熱流体問題を対象としたマルチスケールトポロジー最適化手法の開発

抄録

In this presentation, we discuss how to optimize both microstructure and macrostructure in the topology optimization of 3D thermal fluid. First, the 3D thermal fluid model is approximated by a 2D thermal fluid field. Next, a proxy model of microstructural properties is constructed by learning the relationship between microstructure geometry, permeability, thermal conductivity, and heat transfer coefficient through off-line numerical analysis. The macrostructure optimization is then formulated as a microstructure selection problem from the surrogate model. Finally, a simple numerical example confirms the validity of the method. In this study, as a specific example, topology optimization is performed with two design variables on a two-dimensional model of a 3D forced air-cooled heat sink with prismatic fins. The properties of the 3D prismatic fins are analyzed individually in advance and learned offline using an RBF network so that they can be used in the two-dimensional topology optimization, which requires differentiation.