Algorithmic Models for Determining the Flow Patterns of Oil Pipelines in Gravity Sections

Transportation of liquid energy carriers by pipelines is accompanied by obligatory commodity-transport operations, in particular - determination of the volume of the pumped product. In the process of operation there are complicated sections of pipelines - gravity pipelines, where the determination of the filling factor is practically impossible. In order to solve the problem, the most probable flow patterns of multiphase flow in the pipeline were determined using Ansys software. Then the approach was developed through determination of the flow pattern inside the pipeline using the previously developed computer program, which includes determination of the filling level of the pipeline by the Newton-Gauss method, determination of one of ۸ classes of the flow pattern and recalculation of the filling factor of the pipe body. The result of the research was the developed computer vision model using the resnet-۵۰ architecture. The possibility of determining the oil flow regime at an acceptable high level has been established. The performance of this model was evaluated through an acceptable ROC-AUC value for each class (all metrics were above ۰.۵, namely, almost all classes have a value of ۰.۸, which is high for this kind of multiclass classification). Based on the results of the work, a stand and a prototype control board has been proposed, with which it will be possible to obtain digital analogues of images of each mode and automate the process of determining the flow mode. A scheme is proposed for improving the oil metering unit by adding a module for calculating the filling factor. It is planned to conduct a full-scale experiment and assemble a control board and a stand, also proposed in the sketch format in this article.

Keywords:

Computer vision model , Two-phase flow , Gravity flow section , Multiclass classification , Multiphase fluids , Oil pipeline