Perpustakaan Digital ITB

One of the optimizations in oil wells is the artificial lift method. An Electrical Submersible Pump (ESP) is one of the common tools in oil wells, especially in offshore. ESP operations are inseparable from the monitoring and troubleshooting process. In troubleshooting, Ammeter chart interpretation is one of the methods used for the early detection of ESP problems. Operators or engineers are the main actors in interpreting the ammeter chart in the power control room to identify problems with ESP. It is to prevent operational problems from becoming a major problem. The ammeter chart is used to measure and record the current drawn by the ESP motor and is a representation of the ESP condition based on the electric current pattern on the chart. Based on the pattern, the ammeter chart is divided into several classes according to the problem in ESP. In this study, the ammeter chart was divided into 15 classes and 49% of ESPs were operating under normal conditions. Some of the problems that occur in this case are gas problems, by 11% and followed by other problems. The interpretation carried out is the starting point for inspection of problems in ESP parts, such as pumps, motors, power supply systems, impellers, diffusers, and other components of ESP. To help operators and technicians who are offshore and to overcome the problem of misinterpretation of the ammeter chart as well as differences in interpretation between operators or technicians, the proposed solution is the creation of software that aims for ammeter chart interpretation efficiently, as well as equalize the interpretation results. A machine learning was created to solve this problem through the construction of the image recognition feature and is used as the basis of this software. Besides that, the mobile application construction is carried out as a user interface that is integrated with programs that have been created. This study discusses the prediction of problems in ESP through ammeter chart interpretation with image recognition-based machine learning programs supported by mobile applications. This study recommends a predictive program model to recognize ammeter chart patterns with an accuracy rate of 89.5% as the best scenario.