摘要
针对压气机传统性能预测方法存在精度和复杂度之间制约的问题,采用神经网络模型对流线曲率法进行改进优化。首先,利用神经网络作为流线曲率法中经验公式的代理模型,实现了从叶型和来流参数到气动参数的映射;然后,搭建了单转子压气机试验台,对叶片展向气动参数分布进行测试,并与耦合了神经网络模型的流线曲率法计算结果进行了对比。计算和试验结果对比表明:基于神经网络的流线曲率法通过对复杂流场建模,能够实现对流场的准确预测,对单转子试验件在主流区和近壁区最大预测相对误差分别为5%和9.3%,以此形成了新的正问题计算程序,在压气机中取得了有效应用。
Traditional performance prediction methods face a trade-off between accuracy and complexity.To solve that problem,a neural network model was introduced to enhance and refine the streamline curvature method.By utilizing neural networks as surrogate models for empirical formulas in the streamline curvature method,a mapping from blade profiles and inflow parameters to aerodynamic parameters was achieved.A single-rotor compressor test bench was established to assess the distribution of aerodynamic parameters across the blade span,and those results were compared with those obtained from the streamline curvature method coupled with a neural network model.The comparison of computational and experimental results demonstrates that the streamline curvature method enhanced by neural networks can precisely model complex flow fields,the maximum predicted relative errors for the single rotor test object in the mainstream region and the near-wall region are 5%and 9.3%respectively,thereby enabling the development of a novel forward problem computation procedure that has found effective application in compressors.
作者
何中海
介石
吴亚东
HE Zhonghai;JIE Shi;WU Yadong(School of Mechanical Engineering,Shanghai Jiaotong Univ.,Shanghai 200240,China)
出处
《海军工程大学学报》
北大核心
2025年第1期65-71,共7页
Journal of Naval University of Engineering
基金
国家部委基金资助项目(J2019-II-0004-0024,P2022-B-V-004-001)。
关键词
流线曲率法
压气机
神经网络
性能预测
streamline curvature method
compressor
neural network
performance prediction
