Chad Dueck, Jaime Hecht, and Burke Pond
Ambyint
The upstream oil and gas industry faces significant challenges in optimizing production from aging assets, particularly in managing the vast amounts of unstructured data generated by rod lift systems. This paper presents field results from the deployment of Cognitive Card Recognition (CCR), a machine learning-based solution for automated dynacard analysis and anomaly detection in rod lift operations.
The CCR system, developed through collaboration between rod lift subject matter experts and data scientists, employs multiple machine learning models trained on millions of expert-labeled dynacards. Current models achieve 85-95% accuracy in identifying twelve distinct non-normal operating conditions, including fluid pound, gas interference, worn pumps, and rod parts. The system continuously improves through regular incorporation of additional labeled data and model retraining.
Field case studies demonstrate CCR's ability to identify critical operational issues days to weeks earlier than traditional methods. In one documented instance, CCR detected a hole in barrel condition before production decline occurred, enabling proactive maintenance scheduling. In another case, early detection of a rod part reduced failure cycle time by 1-2 days, minimizing deferred production and preventing cascading equipment damage.
Results show that CCR implementation enables operations teams to transition from reactive to proactive maintenance strategies, leading to reduced deferred production, decreased well downtime, and optimized maintenance scheduling. This technological advancement represents a significant step forward in leveraging artificial intelligence to improve oil production efficiency and equipment reliability in aging fields.
Keywords: artificial intelligence, rod lift optimization, predictive maintenance, machine learning, oil production, dynacard analysis.