航空学报 > 2022, Vol. 43 Issue (11): 526531-526531   doi: 10.7527/S1000-6893.2021.26531

基于时程卷积自编码的机翼绕流特征识别方法

战庆亮1, 白春锦1, 张宁1, 葛耀君2   

  1. 1. 大连海事大学 交通运输工程学院, 大连 116026;
    2. 同济大学 土木工程防灾国家重点实验室, 上海 200092
  • 收稿日期:2021-10-18 修回日期:2021-11-08 出版日期:2022-11-15 发布日期:2021-12-09
  • 通讯作者: 战庆亮,E-mail:zhanqingliang@163.com E-mail:zhanqingliang@163.com
  • 基金资助:
    国家自然科学基金(51778495,51978527);桥梁结构抗风技术交通行业重点实验室(上海)开放课题(KLWRTBMC21-02);辽宁省教育厅研究计划(LJKZ0052)

Feature extraction method of flow around airfoil based on time-history convolutional autoencoder

ZHAN Qingliang1, BAI Chunjin1, ZHANG Ning1, GE Yaojun2   

  1. 1. College of Transportation and Engineering, Dalian Maritime University, Dalian 116026, China;
    2. State Key Laboratory for Disaster Reduction in Civil Engineering, Tongji University, Shanghai 200092, China
  • Received:2021-10-18 Revised:2021-11-08 Online:2022-11-15 Published:2021-12-09
  • Supported by:
    National Natural Science Foundation of China (51778495, 51978527); Industry Key Laboratory of Bridge Structure Wind Resistance Technology (Shanghai) Open Project (KLWRTBMC21-02); Research Project of Liaoning Education Department (LJKZ0052)

摘要: 机翼周围的流动状态直接影响其受力特性,流动特征的识别与分析对保证机翼的气动力研究尤为关键。基于空间流场参数的流动特征识别结果受主观阈值影响大;基于流场快照数据的流动特征分析难以完整表征流场的时变特征,且大范围的流场快照获取难度大,因而其实用性受限。本文基于流场时程数据的低维表征模型提出了无监督自动编码的流场时程特征识别方法。采用深度学习技术充分挖掘时程信号中的隐含的流动特征,建立流场时程数据的低维表征模型;进一步对低维的表征编码进行分析,将包含不同时序特征的测点样本进行特征聚类,实现了基于空间点时程数据的流场特征提取与识别。通过对NACA0012翼型的流场进行特征提取与分析验证了所得流动特征低维表征的准确性,实现了基于流场时程数据的流动分离区自动识别。本文可为相关流场特征提取、特征分析和特征表征等问题的研究提供新的方法与参考。

关键词: 流场时程, 卷积自编码, 深度学习, 特征提取, 机翼绕流

Abstract: The flow around an airfoil directly affects its aerodynamics, and flow feature analysis is critical to ensure its aerodynamic characteristics. The feature recognition based on spatial parameters is greatly affected by subjective threshold. And flow analysis based on the snapshot data, which is difficult to obtain a large-scale snapshot in experiment, is difficult to fully represent the time-varying features of the flow evolution. Based on the low-dimensional representation model of flow time-history data, an unsupervised autoencoder method is proposed to identify flow features automatically. Deep learning is used to mine the implicit flow features in the time-history signal, and a low-dimensional representation model of the time-history data is established. The low-dimensional representation code is further analyzed, and the measurement point samples containing different features are clustered to extract and identify flow features. Analysis of the flow field around the NACA0012 airfoil with a low Reynolds number is carried out. The flow separation zone was successfully identified using the proposed feature classification method. Results proved that the proposed method can be used to solve the problems such as flow feature extraction, feature analysis, and feature characterization.

Key words: flow time-history, convolutional autoencoder, deep learning, feature extraction, flow around airfoil

中图分类号: