摘要
参数化设计和计算流体力学被广泛应用于流体机械的优化设计.采用旋转机械设计软件CFturbo对轴流泵进行水力设计.为缩短优化周期,基于Isight多学科优化平台,通过编写批处理命令将CFturbo与PumpLinx集成,实现了轴流泵的CFD自动优化.以提高轴流泵的水力效率为优化目标,采用最优拉丁超立方设计对叶轮和导叶的7个设计变量进行空间采样,设计了72组方案.基于PumpLinx的数值模拟结果,建立了目标函数与设计变量之间的径向基神经网络模型,并采用多岛遗传算法对其进行优化,结果表明:数值模拟结果与试验结果吻合较好,且径向基神经网络模型能准确预测轴流泵效率与设计变量的关系.优化后,设计点效率提高了4.46%,而扬程几乎不变.通过Pareto图分析,获得了设计变量对目标影响的显著水平,可为轴流泵的优化设计提供一定的参考.
Parametric.＇ design and computational fluid dynamics （CFD） techniques have been widely applied to the optimum design of fluid machinery. The axial flow pump was designed by turbomachinery design software - CFturbo, which was integrated with PumpLinx by batch commands based on Isight software to realize the CFD automatic optimization of axial flow pump. Seven design variables of the blade profile and diffuser were selected by Optimal Latin Hypercube Design （Opt LHD） to generate design points within the selected design space. The hydraulic efficiency at a designed flow rate was set as the objective function. An surrogate model, Radial Basis Functions （RBF） neural network was constructed for the objective function based on the numerical results. Finally, the best combination of the design variables were obtained by solving the approximation model with the Muhi-lsland Genetic Algorithm （MIGA）. The results show that the performance curve simulated by CFD had a good agree- ment with that of the experiment, thus the RBF model can predict the efficiency accurately. After optimization, the efficiency was improved by 4. 46%, while the head kept almost constant under the design condition. After the Pareto graph analysis, the significance levels of design variables were obtained, which can provide a certain reference for optimum design of the axial flow pump.
出处
《排灌机械工程学报》
EI
CSCD
北大核心
2017年第6期481-487,共7页
Journal of Drainage and Irrigation Machinery Engineering
基金
国家科技支撑计划项目(2015BAD20B01)
江苏省水利科技项目(2015042)
江苏高校自然科学研究项目(09KJB570001)
关键词
轴流泵
程序集成
最优拉丁超立方设计
优化设计
径向基神经网络
axial flow pump
program integration
Optimal Latin Hypercube Design
optimum design
radial basis functions neural network
作者简介
陆荣（1990-），男，江苏宜兴人，硕士研究生（349568559@qq．com）．主要从事低扬程泵优化设计研究．
袁建平（1970-），男，江苏金坛人，研究员，博士生导师（yh@ujs．edu．cn），主要从事流体机械内部流动及优化设计研究．
