论文部分内容阅读
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）由气动计算方法确定后，人为地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．实际平行直壁扩压器等厚叶片的数值实验表明，该神经网络经有限样本训练后，能成功地求解出满足给定速度分布的叶片形状
An artificial neural network method was proposed to solve the inverse proposition design of centrifugal compressor diffuser vane line given the surface velocity distribution of the blade. The neural network used has a 4-layer feedforward network structure. After the basic structure size of diffuser cascade (such as inlet and outlet diameter, inlet and outlet installation angle, height, number of blades, etc.) is determined by the aerodynamic calculation method, a certain number of blade shapes are artificially constructed. Through the existing CFD analysis program, The surface speed analysis and calculation. The obtained blade velocity distribution and blade shape are taken as samples, and the standard BP algorithm is used to train the neural network. Numerical experiments on thick blades such as parallel straight wall diffusers show that the neural network can successfully solve the shape of the blade that satisfies the given velocity distribution after being trained by finite samples