[SIS14] DX in PM

[16E-SIS14-05] Data-driven Architectured Material Design for Additively Manufactured Heat Sinks

*A. Suzuki¹, H. Nakatani¹, M. Kobashi¹ (1.Nagoya University, Japan)

キーワード：Additive manufacturing, machine learning, architectured material, Optimization, Computational Fluid Dynamics

Additive manufacturing enables the fabrication of architectured materials with complex topology. These materials can be used for heat sinks because of their expanded surface area. It is required to establish a methodology for optimizing architectures. In this study, an attempt was made to optimize the complex-shaped heat sink structure using machine learning and the Voronoi tessellation. The Voronoi tessellation can design complex architectures by determining the arrangement of seed points and has the advantage of uniquely linking the seed point coordinates with architectures. 800 Voronoi structures were designed by randomly changing the coordinates of nine seed points, and their heat transfer and pressure loss were calculated by computational fluid dynamics. The data of seed point coordinates, thickness of solid, and calculated properties was learned to a neural network. Then, the neural network was inversely analyzed using optimization algorithms to optimize the seed point coordinates and solid thickness.