Title:	AI driven design of complex structured materials for wearable cooling
DNr:	Berzelius-2024-450
Project Type:	LiU Berzelius
Principal Investigator:	Ricardo Vinuesa <rvinuesa@mech.kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2024-11-12 – 2025-06-01
Classification:	20301
Homepage:	https://www.vinuesalab.com/
Keywords:

Abstract

Air conditioning currently accounts for 10 % of global electricity use. Optimizing the cooling technology of vehicles, houses or individuals is a key area for reducing emission and climate change. In this project we try to develop high-power wearable materials based on multi-level kirigami materials (MKMs), which combines the unparalleled performance ofconventional metals with the unique electromechanical properties of emerging stretchable composites. As MKMs are multi-layeredstructured materials, their design space is huge, and their full potential cannot be explored by manualdesign approaches. By exploring Artificial intelligence (AI) driven algorithms in combination with Multiphysics modelling of MKMs, thecomputational cost for structural optimization will be minimized, which enables fast designs of highlycomplex structured materials. In particular, we will use Bayesian optimization (BO) with Gaussian process regression (GPR) to explore optimal structure in the design space. The bestperforming AI designed materials will be fabricated and characterized to verify the process. The main goalsof the project are to: 1) Develop a parameterizable multiphysics model that can be used for AI driven design. 2) Explore AI driven approaches to achieve optimal designs with minimal computational cost. 3) Fabricate, characterize, and verify the best performing generated designs. Altogether the project will establish new tools and methods for efficient material and device designsbased on AI driven methods. This will reduce time and improve performance in material design. The first goal of the project has been already achieved and the framework developed has been tested in local computers for several toy cases with positive results. This framework joins in a modular manner the three main pillars of the project: BO-GPR algorithm, kirigami geometry generation and the multiphysics model. In order to achieve the second goal of the project, we need to scale up the problem to a more complex BO-GPR process due to the higher number of parameters of our real case. Therefore, the use of HPC is essential to run the framework.