Inertial Effect of Drill Pipe Friction Trend and Early Warning of Stuck Pipe

Abstract Common stuck pipe caused by borehole uncleaning and wellbore geometry are a major threat to oil and gas drilling engineering, resulting in significant time and economic costs. These are highly correlated with the dynamic friction of the drill string. Drillstring friction is an important function of cuttings bed and wellbore geometry, etc. and is a depth sequence. Existing research mainly focuses on machine learning and friction calculation for analyzing the stuck pipe trend. However, these methods often overlook the key mechanism parameters contributing to two types of stuck pipe, such as the cuttings bed and wellbore trajectory, etc., resulting in low accuracy. In addition, these methods are primarily real-time calculation of friction coefficient rather than prediction in advance for analyzing the stuck pipe trend, leading to a lack of timeliness. To solve these problems, a prediction method of drill string friction coefficient in advance during tripping operations was developed, which can be employed for early warning of stuck pipe. In the data preparation stage, the friction coefficient is calculated in real time using the Torque & Drag model to characterize the stuck pipe trend during tripping operations. Additionally, key mechanism parameters such as additional cuttings friction, trajectory, and formation characteristics were introduced to characterize the mechanism of borehole uncleaning and wellbore geometric stuck risk, respectively. In the modeling stage, the Graph Attention Network (GAT) using different depth friction coefficients was stablished as graph nodes and key mechanism parameters as auxiliary features. Unlike conventional machine learning techniques, GAT can autonomously uncover potential relationships between nodes and subsequently predict the next node. The Self-Attention mechanism enables the automatic computation of weight values between nodes representing the friction coefficient to simulate the autocorrelation of the friction coefficient at different depths. The results indicate that the friction coefficient usually maintains the internal motion pattern of the previous two single strings, which reveals and simulates the depth inertial effect of the friction coefficient. Employing this effect, the friction coefficient of the next single string under tripping operations can be predicted. This method facilitates the early warning of the stuck pipe caused by borehole uncleaning and wellbore geometry. The reliability of the model is verified by the data from 21 Wells in two oil fields. The results show that the mean absolute percentage error (MAPE) of the friction coefficient ahead of one single string during tripping in and tripping out is 5.4% and 13.1%, respectively, and can quantitatively simulate the inertial effect of the friction coefficient. In the actual stuck pipe case, the early warning occurred eight (8) minutes before the stuck event was confirmed on the field, which can reserve sufficient handling time. This method overcomes the problems of insufficient timeliness and accuracy in existing methods and realizes the early warning of borehole uncleaning and wellbore geometry stuck pipe consequently.