THE PREDICTION OF CASING RUNNING DRAG LOADS USING MACHINE LEARNING TECHNIQUES

MƏQALƏ VƏ TEZİSLƏRİN TƏRTİBİ QAYDALARI THE PREDICTION OF CASING RUNNING DRAG LOADS USING MACHINE LEARNING TECHNIQUES. Vusal Iskandarov Akifkhan 1 [email protected] 1Azerbaijan State Oil and Industry University ABSTRACT Key words: Drilling, Machine Learning, Torque and Drag, Drilling Optimization. The objective of this study was to build a model in order to predict tripping in weight from the real time data. The study is aiming to support comprehensive torque and drag analysis for casing running operations and further optimize the modelling job by cutting the time required to analyse the real time data. This study involves several machine learning algorithms so as to build the model using real time data. The real time data from offshore fields located in South Caspian Basin was applied to train the model. The different machine learning techniques, such as linear and non-linear machine learning and deep artificial neural networks, trained model. The evaluation metric for training is Root Mean Square Error, however, the performances of the regressions are evaluated on the data using R-squared for their comparison. The need to build the model to optimize casing running torque and drag simulations was raised when the side-track operations were commenced in the mature field where it was required to run 4.5” liner in slim-hole conditions with challenging well trajectory and severe dogleg severities. A couple of failures had already been occurred in offset wells, in which the 4.5” liner got stuck in slim open hole section. Also, in some cases, even though the liner was successfully run to the required depth, cementing operations were poorly performed due to poor centralization in the well. Therefore, as one of the improvements to manage the risks and optimize the operations, comprehensive model is determined to be built to further optimize and precisely predict torque and drag loads for casing running operations. The paper will look at in depth look at how model was built for the first well, then calibrated with drilling data, and how the friction factors together with post job friction factors were easily analyzed. This paper also covers why this study is important, how it is different from the existing workflows and how machine learning can help to automate the time-consuming process in drilling engineering. GİRİŞ (Introduction) South Water Gunashli is an offshore oil field located in South Caspian Basin. The start of the production from the field was dated back to 1982 and boomed since (Leonid , George , & Fred , 2001). Even though the field development started more than 40 years ago, the first sidetracked in-fill production well was drilled in 2023. During the drilling operation, a lot of difficulties had been happened. During the side-track operation planning and execution, it was becoming apparent that special engineering calculations and care should be considered for drilling and casing running operations as the side-tracked wells will have slim-hole sizes because of the old well design. In many cases, planned wells need 4 ½” production liner across sidetracked hole section, and this created a lot of challenges regarding torque and drag forces and ECD limits in both liner running and cementing operations. At the beginning of the drilling campaigns, these challenges had resulted in liner stuck across open hole, even in one case, liner got stuck approximately 500 m below casing exit window. This has triggered drilling engineering and well planning team to be more pre-cautious about casing running. The goal of this research is to share a case study of new engineering tool that helps to forecast liner running drag values in real-time from Leuza 2 mud-logging system which was preliminary used in the drilling operations. The machine learning techniques are frequently being used to address challenging and timeconsuming work process in the industry nowadays. One of the important but time-consuming works that drilling engineers do is analysing real time data to further optimize torque and drag simulations. Usually, the drilling engineers build torque and drag model in advanced software, such as WellPlan, DrillBench etc, using essential input parameters: wellbore information, open hole size, casing restrictions, well trajectory, casing string data, liner running tool information, liner hanger restrictions, casing running equipment data, rig limitations, centralizer data and centralizer placement. To correlate the model to actual expected weight data, one of the most essential parameters is “friction factor”. The term “friction factor”, is obtained from friction, an important source of energy loss. Since the tubulars and downhole equipment are being moved in and out during the casing running operations, the resistive force – friction force is observed in the form of drag. During the casing running operations, various operations have followed each other – tripping in, tripping out, tripping in or out with and without rotation. These obviously complicated the calculations, as a result various friction factors are used to determine those weights (Samuel, Friction factors: What are they for torque, drag, vibration, bottom hole assembly and transient surge/swab analyses?, 2010). To add more context on friction factors as they are core of this study - The Friction Factor used in Torque and Drag calculations represents the multiplier applied to the side force to determine the resulting frictional force. Friction within the wellbore arises from various physical mechanisms and is influenced by factors such as the characteristics of the contacting surfaces, the type of drilling fluid, the wellbore’s trajectory, the extent of contact, and the presence of obstructions like drilled cuttings. In most scenarios, the lubricating properties of the drilling fluid play the most significant role in defining the appropriate friction factor. It's common practice to use different friction factors for Cased Hole and Open Hole sections. Considering the complexities of the simulations to accurately determine the casing running weights, sensitivity analysis of friction factors is usually performed upon any casing running operations (Figure 1). Figure 1 - Torque and Drag Loads with various Friction Factors. Drilling engineers may spend a couple of days on running those sensitivities and combining them in a single place. The range of friction factors usually depends on the field, length of the casing string, hole size, well trajectory, casing size and restrictions in the wellbore, and many other factors. The friction factors are also varied for open hole and cased hole. For complex operations, they are much more complicated to analyze because they depend