A novel PINNs based surrogate model for multi-objective reliability-based design optimization of airfoil-shaped printed circuit heat exchangers

Printed circuit heat exchangers (PCHE) are leading solutions for intermediate heat exchangers in sodium-cooled fast reactors. Since higher heat transfer and lower flow consumption are both required, the design of PCHE is a complicated multi-objective optimization problem. Traditional optimization methods always give the objective parameters, which cannot provide accurate physical field distributions. This study proposes a novel physics-informed neural-networks (PINNs) based surrogate model combined with NSGA-II approach to address the multi-objective design optimization for airfoil-shaped fins of PCHE and further provide accurate physical field distributions. The PINNs-based surrogate model of flow distributions with Navier-stokes and energy terms is first established and thermal-hydraulic parameters including heat transfer coefficient, friction factor, max velocity, and pressure drop can obtain from physical field distributions. The surrogate model achieves normalized absolute error less than 10.103 % in physical field distributions and relative error less than 2.799 % in thermal-hydraulic parameters. Additionally, the first-order second-moment reliability analysis approach combined with NSGA-II is developed to prevent the impact of excessive flow velocity and pressure drop on airfoil fins, which effectively generates a set of Pareto frontier solutions. This work highlights the application of PINNs as surrogate model of multi-objection optimization in airfoil fins geometry structure parameters selections for PCHE.