混流泵三元叶片优化设计方法研究  

RESEARCH ON THE OPTIMIZATION METHOD OF 3-D BLADE IN MIXED-FLOW PUMP

作　　者：席光[1] 卢金铃[1] 祁大同[1] 

机构地区：[1]西安交通大学能动学院,陕西西安710049

出　　处：《工程热物理学报》2004年第6期952-955,共4页Journal of Engineering Thermophysics

基　　金：国家自然科学基金重点项目(No.50136030)

摘　　要：以叶片角的分布为优化变量,叶轮效率为优化目标,将三维粘性流动分析与神经网络相结合对混流泵叶片进行了优化设计。在设计空间内根据实验设计理论安排神经网络的训练样本,选择径向基函数网络建立设计变量和目标函数间复杂的响应关系,利用遗传算法进行优化。针对研究对象设计变量较多、变化范围较大的问题,提出子系统优化的初步思想,即:将优化变量进行分类,对每类变量建立子系统,分别在子系统内进行局部优化,然后再考虑子系统之间的相互影响进行全局优化。对一混流泵的三元扭曲叶片进行优化,叶轮效率明显提高,从而验证了该方法的简便性与有效性。The blade in a mixed-flow pump was optimized based on three dimensional viscous flow analysis and neural networks with the blade angle being treated as design variable and the efficiency being treated as objective function. RBF neural networks were adopted to construct the response relation between the design variable and the objective function, and the training sample data were schemed according to the design of experiment theory. As there was many design variables and the design space was very large, a new subset based approach was proposed. Firstly, the design variables were divided into several groups, and a subset was formed in each group. Then, in each subset, part of the blade was optimized respectively and independently. Finally, the whole blade was optimized with the effect of each subset being taken into account. An optimized blade in a mixed-flow pump, where te efficiency was highly improved, confirms the validity of this newly proposed method.

关 键 词：优化 混流泵 神经网络 遗传算法 

分 类 号：O357[理学—流体力学]