Predicting the effective thermal conductivities of composite materials and porous media by machine learning methods

Composite materials have a wide range of engineering applications, and their effective thermal conductivities are important thermo-physical properties for real applications. The traditional methods to study effective thermal conductivities of composite materials, such as the effective medium theory, the direct solution of heat diffusion equation, or the Boltzmann transport equation, are all based on developing good physical understanding of heat transfer mechanisms in those composite materials. In this work, we take a completely different approach to predict the effective thermal conductivities of composite materials using machine learning methods. With a set of trustable data, the support vector regression (SVR), Gaussian process regression (GPR) and convolution neural network (CNN) are employed to train models that can predict the effective thermal conductivities of composite materials. We find that the models obtained from SVR, GPR, and CNN all have a better performance than the Maxwell-Eucken model and the Bruggeman model in terms of predicting accuracy. Our work demonstrates that machine learning methods are useful tools to fast predict the effective thermal conductivities of composite materials and porous media if the training data set is available. The machine learning approach also has the potential to be generalized and applied to study other physical properties of composite materials.