Deep learning-based model for predicting physical fields during frozen startup of high-temperature heat pipes

High-Temperature Heat Pipes (HTHPs) are widely used due to their excellent thermal performance and passive characteristics. However, their complex frozen startup process presents significant challenges for numerical simulations, particularly regarding efficiency and accuracy. This study introduces a Convolutional Neural Network (CNN) framework to develop an end-to-end model that predicts and analyzes the physical fields of HTHPs based on operating parameters, enabling rapid and accurate predictions of the frozen startup process. A large-scale dataset encompassing various operating conditions was generated through numerical simulations to train the CNN model. Convergence analysis results indicated that a training size of 1.0 and a network depth of 4 layers are the optimal parameters for the model. The CNN model accurately predicted the physical fields, achieving mean absolute errors of 0.41 K for temperature, 5.12 × 10⁻⁵ m/s for axial velocity, 3.24 × 10⁻⁶ m/s for radial velocity, and 23.53 Pa for pressure. Additionally, the model demonstrated a prediction speed nearly four orders of magnitude faster than traditional Computational Fluid Dynamics (CFD) methods. It also accurately predicted the wall temperature of HTHPs, with a mean absolute error of only 0.47 K. This study highlights the potential of deep learning for advancing HTHP analysis.