基于域对抗神经网络的双模态燃烧室跨构型燃烧模态识别  

Identification of combustion modes in dual-mode combustors across configurations using domain adversarial neural network

作　　者：宋婷[1] 刘和东 黄玥 陈玉乾 尤延铖 SONG Ting;LIU Hedong;HUANG Yue;CHEN Yuqian;YOU Yancheng(Institute of Artificial Intelligence,Xiamen University,Xiamen 361000,China;School of Aerospace Engineering,Xiamen University,Xiamen 361000,China)

机构地区：[1]厦门大学人工智能研究院,福建厦门361000 [2]厦门大学航空航天学院,福建厦门361000

出　　处：《推进技术》2025年第2期129-144,共16页Journal of Propulsion Technology

摘　　要：双模态冲压发动机燃烧室在宽马赫飞行过程中会呈现不同燃烧模态来保持稳定工作,燃烧模态的准确识别对燃烧室乃至发动机的控制和稳定运行具有重要意义。基于域对抗网络的领域适应策略,提出了一种针对不同构型双模态燃烧室的燃烧模态识别方法。首次将域适应的解决思路应用于燃烧及流体的问题中,以提高双模态燃烧室跨域数据集模态识别模型的泛化性能。通过数值模拟得到三种构型燃烧室的数据集,利用原始构型数据集训练模型,对上凹腔扩张构型和下凹腔扩张构型的数据集验证其泛化性能,并将获得的识别准确率与其他识别方法(包括支持向量机、K近邻、决策树)进行对比分析。研究结果表明:对亚燃模态和超燃模态进行识别,在上凹腔扩张构型和下凹腔扩张构型的密度梯度分布图的验证中分别取得了93.5%和96.3%的准确率,在温度分布图的验证准确率为91.8%和97.1%。本文的方法可以获得更易于识别燃烧模态的图像信息,以获得更高的跨领域数据识别准确率和更好的泛化性能,为发展适用于不同构型双模态燃烧室的燃烧模态识别方法提供了有力支撑。The dual-mode ramjet engine could operate in both ram and scram combustion modes over a wide range of flight speeds.Accurate identification of the combustion modes is of great significance for the control and stable operation of the combustors of the engine.Based on domain adversarial networks incorporating domain adaptation strategies,an identification method of combustion mode for dual-mode combustors with different con⁃figurations is proposed.This approach introduces domain adaptation concepts to address combustion and fluid-re⁃lated challenges,aiming to enhance the generalization of model for combustion mode identification of across di⁃verse datasets of dual-mode combustors.The data comes from numerical simulations and includes three configura⁃tions.The data is divided into training set and test set,the former includes original construction,and the latter in⁃cludes upper-cavity expansion construction and lower-cavity expansion construction.And compare the perfor⁃mance of this method with other recognition methods,including:SVM(Support Vector Machine),kNN(k-Near⁃est Neighbor)and decision tree.The research results show that:for the identification of ram and scram combus⁃tion modes,the accuracy of 93.5%and 96.3%are achieved in the density gradient magnitude contour of the up⁃per-cavity expansion construction and the lower-cavity expansion construction,and the accuracy for temperature contour are 91.8%and 97.1%,respectively.The method proposed in this study can extract image features that are more suitable for recognizing combustion modes,leading to higher cross-domain data recognition accuracy and better generalization.This is crucial for developing combustion mode recognition methods for dual-mode com⁃bustors with different configurations.

关 键 词：双模态燃烧室 燃烧模态识别 域自适应 对抗学习 神经网络 

分 类 号：V235.21[航空宇航科学与技术—航空宇航推进理论与工程]