Machine learning boosting the discovery of porous metamaterials with an abnormal thermal transport property

Abstract

Normally, the introduction of porous structures into materials can tune their thermal conductivity, showing great applications in thermal management and thermoelectric energy harvesting. However, the ability of disorder changing the thermal conductivity of porous materials has seldom been explored. In this work, we show that an introduction of disorder into the macroscopic porous materials with a certain porosity can lead to a desired effective thermal conductivity over a large range, where an abnormal enhancement of ∼7.9% and a normal reduction of ∼44% at room temperature are predicted by the machine-learning-optimized algorithm. All of these theoretical calculation results are further verified by our experiments performed in the current work by using the steady-state thermal flux method. Moreover, when these periodic units are artificially connected, a structural anisotropy up to 40 is achieved, which can be further used to adjust the direction of the thermal flux in a well-controlled way. Our work provides an efficient and convenient approach for designing high-performance porous materials with specific thermal conductivity and high structural anisotropy for various applications in thermal management.