drill bit wear level accurately. This model provides a solution for monitoring drill bit wear, assisting on-site personnel in making decisions on when to stop drilling, and enhancing the level of intelligence at drilling sites. BACKGROUND The drilling operation in the oil industry is a complex engineering process, and the drill bit is a critical component. The drill bit not only has to endure immense pressure but also fend off the effects of wear, heat, impact, corrosion, and other factors (Weeden, R, et al. 2011)1. If the wear on the drill bit reaches a certain level, it not only lowers drilling efficiency but also increases the safety risks associated with drilling (Wilson, A, et al. 2018)2. Therefore, monitoring the wear condition of the drill bit is highly important. The current technology for determining when to replace worn drill bits is based on the inferred dullness of the drill bit, derived from the trend of drilling parameters, rather than a reliable engineering method (Waughman, R, et al. 2003)3. Most scholars primarily characterize the trend of drill bit wear failure through changes in mechanical rate of penetration (ROP) (SONG Xianzhi., et al. 2022)12 and drill bit rock mechanical specific energy (MSE) to determine the appropriate time to remove the drill bit (Dupriest, F. E., et al. 2005)4. Quantitatively monitoring the degree of drill bit wear remains a challenging issue. This is primarily due to the influence of complex drilling conditions on subsurface information, (Abbas, R. K., et al. 2014)5 which leads to unavoidable deviations and noise in the recorded well data when reflected to the surface, and the limited number of factors considered in the drill bit wear models obtained from laboratory experiments, resulting in poor model performance in field applications (Rashidi, B., et al. 2014)6. On the other hand, the nature of subsurface operations is characterized by its concealment, making it difficult to quantitatively characterize the drill bit wear process. In recent years, with the development of digital and intelligent technologies, (XIAO Hua., et al. 2018)11 experts both domestically and abroad have conducted extensive research on the intelligent recognition of drill bit efficiency, (LI Hongbo., et al. 2022)13 significantly improving the accuracy of drill bit efficiency and rock breaking condition recognition. However, stability still needs to be improved (Chen, X., et al. 2018)7.