The deformation of air bubbles in a liquid flow field is of relevant interest in phenomena such as cavitation, air entrainment, and foaming. In complex situations, this problem cannot be addressed theoretically, while the accuracy of an approach based on Computational Fluid Dynamics (CFD) is often unsatisfactory. In this study, a novel approach to the problem is proposed, based on the combined use of a shadowgraph technique, to obtain experimental data, and some machine learning algorithms to build prediction models. Three models were developed to predict the equivalent diameter and aspect ratio of air bubbles moving near a plunging jet. The models were different in terms of their input variables. Five variants of each model were built, changing the implemented machine learning algorithm: Additive Regression of Decision Stump, Bagging, K-Star, Random Forest and Support Vector Regression. In relation to the prediction of the equivalent diameter, two models provided satisfactory predictions, assessed on the basis of four different evaluation metrics. The third model was slightly less accurate in all its variants. Regarding the forecast of the bubble's aspect ratio, the difference in the input variables of the prediction models shows a greater influence on the accuracy of the results. However, the proposed approach proves to be promising to address complex problems in the study of multi-phase flows.

Deformation of air bubbles near a plunging jet using a machine learning approach

Di Nunno F.;de Marinis G.;Gargano R.;Granata F.
2020

Abstract

The deformation of air bubbles in a liquid flow field is of relevant interest in phenomena such as cavitation, air entrainment, and foaming. In complex situations, this problem cannot be addressed theoretically, while the accuracy of an approach based on Computational Fluid Dynamics (CFD) is often unsatisfactory. In this study, a novel approach to the problem is proposed, based on the combined use of a shadowgraph technique, to obtain experimental data, and some machine learning algorithms to build prediction models. Three models were developed to predict the equivalent diameter and aspect ratio of air bubbles moving near a plunging jet. The models were different in terms of their input variables. Five variants of each model were built, changing the implemented machine learning algorithm: Additive Regression of Decision Stump, Bagging, K-Star, Random Forest and Support Vector Regression. In relation to the prediction of the equivalent diameter, two models provided satisfactory predictions, assessed on the basis of four different evaluation metrics. The third model was slightly less accurate in all its variants. Regarding the forecast of the bubble's aspect ratio, the difference in the input variables of the prediction models shows a greater influence on the accuracy of the results. However, the proposed approach proves to be promising to address complex problems in the study of multi-phase flows.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11580/77647
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 7
social impact