Duct optimization using CFD software `ANSYS Fluent Adjoint Solver'

Abstract

The ordinary duct development procedure in Climate sector at VVC relies on the manual evaluation of the CFD results and the identi cation of the proper geometry changes in order to optimize the ducts shape. A newly introduced methodology in automotive sector, the adjoint procedure, could reduce the duration of a duct development process through an automatic iterative procedure. Having computed the ow results, the adjoint obtains the geometry's sensitivity eld which depicts the potential changes in the domain with respect to the cost function. Then a morphing tool utilizes that data and properly modi es the geometry. The current thesis is focused on experiencing and demonstrating the merits of the adjoint solver aiming to minimize the cost function, the pressure drop, over a duct domain. The discrete adjoint method was applied to di erent geometries from the Climate sector examining the prerequisites for extended adjoint use. Through those geometries both the adjoint solver and the morphing tool limitations were discovered. The limitations are mostly related to the mesh, the adjoint discretization scheme as well as the boundary constraints. During the thesis, the limitations were identi ed and properly examined, proposing solutions. An adequate adjoint solution was obtained when using a coarse Polyhedral mesh with cell skewness around 0.8, a low discretization scheme and a simpli ed geometry. Even if some compromises were required in whole procedure, the adjoint has managed to handle all the given geometries. Moreover, new optimized domains with respect to the cost function, the pressure drop minimization,were developed. The result of the thesis was a guide of how to overcome adjoint constraints as well as an analytical illustration of the adjoint results. In some cases, the pressure drop was minimized by 60% resulting in a smooth pressure distribution over the given domain. However, a compromise in the morphing tool leaves room for improvement on the suggested methodology.