损失及落后角代理模型在多级轴流压气机特性预测中的应用

Application of Loss and Deviation Surrogate Model in Prediction of Multistage Axial Compressor Characteristics

关键词:  代理模型  损失  落后角  压气机特性预测  三元流动

Key words:  Surrogate Model  Loss  Deviation  Characteristics  Three-dimensional Flow Theory

摘要:

为了提高轴流压气机设计能力，进而提高发动机的特性，研究人员需要掌握一种能够准确预测轴流压气机的压比、效率等特性的方法。结合运用三元流动理论和传统损失落后角模型计算出的压气机流场数据，本文利用正则化径向基函数神经网络取代经验公式搭建了一种新的损失及落后角模型，计算了E3十级高压压气机的特性；并分别研究了不进行正则化和进行正则化对损失及落后角预测的影响与其对压气机效率压比特性预测的影响。结果表明在多级压气机中，在训练样本区分转静子，区分转速，区分工况条件下，使用正则化的径向基神经网络代理模型在大部分情况下能够较为准确的预测损失、落后角及多级压气机整体特性，但是对沿叶高分布的损失及落后角预测能力还有待提高。

Abstract:

In order to improve the design capability of axial compressor and characteristics of the engine, researchers need to master a method which can accurately predict pressure ratio and efficiency of axial compressor. Combining the data of compressor flow field calculated by theory of three-dimensional flow and empirical loss and deviation formulas, a new loss and deviation model was established by using regularized radial basis function neural network instead of empirical formulas, and the characteristics of E3 10-stage high pressure compressor were calculated. The effects of non-regularization and regularization on loss and deviation prediction were studied respectively as well as the influence of compressor efficiency and pressure ratio prediction. The results showed that in a multistage compressor, under the conditions of distinguishing rotor and stator, rotating speed and operating conditions, the regularized radial basis function neural network surrogate model could accurately predict the loss and deviation comparatively and overall characteristics of a multistage compressor in most cases, however, this kind of work couldn’t have a satisfying performance on the prediction of loss and deviation from shroud to hub.