融合空间迭代与深度学习的气膜叠加模型  

A film superposition model integrating spatial iteration and deep learning

作　　者：刘星 楼健 余鸿乾 汪奇 杨力 饶宇[1,2] 刘宇阳 LIU Xing;LOU Jian;YU Hongqian;WANG Qi;YANG Li;RAO Yu;LIU Yuyang(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Key Laboratory of Icing and Anti/De-icing,China Aerodynamics Research and Development Center,Mianyang Sichuan 621000,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]中国空气动力研究与发展中心结冰与防除冰重点实验室,四川绵阳621000

出　　处：《航空动力学报》2025年第2期427-436,共10页Journal of Aerospace Power

基　　金：中国空气动力研究与发展中心结冰与防除冰重点实验室开放课题资助(2201IADL20220403)。

摘　　要：结合神经网络方法发展一套隐含物理变量的空间迭代算法用于气膜叠加模型,并通过全覆盖气膜平板综合冷效实验和绝热数值模拟分别进行验证,解决了直接采用神经网络方法不蕴含真实物理规律以及缺乏可解释性的问题。通过精度对比和隐含变量挖掘,获得垂直平板方向5倍平板前缘边界层厚度位置处沿侧向的速度等4个可解释的气膜叠加模型关键封闭变量,验证了融合模型的精度和可靠性。A set of spatial iterative algorithms with implicit physical variables were developed in combination with the neural network method for film superposition model,and this model was verified by comprehensive film effectiveness experiment of a flat plate fully covered film cooling and adiabatic numerical simulation respectively.And it solved the problem that the directly used neural network method does not contain real physical laws and lacks of interpretability.At the same time,by comparing accuracy and mining implicit variables,the interpretable 4 key closed variables of the film superposition model,such as the lateral velocity at the position of 5 times the thickness of the front edge boundary layer in the vertical plate direction,were obtained,thus verifying the accuracy and reliability of the fusion model.

关 键 词：气膜冷却 叠加模型 深度学习 空间迭代 隐含物理变量 

分 类 号：V233.1[航空宇航科学与技术—航空宇航推进理论与工程]