Application of ANN to the water-lubricated flow of non-conventional crude

Abstract

An annular water-film lubricates an oil-core by functioning as a barrier between the core and the pipe wall in the water-lubricated flow of non-conventional viscous oil. A dependable model for appraising the frictional energy losses in such a core annular flow system is necessary to ensure its widespread implementation in the industry. In the current study, the modeling was conducted using an artificial neural network (ANN) based on 223 data sets. Seven input variables applied in the current ANN model are pipe diameter, average velocity, fluid properties, and water fraction. The optimum architecture was identified as a feed-forward neural network with backpropagation technique involving two hidden layers, each of which was consisted of 20 neurons or nodes. Comparative statistical analysis demonstrated promising accuracy of the current model, the coefficient of determination was 0.992, and the root mean square error was 0.111. In addition to validating the model, the relative significance of the input parameters was evaluated with a sensitivity analysis.

Keywords:

• Core annular flow
• feed-forward back-propagation neural network
• heavy oil
• pipeline transportation
• pressure loss

Acknowledgments

We thank Pipe Flow Technology Centre, Saskatchewan Research Council, Regina, SK, Canada, and Flow Process Assurance Research Laboratory, Cranfield University, Cranfield, the UK, for their high-quality data used for the current analysis.

Additional information

Funding

The authors acknowledge the Deanship of Scientific Research at King Faisal University (KFU) in Al-Ahsa for the financial support under the Research Group Support Track (grant no. 17122013).