中国机械工程

• 智能制造 • 上一篇    下一篇

基于NSGA-Ⅲ算法的低比转速离心泵多目标优化设计

童哲铭1;陈尧1;童水光1;余跃1;李进富2;郝国帅3   

  1. 1.浙江大学机械工程学院,杭州,310027
    2.杭州碱泵有限公司,杭州,310027
    3.沈阳透平机械股份有限公司,沈阳,110869
  • 出版日期:2020-09-25 发布日期:2020-10-07
  • 基金资助:
    国家重点研发计划资助项目(2018YFB0606105);
    国家自然科学基金资助项目(51708493,51821093);
    浙江省自然科学基金资助项目(LR19E050002);
    浙江省重点研发计划资助项目(2018C01020,2018C01060,2019C01057)

Multi-objective Optimization Design of Low Specific Speed Centrifugal Pumps Based on NSGA-Ⅲ Algorithm

TONG Zheming1;CHEN Yao1;TONG Shuiguang1;YU Yue1;LI Jinfu2;HAO Guoshuai3   

  1. 1.School of Mechanical Engineering,Zhejiang University,Hangzhou,310027
    2.Hangzhou Alkali Pump Co.,Ltd.,Hangzhou,310027
    3.Shenyang Turbo Machinery Co.,Ltd.,Shenyang,110869
  • Online:2020-09-25 Published:2020-10-07

摘要: 为提高IJ125-100-400-00型低比转速离心泵的水力性能,采用基于参考点的非支配排序遗传算法(NSGA-Ⅲ)结合近似模型、数值模拟等方法对叶片进行了多目标优化。根据水力损失模型和灵敏度分析结果,选取叶轮出口宽度、叶轮出口直径以及叶轮出口角度3个参数作为设计变量; 使用拉丁超立方抽样随机生成了60组设计方案,针对非简化模型进行了数值模拟,得到了对应的外特性值,并建立了高效的BP神经网络近似模型;最终采用NSGA-Ⅲ算法对近似模型进行了多目标寻优,得到了最优设计变量组合。研究结果表明:初始模型的数值模拟结果与实验值具有很好的一致性,两者在额定工况下扬程的绝对误差小于2.14%,;BP神经网络拟合预测效果良好;优化后的离心泵水力效率提高了6.86%,扬程满足设计需求,泵内部流场明显改善。

关键词: 离心泵, 数值模拟, 神经网络, 多目标优化, NSGA-Ⅲ算法

Abstract: In order to improve the hydraulic performance of the IJ125-100-400-00 low specific speed centrifugal pumps, a non-dominated sorting genetic algorithm based on reference point(NSGA-Ⅲ) combined with approximate model and numerical simulation was used to optimize the multi-objective optimization of the blades. According to the hydraulic loss model and sensitivity analysis results, three parameters of impeller exit width, impeller exit diameter and impeller exit angle were selected as design variables. Sixty sets of design schemes were randomly generated by using Latin hypercube sampling, and the numerical simulations were performed for non-simplified models. The external characteristic values were obtained, and an efficient BP neural network approximation model was established. Finally, the NSGA-Ⅲ algorithm was used to optimize the approximate model and the optimal design variable combination was obtained. The results show that the numerical simulation results of the initial model are in good agreement with the measured values. The head absolute errors of the two values under rated conditions are less than 2.14%. The BP neural network has good prediction results. Optimized centrifugal pump hydraulic efficiency is increased by 6.86%, the lift meets the design requirements, and the internal flow field of the pumps is significantly improved.

Key words: centrifugal pump, numerical simulation, neural network, multi-objective optimization, NSGA-Ⅲ algorithm

中图分类号: