Comparative Study on Modal Decomposition Methods of Unsteady Separated Flow in Compressor Cascade

压气机叶栅非定常分离流动的模态分解方法对比研究

¹ School of Power and Energy, Northwestern Polytechnical University, Xi'an 710072, China
² Yangtze River Delta Research Institute of Northwestern Polytechnical University, Taicang 215400, China
³ Beijing Changcheng Institute of Metrology and Measurement, Beijing 100095, China

Received: 1 April 2019

Abstract

The present work investigated the vortex structure and fluctuation frequency characteristics generated by boundary layer separation of a high-load compressor cascade using modal decomposition methods. The dominant modes and dynamic behaviors of unsteady flow in the cascade were obtained, and the differences of three modal decomposition methods (Proper Orthogonal Decomposition, Dynamic Mode Decomposition and Spectral Proper Orthogonal Decomposition) in feature recognition of cascade flow were discussed. The results show that:(1) The POD method can accurately extract the dominant spatial structure of the flow field, but the modal coefficients are multi-frequency coupled, which makes the dominant modal characteristics of cascade flow unclear. (2) The standard DMD method can obtain the spatial-temporal single frequency mode of cascade flow, as well as their growth rates and frequencies. However, this method is likely to capture the suboptimal mode of large amplitude with large attenuation rate, and fails to obtain the high-frequency coherent structure, which makes it impossible to obtain the dominant feature with limited mode number. (3) The SPOD method, based on spectral characteristics, can obtain spatial and temporal single frequency modes, and there is no modal screening problem. The use of spectral estimation method (SPOD) reduces the sensitivity to numerical noise. This method can obtain the low-rank behavior of cascade flow, which is helpful to understand cascade flow mechanisms. Therefore, SPOD method is more advantageous for the modal analysis of unsteady separated flow in high-load compressor cascade.

摘要

以某高负荷压气机叶栅为研究对象，应用模态分解方法研究叶栅附面层分离所产生的旋涡结构形式和波动频率特征，获得了叶栅非定常流动的主导模态和动态行为，探讨了3种模态分解方法（本征正交分解（POD）、动态模态分解（DMD）和谱本征正交分解（SPOD））在叶栅流动特征识别的差异性。研究结果表明：①POD方法准确提取叶栅分离流动主要空间结构，但模态系数存在多频率耦合问题，导致对于叶栅流动的主导模态特征无法清晰识别；②标准DMD方法可以获得叶栅流动时空单频模态，及其增长率和频率，但是该方法捕捉到流场大衰减率的大振幅次优模态，未能获得高频相干结构，导致无法用有限的模态数得到主导特征；③基于谱特征的SPOD方法能够获得时空单频模态，且不存在模态筛选问题，谱估计方法的使用降低了对数值噪声的敏感性，同时该方法获得了叶栅流动的低秩行为，有助于增强对叶栅流动机理的理解。因此，对高负荷压气机叶栅分离流动进行模态分析时，SPOD方法更具优势。