Diagnostic of Operation Conditions and Sensor Faults Using Machine Learning in Sucker-Rod Pumping Wells.

Area of Science:

• Petroleum Engineering
• Artificial Intelligence
• Data Science

Background:

• Sucker-rod pumping wells face operational challenges due to undetected issues and sensor faults, leading to increased downtime and production loss.
• Current diagnosis relies on manual analysis of downhole dynamometer cards, which is labor-intensive and prone to errors.
• Existing machine learning applications show promise but lack clarity on task difficulty and optimal methodologies.

Purpose of the Study:

• To investigate the efficacy of machine learning algorithms in diagnosing operating conditions and detecting sensor faults in sucker-rod pumping wells.
• To evaluate different machine learning algorithms, data descriptors, and automated machine learning pipelines for dynamometer card analysis.
• To address uncertainties regarding the classification difficulty and best practices for applying machine learning to dynamometer card data.

Main Methods:

• Conducted sixty tests using over 50,000 dynamometer cards from 38 wells.
• Evaluated three algorithms: decision tree, random forest, and XGBoost.
• Utilized three descriptors (Fourier, wavelet, card load values) and automated machine learning pipelines, assessing hyperparameter tuning, dataset balancing, and various metrics.

Main Results:

• Demonstrated the capability to detect sensor failures using dynamometer card data.
• Achieved high accuracy in 75% of tests, with results exceeding 92%.
• Reached a maximum accuracy of 99.84% in fault detection and condition diagnosis.

Conclusions:

• Machine learning models, particularly when optimized, can reliably detect sensor failures in sucker-rod pumping wells.
• The study provides valuable insights into the performance of different algorithms and descriptors for dynamometer card analysis.
• Implementing these machine learning techniques can significantly enhance the efficiency and reliability of oil well operations.