基于神经网络的叶栅气动性能影响研究  

Research on Aerodynamic Performance Influence of Cascade Error Based on Deep Neural Network

作　　者：马峰 杜亦璨 王掩刚[1] 陈为雄[1] 刘汉儒[1] 尚珣 MA Feng;DU Yican;WANG Yangang;CHEN Weixiong;LIU Hanru;SHANG Xun(School of Power and Energy,Northwestern Polytechnical University,Xi'an 710129,China;School of Mechanics,Civil Engineering and Architecture,Northwestern Polytechnical University,Xi'an 710129,China;Aecc Aviation Power Co,Ltd,Xi'an 710021,China)

机构地区：[1]西北工业大学动力与能源学院,西安710129 [2]西北工业大学力学与土木建筑学院,西安710129 [3]中国航发动力股份有限公司,西安710021

出　　处：《工程热物理学报》2024年第12期3680-3690,共11页Journal of Engineering Thermophysics

基　　金：国家科技重大专项(No.2017-II-0009-0023)。

摘　　要：叶栅气动性能受叶型误差变化的影响较大,为了研究微小几何误差变化对压气机叶栅气动性能影响,本文提出了一种基于深度神经网络的可变叶栅上不可压缩层流定常流场预测方法。该方法可以将流场近似为叶栅变形误差和攻角以及来流马赫数的函数,不需要求解传统方法所使用的Navier-Stokes(N-S)方程。采用拉丁超立方结合蒙特卡洛方法生成大量几何误差参数结合流场参数作为输入,训练深度神经网络模型并进行预测,短时间内就可以完成大量样本,并利用预测结果得到叶栅误差和总压损失系数之间的敏感性关系。The aerodynamic performance of the blade cascade is greatly affected by the change of the blade shape error.In order to study the inffuence of the small geometric error change on the compressor performance,statistical theory often requires a large number of example data to verify,and the process often needs too much time and cost.This paper proposes a method for predicting the steady fow field of incompressible laminar flow on variable cascades based on Deep Neural Network(DNN).This method is used to analyze the influence of the aerodynamic performance of the cascade on the refined error model.The method proposed in this paper can approximate the fow field as a function of the cascade deformation error and the angle of attack and the incoming flow Mach number during the work,without the need to solve the Navier-Stokes(N-S)equation used in the traditional method.The Latin hypercube combined with Monte Carlo method is used to generate a large number of geometric error parameters combined with fow field parameters as input,and the prediction is made through the trained DNN.The flow field prediction of a large number of samples can be completed in a short time,and the sensitivity relationship between the cascade error and the total pressure loss coefficient can be obtained by using the prediction results.

关 键 词：叶轮机械 几何误差 气动敏感性 流场预测 深度神经网络 

分 类 号：V231.3[航空宇航科学与技术—航空宇航推进理论与工程]