Flow regime identification in the subsea jumper based on electrical capacitance tomography and convolution neural network

Abstract

The subsea jumper plays an indispensable role in the subsea production system as a connecting component between Christmas trees, manifolds and risers. Flow pattern changes inside the jumpers directly affect their safe operation. However, the research work on flow pattern identification of jumpers is limited at present. In this paper, the M-shaped jumper is chosen as the research object and two-phase flow experiment system is constructed. In the range of gas superficial velocity 0–5 m/s and liquid superficial velocity 0.1–4 m/s, firstly, four typical flow patterns, namely, bubbly flow, slug flow, annular flow, and stratified flow, are collected by the electrical capacitance tomography (ECT) across the jumper to establish the flow pattern data sample set. Secondly, the EfficientNet convolutional neural network (CNN) is used to train and learn the flow pattern. It is proved that EfficientNet-B5 is the most suitable and the flow pattern recognition accuracy reaches 95.6%. Finally, to enhance the recognition accuracy of similar flow patterns, EfficientNet-B5 is optimized by using the advanced optimizer Adam, with an accuracy rate of 97.55%. This research provides a new method for monitoring the flow pattern of deep-sea oil exploration and transportation, which is meaningful for the safe operation of the subsea jumper.