Detection of electric submersible pump cable damages in crude oil wells using current harmonics data with artificial intelligence methods

In crude oil production, it is preferred to use electric submersible pump (ESP) type pumps in wells with the highest efficiency. In crude oil production wells where ESP is used and high economic value, the high cost of possible failures makes failure detection important. Many different methods have been used from past to present for the purpose of failure detection in oil wells. In this study, both the current failure detection methods in use were examined and especially artificial intelligence-based methods related to the prediction of cable failures before they occur were discussed. For this purpose, an artificial oil well pump system was established and experimental data sets consisting of current harmonics recorded with an energy analyzer were created by creating different levels of failures that were not at a level that would activate the protection elements in the cable feeding the pump. The data were divided into four classes as undamaged, slightly damaged, moderately damaged and very damaged. Appropriate methods were investigated for the prediction of failures. Using the data set consisting of current harmonics obtained for different fault conditions, fault prediction was made with artificial neural networks (ANN), support vector machine (SVM), convolutional neural networks (CNN) and long short-term memory networks (LSTM) methods. According to the results obtained, it was shown that cable fault prediction could be made with 92% accuracy with the LSTM method and this method was the most suitable method. The success rates of ANN, CNN and SVM methods were obtained as 89%, 84% and 82%, respectively.