基于深度卷积神经网络对CH自由基分布预测  

Prediction of CH Radical Distribution Based on Deep Convolutional Neural Network

作　　者：潘玮 金武 汤朝伟 李夏飞 李建中[1] 袁丽 Pan Wei;Jin Wu;Tang Chaowei;Li Xiafei;Li Jianzhong;Yuan Li(College of Energy and Power,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;College of Defense Engineering,Army Engineering University of PLA,Nanjing 210007,China)

机构地区：[1]南京航空航天大学能源与动力学院,南京210016 [2]陆军工程大学国防工程学院,南京210007

出　　处：《燃烧科学与技术》2023年第3期337-346,共10页Journal of Combustion Science and Technology

基　　金：国家自然科学基金资助项目(51906106);江苏省自然科学基金资助项目(BK20190423);先进航空动力创新工作站资助项目(HKCX2020-02-027);国家科技重大专项资助项目(2019-Ⅲ-0014-0058).

摘　　要：针对低当量比条件下CH自由基浓度较低、存活时间短、测量得到的PLIF信号信噪比较低的问题,采用深度卷积神经网络,通过OH-PLIF测量结果对CH自由基分布进行预测.研究过程中,首先采用高当量比条件下获得的OH-CH-PLIF同步测量图片作为训练数据对神经网络进行训练,并验证优化得到的神经网络预测结果,之后将模型推广到低当量比CH自由基分布的预测中,实现了基于OH-PLIF对CH自由基分布的智能诊断.对比低当量比CH-PLIF数据和预测结果,在CH-PLIF数据较为模糊的情况下,预测结果依然能够十分清晰.由此可以认为,使用深度学习的方法能够弥补现有实验设备的不足,为燃烧诊断提供数据支持.Aiming at the problems of low concentration,short survival time and low signal-to-noise ratio of CH-PLIF signal under the condition of low equivalence ratio,deep convolutional neural network was utilized to pre-dict the distribution of CH radicals from OH-PLIF measurement.Firstly,the OH-CH-PLIF simultaneous meas-urements obtained under the condition of high equivalence ratio are used to train and verify the neural network.Then,the model is extended to the prediction of CH radical distribution at low equivalence ratio,and the intelli-gent diagnosis of CH radical distribution based on OH-PLIF is realized.The CH-PLIF data are compared with the predicted results by CNN at low equivalence ratio,and the result shows that the predicted CH radical distribution is as good as expected.The intelligent prediction of CH from OH-PLIF through deep learning method can make up for the shortage of existing experimental equipment and provide data support for combustion diagnosis.

关 键 词：卷积神经网络 燃烧诊断 预混火焰 火焰结构 PLIF 

分 类 号：TK11[动力工程及工程热物理—热能工程]