Efficient inverse reconstruction of internal flow fields in rocket engine combustion chambers based on diffusion models

The internal flow field of a rocket engine combustion chamber is crucial for performance evaluation and design optimization. Traditional numerical simulations suffer from long computation times and low inversion efficiency and are difficult to apply in the real engine design and optimization. This study employed the data-driven reconstruction method for the combustion chamber's internal flow field based on measurable data from the outside. The present study is based on a 490 N attitude and orbit control engine as the test case, and a multi-condition dataset was generated through high-precision simulations. A diffusion model-based neural network was then trained, where measurable data guided the generation process to achieve fast and accurate flow field inversion. The results show that the proposed method achieves an average error of only 2.13% on the validation set, with inversion speed three orders of magnitude faster than conventional simulations. The effects of observation point density and noise intensity were also analyzed: as the number of observation points decreased from 80% to 20%, the average prediction error increased linearly from 2.13% to 4.03%. When the maximum signal-to-noise ratio varied from 40 to 10 dB, the error first decreased to 1.81% and then rose to 3.59%, demonstrating strong robustness against noisy data.