on various other factors, including temperature, asperities between the surface, type of materials, sliding speed, and rolling speed in case of an object rotating relative to another. Also, the real unknown is the measurement of the hookload at the surface (Samuel, Friction factors: What are they for torque, drag, vibration, bottom hole assembly and transient surge/swab analyses?, 2010). The sensitivity analysis results for casing running operations gives the sense of surface hookloads expectations and the range where the casing running weights may be expected (John & David , 2013). The usual range for the casing running friction factors are within 0.15 to 0.30. However, in extreme conditions, they may be well over 0.45 as well (Rabbat, 1985). Once the casing running operations are completed, the drilling engineers collect the mud logging outputs with all the real-time data from the rig and analyse the casing running loads (Samuel & Jamal, Drilling Engineering, 2007). They match the tripping in, tripping out, tripping in or out with and without rotation weights reading from the mud logging system and check them against the mode outputs. This process helps drilling engineers and well planners to check the friction factors and calibrate their models for the future well planning activities MATERİAL VƏ METODLAR (Methods) The main concept of the tools lies on machine learning (ML) techniques and combination of ML techniques with torque and drag simulations. Machine learning techniques were analysed to determine liner running forces from the real-time data sets from mud-logger system and then to calibrate it for optimum selection of friction factors for the next well delivery. The main problem was running liner to required depth. It may sound easy, yet due to long open hole section, covering multiple layers, requirement to isolate high pressure ramp and depleted reservoirs in a single section, as well as high dog-leg severity across casing exit, exacerbated the situation in almost each well. As the 4 ½” liner length was approximately in range of 1200 – 1700 m, extensive engineering calculations were required to be done in well planning stage. ƏLDƏ OLUNAN NƏTİCƏLƏR (Results) The engineering solution was created and tested against the multiple well data sets. The models are trained using a variety of machine learning algorithms using mud logger data from the Leuza 2 program. The suggested approach offers precise friction factors and its calibration for torque and drag software in addition to help with drag force in well planning. Of all the methods examined, Random Forest turns out to be the best option because it performs better in terms of accuracy and computing efficiency. This study fills a significant gap in the field by providing a workable solution for improved operational planning in non-traditional wellbore circumstances. This project's main goal is to create and evaluate a predictive model for casing running drag values. With the use of machine learning techniques and mud logger data from the Leuza 2 program, the model aims to improve well delivery and well integrity by calibrating FFs for well planning. This model can be used in practice to provide accurate friction factors for torque and drag software as well as real-time monitoring. With an emphasis on efficiency, accuracy, and computational effectiveness, the project seeks to enhance operational planning in unconventional wellbore situations, ultimately advancing drilling technologies. To create a predictive model for liner running drag values, this research makes use of mud logger data from the Leuza 2 software. Basically, from Leuza data and considering liner running procedures, it is manually challenging to obtain optimum running loads and get exact running loads throughout the wellbore. Therefore, this model was created. Furthermore, the research data is obtained from real-time drilling parameters using a variety of supervised machine-learning algorithms, including linear and non-linear models. The Random Forest approach is preferred in a comparative study due to its higher accuracy and computational efficiency. The correctness of the model is evaluated through the use of the Root Mean Square Error (RMSE). The model's final use will enable real-time drag force monitoring, exact friction factor provision for torque and drag software, and operational planning optimization in non-standard wellbore circumstances. Figure 2 - Determined running loads of liner. Optimum friction factors were obtained by using ML techniques from the very first well and these data were implemented into the next well campaign to improve the well centralization plan which allowed for to enhancement of cement bond long and zonal isolation. During the execution, this tool was quite crucial for keeping an eye on drag forces when deploying a 4.5" liner in difficult and tight wellbore conditions. This result was initially checked manually with drillers running loads and was confirmed to be correct. MÜZAKİRƏ (Discussion) In conclusion, this study successfully establishes the Random Forest algorithm as an optimal solution for predicting liner running drag values in slim-hole sidetrack operations, achieving a remarkable average R-squared of 0.90. The practical viability of the developed model in realtime drag force monitoring, coupled with its ability to provide accurate friction factors, underscores its significance for operational planning in unconventional wellbore scenarios. These findings contribute to advancements in drilling technologies, offering a reliable tool for optimizing efficiency and precision in predicting and managing liner running drag values. ƏDƏBİYYAT SİYAHISI John , M., & David , W. (2013). A Work Method to Analyzing Friction Factors in Torque and Drag Modeling. SPE Unconventional Resources Conference Canada. doi:https://doi.org/10.2118/167172-MS Leonid , B., George , C., & Fred , A. (2001). Petroleum Geology of the South Caspian Basin. (E. Inc, Ed.) Gulf Professional Publishing. doi:2001 Rabbat. (1985). Friction Coefficient of Steel on Concrete or Grout. Journal of Structural Engineering 111, 505–515. doi: https://doi.org/10.1061/(ASCE)0733-9445(1985)111:3(505) Samuel, R. (2010, September ). Friction factors: What are they for torque, drag, vibration, bottom hole assembly and transient surge/swab analyses? Journal of Petroleum Science and Engineering, Volume 73 (3-4), 258-266. doi:https://doi.org/10.1016/j.petrol.2010.07.007 Samuel, R. (February 2010). Friction Factors: What are They for Torque, Drag, Vibration, Bottom Hole Assembly and Transient Surge/Swab Analyses? IADC/SPE Drilling Conference and Exhibition. doi:https://doi.org/10.2118/128059-MS Samuel, R., & Jamal, A. (2007). Drilling Engineering. PennWell Corp. doi:ISBN-13 : 9781593700720