Machine Learning Solution for Predicting Vibrations while Drilling the Curve Section

Area of Science:

• Drilling Engineering
• Machine Learning Applications
• Vibration Analysis

Background:

• Downhole vibrations significantly impact drilling equipment performance and operational efficiency.
• High vibration levels lead to equipment failure, increased costs, and nonproductive time.
• Existing downhole sensors for vibration detection are expensive and add to operational costs.

Purpose of the Study:

• To develop and validate machine learning models for automated drillstring vibration detection using surface data.
• To assess the effectiveness of various ML techniques in identifying axial, torsional, and lateral vibration modes.
• To provide a cost-effective alternative to downhole sensors for vibration monitoring.

Main Methods:

• Utilized surface drilling data for drillstring vibration detection.
• Employed four machine learning techniques: Adaptive Neuro-Fuzzy Inference System (ANFIS), Radial Basis Function (RBF), Functional Networks (FN), and Support Vector Machines (SVM).
• Developed models through data gathering, wrangling, statistical analysis, model development, and accuracy evaluation using real field data.

Main Results:

• ANFIS and SVM models demonstrated the highest accuracy, with a coefficient of correlation (R) between 0.9 and 0.99.
• RBF and FN models showed good accuracy, with R ranging from 0.82 to 0.96.
• Validated models achieved high prediction accuracy for all three vibration modes (axial, torsional, lateral) on unseen data, with R > 0.93 and AAPE < 2.8% for SVM and ANFIS.

Conclusions:

• Machine learning models, particularly ANFIS and SVM, can effectively detect downhole drillstring vibrations using only surface data.
• This approach offers a technically accepted and cost-effective solution for real-time vibration monitoring.
• The developed ML algorithm can serve as an intelligent tool to enhance drilling operations and reduce costs by eliminating the need for downhole sensors.