Physics-Informed Neural Networks for Forward and Inverse Multiphysics Heat Transfer Problems

Heat transfer modeling plays a key role in different scientific and engineering fields. However, often parameters are not fully known leading to ill-defined heat transfer problems, which cannot be easily tackled with traditional computational methods. Additionally, the coupling with nonlinear fluid flow phenomena further aggravates these issues. One approach for solving these kinds of problems is physics informed neural networks (PINN), which provides a hybrid data-driven and physics-based solution to ill-posed problems. In this talk, we present different applications of PINN in Multiphysics convective heat transfer problems. First, heat transfer in fins where conduction in solid is coupled with convection in fluid is considered. We quantify the base temperature and thermal conductivity of the solid using sparse temperature measurements in the fluid domain. We present the sensitivity of the results to sparse sensor placement strategies. Additionally, we study the challenges of getting a unique solution in inverse problems and propose remedies to constrain the solution space based on prior physical knowledge. Finally, we present forward and inverse modeling of convective heat transfer in a rotating fluid-saturated porous medium.