(422c) Comparative Study of Erosion Prediction in Elbows Using Machine Learning and CFD

Understanding and developing an efficient approach for estimating erosion damage in pipeline facilities remains an industrial problem wherever solid particles are transported either as pure solids or in multiphase flows [1, 2, 3]. Presently, experimental and numerical methods are the most widely used for predicting erosion. Computational Fluid Dynamics (CFD) studies are mostly used to complement experimental erosion studies [1, 4]. Reproducing experimental results by CFD requires a large number of underlying assumption, several factors such as mesh properties, force combinations, collisions, coupling, solid particle and fluid properties, erosion equations etc, need to be investigated, carefully selected, and optimised in order to achieve an approximate predictions. This task can be time-consuming, computationally expensive, and may contribute to uncertainties of predictions. Sometimes experimental observations are not easily reproducible by CFD, which reinforces the need for alternative, data-driven methods for erosion prediction that are simplified, easy to use but reliable

In this study, we investigate the prediction of maximum erosion in a 76.2mm diameter elbow geometry by machine learning (ML) and compare the results with CFD simulations reported by Zamani [4], Vieira [5], and Banakermani [6] for gas-solid system. ML was performed using the Random Forest (RF) and Decision Tree (DT) regressions methods, with a 50-50% data split between the training and testing datasets. Six variables were studied and used for erosion prediction, which include gas superficial velocity, geometry orientation, material hardness, pipe diameter, volume fractions, and particle size.

The R² values obtained from erosion prediction on the testing data after fitting of the training dataset are 0.954 and 0.965 for DT and RF (see Figure 1), respectively. The fitted algorithm was tested using six cases, and the average deviation of ML predictions from the experimental data obtained are 25.33% and 35.85% for DT and RF, respectively, compared to 54.27% for Vieira, 49.86% for Zamani, 45.31% for Banakermani (see Figure 2). These results show that ML can provide an improved prediction compared to CFD, and the prediction for each case by ML can be improved and optimised with the use of tuning parameters. Ranking of the parameters indicates that gas superficial velocity and volume fractions are the most dominant variables affecting erosion.

Our results demonstrate that ML is a promising approach for erosion prediction, which also has the advantage of reducing CFD uncertainties due to mesh quality, erosion equations, etc. It is also important to note, however, that the ML prediction accuracy will significantly depend on the robustness and accuracy of the data source utilised.

References
[1] C. B. Solnordal, C. Y. Wong, and J. Boulanger, “An experimental and numerical analysis of erosion caused by sand pneumatically conveyed through a standard pipe elbow,” Wear, vol. 336-337, pp. 43–57, aug 2015.

[2] “CFD study of Jet Impingement Test erosion using Ansys Fluent® and OpenFOAM®,” Computer Physics Communications, vol. 197, pp. 88–95, dec 2015. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0010465515002945

[3] S. A. Shirazi and B. S. McLaury, “Predicting Solid Particle Erosion in Multiphase Flow: Challenges and Success Stories (Keynote Paper),” in Volume 1: Symposia, Parts A, B and C. ASMEDC, jan 2009, pp. 637–637. [Online]. Available: https://asmedigitalcollection.asme.org/FEDSM/proceedings/FEDSM2009/43727/637/346665

[4] M. Zamani, S. Seddighi, and H. R. Nazif, “Erosion of natural gas elbows due to rotating particles in turbulent gas-solid flow,” Journal of Natural Gas Science and Engineering, vol. 40, pp. 91–113, 2017. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1875510017300434

[5] “Experimental and computational study of erosion in elbows due to sand particles in air flow,” Powder Technology, vol. 288, pp. 339–353, jan 2016.

[6] M. R. Banakermani, H. Naderan, and M. Saffar-Avval, “An investigation of erosion prediction for 15° to 90° elbows by numerical simulation of gas- solid flow,” Powder Technology, vol. 334, pp. 9–26, 2018. [Online]. Available: https://doi.org/10.1016/j.powtec.2018.04.033