Machine Learning based algorithms for modeling natural convection fluid flow and heat and mass transfer in rectangular cavities filled with non-Newtonian fluids

• Published in: Engineering applications of artificial intelligence 2023-03, Vol.119, p.105750, Article 105750
• Main Authors: Tizakast, Youssef, Kaddiri, Mourad, Lamsaadi, Mohamed, Makayssi, Taoufik
• Format: Article
• Language: English
• Subjects: Artificial Neural Networks; Gradient Boosting; Machine Learning; Natural double-diffusive convection; Non-Newtonian fluids; Random Forests
• ISSN: 0952-1976; 1873-6769
• DOI: 10.1016/j.engappai.2022.105750

The importance of studying double-diffusive fluid flows along with the significance of non-Newtonian fluids have been well recognized in the fluid dynamics field for scientific and practical purposes. However, and given the ever-rising complexity of such non-linear flows, the conventional simulation methods are confronted with problems related to accuracy and required computation time and resources. As a result, the Machine Learning approach qualifies as a promising solution to said limitations. The present study investigates the application of Machine Learning models to study double-diffusive natural convection in rectangular cavities filled with non-Newtonian fluids. The flow is governed by four dimensionless parameters, namely: thermal Rayleigh number RaT, Lewis number Le, buoyancy ratio N, and power-law behavior index n, employed as input features. To precisely model fluid flow, three characteristics are predicted: flow intensity |ψc|, average Nusselt number Nu¯, and average Sherwood number Sh¯ using four machine learning models: Artificial Neural Networks (ANN), Random Forests (RF), Gradient Boosted Decision Trees (GBDT), and Extreme Gradient Boosting (XGBoost). The analysis of investigated models’ fine-tuned architectures using various tools confirms the non-linear complexity of the problem and allows to explore and discuss in details the inner-workings of each model. The ANN predicts the test data with R2=0.9999,R2=0.9996, and R2=0.9999 followed by XGBoost with R2=0.9979,R2=0.9802, and R2=0.9979, and that for |ψc|, Nu¯, and Sh¯, respectively. The study shows the strong and correlated effects of RaT and n on the flow characteristics. The models’ generalization is further examined using fluids with power-law behavior indexes outside the learning range. The present paper validates the choice of Machine Learning approach as a promising solution to model non-Newtonian double-diffusive fluid flows and encourages future works in this direction.