二维扩压器叶片形线的神经网络设计

提出了一种人工神经网络方法，用于在给定叶片表面速度分布条件下，求解离心压缩机扩压器叶片形线的逆命题设计，所使用的神经网络具有４层前馈网络结构，在扩压器叶栅基本结构尺寸（如进出口直径，进出口安装角，高度，叶片数等）由气动计算公式确定后，人为均构造一定数量的叶片形状，通过现有ＣＦＤ分析程序，对其表面速度进行分析计算，将由此得到的叶片速度分布和叶片形状作为样本，采用标准ＢＰ算法对神经网络进行训练，实际平

Inverse Design of Diffuser Blade by Neural Networks

A neural network approach is proposed to search for a centrifugal compressor diffuser blade profile with a prescribed velocity distribution on its surface. The applied neural networks have a four layered feed forward structure. Blade profiles are determined via aerodynamic computation by constructing parameters such as the inlet and outlet diameter, the blade inlet and outlet angles, the height, and the blade numbers. A Computational Fluid Dynamics (CFD) solver is then applied to find the velocity distribution on the blade surface. The results are used with the help of the standard BP algorithm to tune neural networks. The new approach is validated through the design of a diffuser with two dimensional blade profile. Desired blade profile and velocity distribution are obtained.