A Data-Driven Method for Design of Foam Structures with High Effective Thermal Conductivity

Abstract

The foam-like structures have been shown to facilitate thermal transport efficiency by forming 3D-network heat conduction pathways, which shows broad applications in thermal management. However, the design of a foam structure with desired thermal transport property is tricky because of the complicated relationship of the property concerning the various parameters. Here, we combine high-throughput numerical simulations and machine learning methods to systematically investigate the quantitative relationship between the structural characteristics and thermal transport properties of a variety of foam structures. The so-called explicit jump immersed interface approach that solves the transport equations by combing the fast Fourier transform (FFT) and GiGGStab methods is used to perform high-throughput calculations of over 700 different foam structures to generate a database. The high accuracy and efficiency of the approach ensure the high quality of the database, which provides a foundation to build robust machine learning models. 66 out of 131 structural descriptors are selected as the inputs of 6 machine learning models, including the DTR, GBR, GPR, RFR, SVR, and ANN, among which we obtain the ANN with the best performance. For the importance of descriptors, VolumeFraction2 and porosity have the dominant impact on the thermal conductivity, but are far from being decisive, which evidences the complexity of the structure-property relationship. The present work provides versatile tools for the design of highly conductive foams, which could facilitate thermal management in broad applications such as solar receivers and batteries, preservation of biological tissues, and so on.

Keywords: Foam, thermal transport, material design