Project Overview
When gases and liquids flow simultaneously in a pipe, the phases can distribute themselves in various configurations depending on many variables. There is no convention in the flow patterns number in two-phase flow due to overlapping and characterization subjectivity, especially in transition zones. Flow patterns change according to different physical properties and characteristics of the medium. The most notable physical properties of liquid and gas are superficial velocity, viscosity, density, and surface tension. Furthermore, the influence properties of the medium are the angle of inclination and the diameter of the pipe. Given the considerable number of variables, determining the flow pattern is not a trivial task, however, its accurate classification is essential for the proper design and planning of systems with pipelines.
In the study of two-phase flows, the correct estimation is vital due to its relationship with the design variables such as phase holdup, pressure drop, and chemical reaction rate. The present research explores multiple artificial intelligence models for the classification of flow patterns since there is no universal model to perform this task. However, over time, progress has been made in defining the flow pattern, but there is still no theory that accurately characterize flow regimes.
Results Highlight
Results presented in this section correspond to experiments performed during the development of this project. For more information refer to the scientific paper [1] linked below.
| Model | Accuracy | Cross-validation | Run Time [sec] |
|---|---|---|---|
| Extra Trees | 0.970 | 0.958 ± 0.013 | 4.304 |
| Random Forest | 0.951 | 0.949 ± 0.007 | 9.269 |
| Support Vector Machine | 0.948 | 0.934 ± 0.012 | 32.02 |
| Gradient Boosting | 0.944 | 0.947 ± 0.011 | 56.61 |
| Decision Tree | 0.934 | 0.919 ± 0.008 | 0.187 |
| K-Nearest Neighbors | 0.897 | 0.887 ± 0.011 | 0.325 |
| Quadratic Discriminant Analysis | 0.713 | 0.684 ± 0.028 | 0.119 |
| Gaussian Naive Bayes | 0.695 | 0.674 ± 0.021 | 0.097 |
| Ada Boost | 0.670 | 0.664 ± 0.017 | 0.724 |
Products
[1] H. B. Arteaga-Arteaga et al., "Machine learning applications to predict two-phase flow patterns," PeerJ Comp. Sci., Nov. 2021, doi: 10.7717/peerj-cs.798.