Predicting the temperature field of composite materials under a heat source using deep learning

Composite materials have different effective material properties with different combinations of material components. However, as the number of all possible combinations is astronomical, it appears complex to establish a clear relationship between the temperature field and layout of composite materials under the heat source. Traditional methods, like the finite element method (FEM) and finite difference method, can be time-consuming and costly for such a large task. To overcome these problems, a deep learning (DL) based surrogate model is developed by mapping the composite layout to temperature field as an image-to-image regression task. This surrogate model can accurately predict the temperature field and its ranking for a two-phase composite under a heat source. Furthermore, we discuss the effect of data size on the surrogate model, and extend the surrogate model to different cases which still achieve good prediction results. The proposed surrogate model can significantly reduce time consumption, and accurately predict the temperature fields of composite materials with arbitrary layouts.