PINN-DTC: Adaptive partitioning of physics-informed neural network for designing thermal cloak

In this study, an adaptive partitioning of physics-informed neural network is proposed for designing thermal cloak (PINN-DTC). In order to realize the intelligent inverse design of thermal cloak structures with different shapes, an adaptive hierarchical inverse design PINN solution framework is established. The application of improved PINN structure for inverse design is somewhat free from the limitation of the background thermal conductivity compared to the thermal cloak research based on the equivalent transformation theory. Consequently, a thermal cloak with considerable functionality can be designed by applying a given material within a certain range. In particular, the presented method avoids the necessity of multiple subdivision and mesh redivision, and simultaneously permits the acquisition of the corresponding full-field temperature and the number of layers of the thermal cloak. Finally, comparing finite element results, the structure of the thermal cloak obtained by PINN-DTC showed superior thermal protection performance.