Revealing and Solving the Hidden Hinders for Efficient Shale Gas Drilling Based on Data Analytics and Machine Learning

Abstract Despite annual production of shale gas exceeding 20 billion cubic meters from Sichuan basin in China, its cost-effective development has been constrained by lengthy drilling cycle of horizontal sections caused by downhole complex such as leakage, stuck pipe, multiple drilling trips due to frequent downhole tool failure, low ROP, bit wear and other related challenges. Progress has been realized in reducing NPT and increasing ROP due to extensive effort in complex prevention and drilling optimization in the last few years. However, little attention has been paid to the influence of downhole vibration and bit wear misjudgment on efficient shale gas drilling. This paper presents the revealing and solving of these two hidden hinders based on data analytics and machine learning. Various sources of data, including over 300 cases of bit records, lithology, elemental logging and downhole vibration measurements from drilled formation for selected representative drill bits, as well as the data from full-scale indoor experiments, were collected for analysis and comparison. The characteristic of bit vibration and rock-breaking efficiency, obtained from five specifically designed 8-1/2″ PDC bits with different levels of worn cutting teeth which were used to drill artificial cores with varying degrees of heterogeneity, were evaluated with data from bit records and downhole measurements. Furthermore, identification of bit wear states was realized to reduce unnecessary POOH by combing RFA and data from bit records. It's revealed that severe downhole vibration, caused by formation heterogeneity and reflected in mineral composition when crossing different layers, is one of previously overlooked causes for downhole tool failure and premature bit wear. Consequently, trajectory optimization was proposed to reduce downhole vibration. Experimental results suggested the wear value of 2-2 should be defined as "critical wear value" for PDC bit as significant difference in rock-breaking performance were observed between this threshold. However, over 30% of POOH due to PR from collected bit records ended up with bit wear grades no greater than this value, underscoring the significance of accurate bit wear identification for avoiding unnecessary POOH. This proposed critical value was regarded as benchmark to differentiate whether the bit is suitable for continued drilling or requires replacement. Multiple bit rock-breaking performance indicators such as DOC, ROP and DOC rate of change were used to identify drill bit wear states with thresholds obtained by model training with RFA. The average prediction accuracy for bit wear states along well depth of selected bit types, based on depth-based rock-breaking performance indicators, was 88.9% and 90.5% for horizontal drilling with and without downhole motor, respectively. The revealing and solving of these hidden hinders highlight the advantages of digital driven techniques in better capturing and utilizing digital data to minimize NPT and optimize drilling operation.