Inverse estimation of heat flux under forced convection conjugate heat transfer in a vertical channel fully filled with metal foam

Abstract

In this work, for the first time, a heat flux at the boundary is estimated for a conjugate heat transfer under forced convection in the presence of high porosity metal foams. For the forward problem a vertical channel experimental set up reported in the literature is considered. The metal foam placed in the vertical channel is subjected to constant heat flux through aluminum plate and airflow of various velocities is passed through vertical channel for removal of heat from the high porosity metal foam placed in the vertical channel. Six different velocities are considered and the required temperature distribution of the aluminum plate is obtained by solving Darcy extended Forchheimer and Local Thermal non-equilibrium models for metal foams. The forward problem, created using computational fluid dynamics in ANSYS-FLUENT, is substituted with Neural Network for faster computation of the forward problem. The maximum errors between the computational fluid dynamics and Artificial Neural Network models for the heat flux values of 466.66, 666.66 and 1133.3 W/m 2 are found to be 0.086, 0.043, 0.092 respectively. The heat flux to the forward problem is treated as unknown and the same is estimated using an inverse method that couples Particle Swarm Optimization with Bayesian framework. The result of inverse estimation of exact temperature data shows that for a heat flux of 1266.64 W/m 2 the error is found to be 1.6e−4%. Similarly, for the noise added temperature data, the absolute % error in heat flux of 599.985, 733.315 and 1266.635 W/m 2 is 4.80e−2%, 2.20e−2%, 2.30e−2% respectively. • Inverse estimation of boundary heat flux using particle swarm optimization method. • Forward problem is a conjugate heat transfer problem in a vertical channel. • Darcy–Forchheimer, Local Thermal Non-equilibrium models for modelling porous medium. • Different Artificial Neural Network models to facilitate faster computations. • Particle Swarm Optimization combined Bayesian framework to quantify modeling error.