An investigation of the thermal behavior of constructal theory-based pore-scale porous media by using a combination of computational fluid dynamics and machine learning

Creative Commons License

Mesgarpour M., Sakamatapan K., Dalkılıç A. S., Alizadeh R., Ahn H. S., Wongwises S.

INTERNATIONAL JOURNAL OF HEAT AND MASS TRANSFER, vol.194, no.September, pp.1-18, 2022 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 194 Issue: September
  • Publication Date: 2022
  • Doi Number: 10.1016/j.ijheatmasstransfer.2022.123072
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Communication Abstracts, Computer & Applied Sciences, INSPEC, Metadex, zbMATH, Civil Engineering Abstracts
  • Page Numbers: pp.1-18
  • Yıldız Technical University Affiliated: Yes


The present study investigates the flow pattern and thermal behavior of constructal theory-based pore-scale porous media (CTPSPM) using computational fluid dynamics (CFD) and machine learning. Due to the increasing complexity of the geometry, as the number of pairs (Np) and clusters (Nc) increase, the CFD approach is unable to present the precise results. For the first time, we present a novel hybrid computational method to predict the flow pattern and thermal behavior of pore-scale porous media (PSPM) based on the direct training data set from high-fidelity numerical simulation. Tensors of component velocity, temperature for fluid and solid, and pressure drop will be used to update the training dataset in accordance with the geometry matrix. A tensor-based combination of CFD and block-based machine learning (BBML) is developed for this target due to the complex interface between flow and solid. An elliptical tube is filled with two distinct types of PSPM: CTPSPM and constant-porosity PSPM (CPPSPM) with different porosity ratios (β=0.6, 0.8, 1.2, 1.4). A laminar flow (Re=300, 500, and 800) of Al2O3 nanofluids through an elliptical tube in constant volume fractions (φ" role="presentation" >φ=0.06) is studied in both CFD and BBML methods. The effects of various combinations of CTPSPM pairs and clusters on pressure drop and heat transfer are evaluated using machine learning in over 1200 states of calculation to determine the optimal configuration. Additionally, we evaluated the effect of multiple cores on calculation time in three distinct optimizations (machine learning, surface response method, and genetic algorithm) approaches in CTPSPM for the first time. The results show that the multiblock neural network could reduce the computational cost by up to 70% compared with the regular CFD. In addition, the results show that the constructal theory significantly influences heat transfer in the low Re range. The results of this study could lead to a new understanding of fluid flow due to complex geometry, such as that of a catalyst and membrane.