摘要
利用反问题计算间接对叶片进行参数化并结合神经网络技术,建立了一种基于三维黏性流动分析的叶轮优化设计方法.该方法将设计变量分为叶片的设计变量和子午流道的设计变量,并分别作为子系统和中心系统.首先,在子系统内保持子午流道形状不变,利用遗传算法对叶片进行优化.然后,在中心系统内利用试验设计理论安排训练样本,采用神经网络建立叶轮性能与子午流道设计参数之间的响应关系,同时计及叶片形状对叶轮性能的影响,对子午流道及叶片进行优化.该方法设计变量个数较少,通过将设计变量进行分类并分步优化,有效地减少了计算量,对三元叶轮的优化问题具有良好的应用前景.通过对一台混流泵叶轮进行优化,扬程和效率分别提高了7.43%和3.37%,从而验证了该方法的实用性.
An impeller optimization method based on the 3-D viscous flow analysis was proposed, where the blade was parameterized using inverse design method and neural networks was adopted. The design variables were assorted into two sets: the meridional channel design variables and the blade design variables that were treated as main system and subsystem, respectively. First, in the subsystem, the blade was optimized according to genetic algorithm with the meridional channel remained constant. Then, in the main system, neural network was adopted to construct the response relation between impeller performance and meridional channel design variables, where the training sample data were schemed according to the design of experimental method, and the effect of blade shape on the impeller performance was taken into account and the meridional channel was optimized. Fewer design variables and less calculation effort were required in this method that may be widely used in the optimization of three-dimension impellers. An optimized im- peller in a mixed-flow pump, where the head and the efficiency increased by 7.43% and 3.37% respectively, confirmed the validity of this newly proposed method.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2005年第9期1021-1025,共5页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金重点资助项目(50136030).
关键词
子午流道
叶片
优化
神经网络
meridional channel
blade
optimization
neural network