02. Master's Thesis

Permanent URI for this collection

https://nur.nu.edu.kz/handle/123456789/8148

Browse

Browsing 02. Master's Thesis by Author "Amanzhol, Elmira"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    ItemOpen Access
    STUCK PIPE PREDICTION IN DEEP WELLBORES DRILLED IN COMPLEX EVAPORITE FORMATIONS USING MLBASED INTELLIGENT CLASSIFIERS
    (Nazarbayev University School of Mining and Geosciences, 2025-04-21) Amanzhol, Elmira
    Stuck pipe events represent a very challenging and high-cost problem in drilling industry. Hence, accurate stuck pipe prediction is crucial for successful field development. Various stuck pipe prediction models using Machine Learning (ML) approaches have been developed in the past. In this study, an attempt was made to address critical problem of stuck pipe incidents prediction while drilling through deep and complex evaporite formations by ML-based models for early hazard detection. Development and testing different intelligent models performed using real field data that included actual drilling parameters along with geological information and drilling mud properties. Actual field data from 10 wells were used to train and test the models using acquired 61 data sets that consist of 610 datapoints. Five supervised classification ML algorithms were used: Logistic Regression (LR), K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Decision Trees (DT) and CatBoost. Before proceeding with training phase, an essential step of data pre-processing was conducted to identify any missing data and detecting outliers. Each model was trained and tested using the 60-40 data splitting strategy to identify stuck pipe condition at both individual hole section level and the entire field level. The developed intelligent models can identify patterns of different stuck pipe conditions including stuck pipe, non-stuck, pack-off, overpull, and tight spot. Error metrices were used to ascertain accuracy, precision, recall, and F1-score of the developed intelligent models in conjunction with ROC analysis to assess performance of the developed intelligent models. Two intelligent models were identified as the most effective classifiers for specified stuck pipe issues: DT model attained 99.59% accuracy at the field level; while, CatBoost reached 100% accuracy in hole section-based assessments. Feature importance analysis together with SHAP (SHapley Additive exPlanations) analysis showed that formation lithology is the leading factor affecting stuck pipe occurrences, as salt and clay contents remain as a primary contributors to this issue. Two key drilling parameters that control stuck pipe occurrence were identified as drilling rotational speed and flow rate. Additionally, mud weight combined with gel strength demonstrated major effects of rheological properties of the mud. SHAP analysis showed that some hole sections outcomes are locally affected by depth and clay content levels. The results of this research demonstrated that successful field development operations require interpretable intelligent models to enhance operational decision-making process to prevent different drilling hazards.