神经网络模型在轴流压气机正问题计算中的应用
Application of Neural Network Model in Calculation of Axial Compressor Direct Problems
何中海 1介石 1吴亚东1
作者信息
- 1. 上海交通大学机械与动力工程学院,上海 200240
- 折叠
摘要
为解决轴流压气机正问题分析过程中传统落后角和总压损失系数经验模型在不同叶型和宽工况范围内预测效果差、更新难度大的问题,根据以往积累的叶型仿真计算结果及叶栅、部件级实验中测得的流场参数,采用BP神经网络模型替代常用的经验模型,改进原有计算程序上得到新的压气机正问题分析程序,并对轴流压气机正问题进行了求解分析.在提高神经网络模型的预测精度方面,利用一定范围内的流场参数作为数据集,训练出从基本参数到落后角和总压损失系数的BP神经网络,并将预测相对误差控制在6％和2％之内.针对两台轴流压气机算例进行计算校验,用新分析程序预测结果对比实验和替代之前的计算结果发现,该方法对Stage 37叶片出口参数的预测结果符合实验结果,对高速四级轴流压气机总体性能的预测误差小于7％.
Abstract
In order to solve the problem of poor performance prediction and difficulty in updating tradi-tional deviation angle and total pressure loss coefficient empirical models in the direct problem analysis process of axial compressors in different blade shapes and wide operating conditions,based on the accu-mulated blade shape simulation calculation results and flow field parameters measured in cascade and component level experiments,a back propagation(BP)neural network model has been used instead of the commonly used empirical model.A new compressor direct problem analysis program is obtained on the basis of improving original calculation program,and the direct problem of axial compressors is solved and analyzed by this program.As for improving the prediction accuracy of the neural network model,the BP neural network is trained from the basic parameters to the deviation angle and the total pressure loss coefficient by using the flow field parameters in a certain range as data sets,which controls the relative error of predictions between 6％and 2％.Computational verification is performed for two axial compres-sor examples.Comparing the results of the new analysis program with the experimental results and the previous calculation results,it is found that the prediction results of the Stage 37 blade outlet parameters by this method are consistent with the experimental results,and the prediction error of the overall per-formance of the high speed 4-stage axial compressor is less than 7％.
关键词
轴流压气机/正问题分析/神经网络/流动损失预测/流线曲率法Key words
axial flow compressor/positive problem analysis/neural network/flow loss predicting/stre-amline curvature method引用本文复制引用
出版年
2024
NETL
NSTL
万方数据