%0 Journal Article
%T 液力变矩器的叶片数神经网络模型<br>Blade Number Neural Networks Model of Hydraulic Torque Converter
%A 王安麟
%A 孟庆华
%A 曹岩
%A 韩继斌
%J 西安交通大学学报
%D 2015
%R 10.7652/xjtuxb201507003
%X 针对一元束流理论无法量化表达叶片数对液力变矩器性能影响的缺陷和基于三维流体解析的液力变矩器叶片数设计中大组合、大计算量等难题,提出液力变矩器的叶片数神经网络模型。在结合台架试验数据确认三维流体解析结果准确的基础上,利用正交试验法合理地安排试验,并以三维流体仿真结果作为反向传播网络的训练样本;为提高反向传播网络的设计效率及收敛精度,引入遗传算法来优化反向传播网络的初始权重,训练后的反向传播网络可以对非训练样本集合的液力变矩器性能实现准确预测。研究结果表明,叶片数神经网络模型是基于整机性能匹配的液力变矩器定制化设计的桥梁,对提升整机作业效率具有重要的工程应用价值。<br>One？？dimensional flow bundle theory is unable to quantitatively predict the influence of blade number on hydraulic torque converter performance. And there exist various combinations and huge calculations in blade number design based on 3D fluid analysis. Thus a blade number neural networks model of hydraulic torque converter was constructed. The accuracy of 3D fluid analysis results was confirmed in comparison with the bench test data, and simulations were arranged reasonably by orthogonal experiment method. 3D fluid simulation results were regarded as the training samples of back propagation (BP) neural networks. To improve design efficiency and convergence accuracy, genetic algorithm was introduced for optimizing initial weights and thresholds of BP neural networks, which achieved accurate predictions for non？？training sample sets. The experiments show that blade number neural networks model serves as a bridge for hydraulic torque converter customization design based on vehicle performance matching
%K 液力变矩器
%K 叶片数
%K 反向传播网络
%K 遗传算法&lt
%K br&gt
%K hydraulic torque converter
%K blades number
%K back propagation neural networks
%K genetic algorithm
%U http://zkxb.xjtu.edu.cn/oa/DArticle.aspx?type=view&id=201507003