Perpustakaan Digital ITB

Electric Submersible Pump (ESP) is a highly effective artificial lift method for boosting oil production in both onshore and offshore fields. The ESP maintenance will be conducted regularly as other artificial lift to prevent costly production disruptions due to unexpected pump failures. Many diagnostic methods have determined the ESP system's status by using the automation system; however, these methods usually only provide backward-analysis after failure events have occurred. This paper involves acquired real-time data to establish an analytical methodology to detect impending ESP failures. The classification will be done on ten minutes-interval data forecasting performance, which shaped up into slope. This is primarily achieved using Supervised Learning Technique; Logistic Regression, Random Forest, Support Vector Machine, Decision Trees, and K-Nearest Neighbors. The models will be built based on individual distinct parameter's characteristics of nine status consists of; low PI, pump wear, tubing leak, higher PI, increase in frequency, open choke, increase in water cut, sand ingestion, and closed valve with an accuracy rate over 95%. These automation and control systems require constant surveillance by a human operator to verify that all processes are running normally. Furthermore, the abnormal behavior is identified in advance, and the operators can early determine the best corrective action to avoid an ESP's failure built upon the recommendations attached. It is the human operator's responsibility to react to any alarm conditions that occur during operation. This introduced technology is an effective way to monitor the ESP system that leverages Artificial Intelligence. The operator can rely on the surveillance system's ability to detect abnormal behavior, allowing the operator to focus on higher priority tasks. Moreover, an engineer's significant advantages are taking pre-emptive action to avoid failure and generating billions of revenues.