Predicting the temperature field of thermal cloaks in homogeneous isotropic multilayer materials based on deep learning

Abstract

Thermal cloaks can realize thermal manipulation and have been widely applied in the engineering field. Traditional methods, like the finite element method (FEM) and finite difference method, obtain the temperature of the thermal cloak with time-consuming. A deep learning model is proposed to predict the temperature field in homogeneous isotropic multilayer materials. The FEM is applied to solve the thermal cloak and the training data are obtained. The layer's properties, including different layers, thickness, relative thermal conductivities and distribution modes, of the thermal control or manipulative ability for temperature and heat diffusion are also investigated. The deep learning mode is developed by mapping the parameters layout to the temperature field as an image-to-image regression task and the relationship between the layer's properties and the cloaking performances is intelligently established. The temperature field of the arbitrary layout composite can be acquired almost with no time cost. The temperature field with the largest node numbers of medium temperature by the FEM and the deep learning model, is exhibited and compared. Furthermore, we extend the surrogate model to different cases with different boundary conditions, which can still achieve good prediction results. Owing to the high efficiency and complete independence from the mesh densities, the proposed method offers a novel and robust technique for predicting the temperature field and suggestion for the design of the thermal cloak.