超燃冲压发动机燃烧室流场超分辨率重建

Super resolution reconstruction of flow field in scramjet combustor

关键词:  超燃冲压发动机  燃烧室  双三次插值  超分辨率  生成对抗网络

Key words:  Scramjet  Combustor  Bicubic  Super-resolution  Generative adversarial network

摘要:

超声速燃烧室受限空间内复杂流场波系结构的获取受到光学测量装置精度的制约。为提升流场时空分辨率特征，本文应用中国空气动力研究与发展中心地面脉冲燃烧风洞获取的试验数据，在发动机入口马赫数2.5的条件下，构建了6种不同当量比下基于压力数据重构的燃烧室流场低分辨率图像数据集，研究了三种提高图像分辨率的方法来提升超燃冲压发动机燃烧室流场重构图像的分辨率。结果表明，本文所提出的流场超分辨率稠密网络（Flow-field Super-Resolution Dense Network， FSRDN）、流场超分辨率生成对抗网络（Flow-field Super-Resolution Generative Adversarial Network， FSRGAN）、传统的双三次插值法（Bicubic interpolation，Bicubic）对流场图像分辨率都提高了4²倍。FSRDN网络所得流场图像结果的峰值信噪比（Peak Signal-to-Noise Ratio，PSNR）、相关性系数（Correlation coefficient，CORR）、感知指数（Perceptual Index，PI）指标均优于双三次插值法，但实际图像存在过于平滑的现象。FSRGAN网络所得流场结果消除了图像平滑现象，使流场图像的细节更加丰富，大幅度优化了PI指标，对燃烧室内的剪切层、斜激波、分离激波等主要波系结构的清晰度有了极大的增强作用。

Abstract:

The acquisition of complex flow field wave structures in a confined space of a supersonic combustion chamber is limited by the accuracy of optical measurement devices. To improve the spatiotemporal resolution characteristics of the flow field，in this paper， based on the test data obtained from the China Aerodynamics Research and Development Center ground pulse combustion wind tunnel， under the condition of engine inlet Mach 2.5， six low resolution image data sets of combustor flow field reconstruction based on pressure data under different equivalence ratios are constructed， and three methods to improve image resolution are studied to improve the resolution of scramjet combustor flow field reconstruction images. The results show that the proposed flow field super-resolution dense network （FSRDN）， flow field super-resolution generative adversarial network （FSRGAN） and the traditional bicubic interpolation（Bicubic） method have improved the flow field image resolution by 4² times. The peak signal-to-noise ratio （PSNR）， correlation coefficient （CORR） and perceptual index （PI） of the flow field image results obtained by FSRDN network are better than those obtained by bicubic interpolation method， but the actual image is too smooth. The flow field results obtained by FSRGAN network eliminate the smooth image，enrich the details of the flow field image，greatly reduce the PI index， and greatly enhance the clarity of the main wave structures such as shear layer， oblique shock wave and separation shock wave in the combustor.