【摘要】 实现航空发动机燃烧室内火焰锋面的实时准确测量对于燃烧器优化设计、燃烧过程优化具有重要意义。基于代数重建算法的化学发光层析技术(CTC)可准确实现火焰锋面测量，但存在重建计算量大、耗时长的问题。本文提出了一种基于深度学习的旋流火焰锋面重建方法。首先搭建了紫外相机阵列CTC系统，并利用联合代数重建算法(SART)算法重建了不同工况下的三维旋流火焰锋面，进而构建了化学发光投影图像–火焰锋面数据集。在此基础上开展了基于卷积神经网络(CNN)的三维旋流火焰锋面重建实验研究，并评价了CNN网络模型的重建性能。结果表明：训练后的CNN网络重建结果与SART算法重建结果的结构相似度平均值为0.89，相关系数平均值为0.86，证实两种方法的重建结果具有较高的结构相似性；CNN网络的火焰锋面重建时间为1.8 s，仅为SART算法重建时间的1.9%。更多还原

【Abstract】 Real-time and accurate measurement of the flame front in the aeroengine combustion chamber is of great significance for the optimization of the combustion process and the improvement of burner design. Chemiluminescence chromatography(CTC) based on the simultaneous algebraic reconstruction algorithm(SART) can accurately achieve the reconstruction of flame front measurement, but it has the problems of large reconstruction calculations and long-time consumption. In this paper, a deep learning-based swirl flame front reconstruction method is proposed. First, a UV camera array CTC system are designed and built, and the SART algorithm is further used to reconstruct the three-dimensional swirling flame front under different working conditions, and then a chemiluminescence projection image-flame front data set is built. On this basis, an experimental study of three-dimensional swirling flame front reconstruction based on convolutional neural network(CNN) is carried out, and the reconstruction performance of the CNN network model is evaluated.Results demonstrated that the average structural similarity between the CNN network reconstruction results and the SART algorithm reconstruction results is 0.89, and their correlation coefficient average is 0.86, validating the feasibility of the deep learning-based swirl flame front reconstruction method. Furthermore, the CNN model achieves a single flame sample reconstruction time of 1.8seconds, which is only 1.9% of the reconstruction time required by the SART algorithm.更多还原