Improved Thermal Dome Structure Optimization Design Based on Neural Network Algorithm

This research introduces an optimized design methodology for thermal domes, which realize thermally invisible. Unlike traditional thermal cloaks that demand exceptional material properties and are environmentally inflexible, the proposed thermal domes employ an open structure for ease of installation and reconfiguration, adaptability, and internal heat source accommodation. The study utilizes a neural network model trained with the Levenberg-Marquardt optimization algorithm to correlate the thermal conductivity of the dome's layers with its performance. The model training involves a dataset derived from finite element simulations, with a focus on optimizing the minimal value of a defined evaluation function to identify the optimal layer configuration. Results show that the neural network model achieves high predictive accuracy, as well as the following validation through the finite element simulation. The combination of the trained neural networks and the additional design samples generated rapidly through Latin hypercube sampling enable efficient and swift evaluation of thermal dome configurations. The cloaking functionality of the neural network-designed thermal domes is validated through comparative temperature field and isotherm distribution analyses. In conclusion, the integration of theoretical analysis with neural network modeling has led to the development of a robust framework for intelligent thermal dome design. The approach not only streamlines the design process but also expands the potential design schemes for thermal domes, offering a versatile solution for advanced thermal device design across various applications.