Remote Monitoring and Predictive Maintenance of O&G Wells

The client, one of the world’s largest oil and gas production companies, wanted to move beyond reactive maintenance by remotely monitoring well health and facilitating predictive, proactive interventions. However, they faced challenges integrating and analyzing complex well data and lacked sufficient data to identify key well states. To address these hurdles, Utthunga built an intelligent well health monitoring system leveraging advanced machine learning techniques and data analysis tools.

We trained four ML models on the client’s data to classify well state with improved precision. The outputs of the individual models were fed into an ensemble model to further improve prediction accuracy and robustness. Our novel approach also incorporated dynacard signature analysis using images and numerical data for a comprehensive view. For previously unavailable well states, we generated synthetic data to train the models effectively. Combining TensorFlow, NumPy, Pandas, Matplotlib and more, we built an enterprise-level framework to ingest data, train models and serve predictions.

These efforts led to significant benefits for the O&G company. The highly accurate machine learning models ensured reliable well state identification. The combined model further enhanced prediction accuracy, leading to fewer false alarms and missed issues. By correlating field parameters, early detection of unhealthy well states became possible, enabling proactive maintenance and minimizing downtime. This resulted in optimized production, minimized downtime, and significant operational efficiency gains. With data-driven visibility into well health, we empowered the client to transform maintenance practices and maximize the value generated from assets through optimized production.