针对传统流场分析方法难以提取非线性与高维度数据特征的问题,结合自适应噪声完全集合经验模态分解(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)与分形盒维数对速度、压力和涡量等时程数据进行分解、筛选及重构,并结合深度学习建立尾流特征识别模型,以实现翼型特征和流动攻角的反向预测.结果表明:当训练集和测试集数据来自同一翼型时,所提模型对流动攻角的识别率最高可达98.8%,当数据来自不同翼型时,准确率仍可达95.6%,验证了所提模型的准确性、可行性及泛化性.
Analysis of Wingtip Vortex Characteristics based on Time History Data Using a Feature Recognition Model
Addressing the challenge of recognizing nonlinear and high-dimensional data features in tradi-tional flow field feature extraction methods,we optimized the complete ensemble empirical mode decom-position with adaptive noise(CEEMDAN)and fractal box dimension methods to decompose,filter,and reconstruct time history data such as velocity,pressure and vorticity.By combining this approach with deep learning,a wake feature recognition model was established to achieve reverse prediction of airfoil characteristics and flow attack angles.The results demonstrate that when the training and test datasets are derived from the same airfoil,the recognition accuracy of the proposed model to the flow attack angle can reach up to 98.8%;when the data originates from different airfoils,the recognition accuracy can be as high as 95.6%,verifying the accuracy,feasibility and generalizability of the proposed method.
NETL
NSTL
万方数据
