データ駆動型マルチフィデリティトポロジーデザインによる自然対流型ヒートシンクの最適設計

抄録

In recent years, the demand for high-performance natural convection heat sinks has been increasing due to the miniaturization of electronic devices and their wide range of applications. It has become important to design a natural convection heat sink that can exhaust the maximum amount of heat with a compact size. In this study, we apply the framework of data-driven multifidelity topology design (MFTD) for the design problem of natural convection heat sinks. In this method, we initially prepare several promising and diverse solutions and iterate generating new solutions from them while preserving diversity and structural features and simultaneously arranging moderate perturbation. At each iteration, better solutions are selected as the seeds for generating the succeeding set of solutions, and they are arranged into new ones through a deep generative model. It is expected that repeating this operation finally reaches superior solutions without gradient information. We applied this framework to the 2D topology optimization problem. The shape of the heat sinks was optimized for several values of the coefficient of volumetric expansion (β), which is a number that determines the strength of convection. Different optimal structures were obtained by changing β, and they were reasonable. For example, the optimal solutions under strong convection had partial structures with rectifying effects.