11 - Operations and Sustainment
PARAMETER INVERSION ALGORITHM FOR AERO-ENGINE γ-PHOTON FLOW FIELD IMAGES BASED ON UNET-GAN ARCHITECTURE
L.Y. Hou¹, J.T. Liu¹, Y.L. Song¹, H. Xiao¹; ¹Nanjing University of Aeronautics and Astronautics, China
γ-photon emission tomography offers a promising non-intrusive solution for monitoring the internal flow fields of aero-engine nozzles. However, the stochastic nature of photon detection results in sparse grayscale density maps that lack direct thermodynamic insights. To address this, this paper proposes a high-precision parameter inversion algorithm based on a Conditional Generative Adversarial Network (cGAN) architecture. The framework employs a U-Net generator with a ResNet101 backbone to map probabilistic photon distributions to continuous multi-channel flow parameters. A novel Physical Aware Enhancement (PAE) module, integrating multi-scale dilated convolutions and dual-attention mechanisms, is embedded to suppress statistical noise and enforce gradient continuity, effectively eliminating non-physical artifacts. Experimental validation on the SW60B turbojet engine demonstrates that the proposed method achieves a Peak Signal-to-Noise Ratio (PSNR) of 33.86 dB and a Structural Similarity Index (SSIM) of 0.86. Comparative analysis confirms its superiority over conventional generative models in maintaining the structural integrity and physical fidelity of the reconstructed flow fields.
DOI: 10.71945/icas2026_0264