Physics-Informed Neural Networks for Heat Transfer and Fluid Flow problems

---

Aygün A., Ata O., Karakuş A.

15th International Conference on Computational Heat and Mass Transfer, Antalya, Turkey, 20 - 22 May 2025, (Unpublished)

• Publication Type: Conference Paper / Unpublished
• City: Antalya
• Country: Turkey
• Middle East Technical University Affiliated: Yes

Abstract

Physics-informed neural networks (PINN) have emerged as a viable alternative to the classical
numerical methods while solving partial differential equations. Their ability to obtain the solution
without generating any mesh or using a discretization method without any labeled data have drawn
attention in recent years. PINNs struggle to obtain the correct solution when the problem has multiscale
behavior or highly nonlinear behavior. In this talk, we will show the solutions for thermal convection
problems in different flow regimes. We will show the results for different thermal convection problems
and show the applicability of PINNs on heat transfer and fluid flow problems.