Tese: Sensitivity Analysis for Design and Control of Thermal-Fluid Systems Using Computational Fluid Dynamics-Based Adjoint Method
Aluno(a) : Javier Aliaga RiveraOrientador(a): Marcos Sebastião Gomes
Área de Concentração: Termociências
Data: 31/05/2022
Resumo: In this work, we develop the sensitivity analysis for design and control of thermal-fluid systems using computational fluid dynamics-based adjoint method. The principal motivation for this work comes from problems related to the optimal design of valves and flow control, as well as for devices subject to heat and mass transport phenomena. To illustrate the method, the fluid is modeled with the incompressible Navier-Stokes-Brinkman equation and the heat transfer is modeled by a convection-diffusion equation in steady-state. The system of equations is discretized using both the Finite Element Method and the Finite Volume Method, implemented in Matlab and OpenFOAM respectively. To perform the design and control of the thermal-fluid system, we consider three kinds of actuators: first, the components of velocity at inlet boundaries; second, the position of sources actuating as flow blockage; and finally, the pseudo-density that determines the material distribution in the computational domain. The sensitivity analysis begins by comparison of the continuous and discrete adjoint variables for specific cost functions. Thereafter, we checked the sensitivities by comparison with sensitivities obtained by the finite difference method, obtaining good agreement in all cases, and proving the robustness of the method. Next, two case studies are presented. The optimal location and magnitude of discrete sources in the domain, to achieve a determined steady-state condition; and the topology optimization of a forced convective heat exchanger, where the goal is to maximize the heat transfer while constraining the power dissipation across the duct. Furthermore, several numerical studies have been carried out for different configurations and operating conditions. The different cases demonstrate the method's capability by applying it to the optimal design with little computational effort.