Abstract
Efficient thermal management holds pivotal significance in the rapidly evolving realm of energy-dense applications, notably in high-performance computing, where the push toward miniaturization encounters limitations in controlling heat flux within confined spaces. The advent of metamaterials presents an energy-free opportunity to transform thermal control, essential to boosting system efficiency. This work marks a step toward integrating AI in the design and optimization of compact systems, such as printed circuit boards, solar collectors, and power electronics. We present an automated generative design exploration framework for thermal metamaterials capable of effectively controlling heat transfer to achieve multiple objectives: unperturbed temperature distribution (cloaking) and targeted heat flux insulation and concentration. The architected structures developed leverage the optimization of unit cell arrangements with contrasting thermal properties to exhibit characteristics not commonly occurring in nature. This framework employs a variational autoencoder (VAE), trained on bio-inspired leaf and mushroom microstructures, to assemble bi-material unit cells on a flexible lattice for arbitrary domains. By operating a genetic optimizer within the reduced design space of the VAE, we develop these assemblies to address adversarial temperature and heat flux objectives using a finite element solver. We demonstrate the modularity and scalability of the framework, outputting multi-objective metamaterial solutions for user-specified thermal requirements within a total run-time of 5 min only.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call