液压机械无级变速箱换段液压故障诊断的BP方法  

Hydraulic fault diagnosis of hydro-mechanical continuously variable transmission in shift based on BP method

作　　者：张晓辉[1] 太健健 王光明[1] 张海军[2] 王成飞[2] 钟成义[2,3] 

机构地区：[1]山东农业大学机械与电子工程学院,山东泰安271018 [2]南京农业大学工学院,南京市210031 [3]农业部南京农业机械化研究所,南京市210014

出　　处：《中国农机化学报》2016年第10期133-139,共7页Journal of Chinese Agricultural Mechanization

基　　金：江苏省科技支撑计划资助项目(BE2014134);江苏省产学研联合创新资金自助项目(BY2014128-04);山东省现代农业产业技术体系棉花创新团队资助项目(SDAIT-0701110);山东农业大学博士后研究经费资助项目(76463)

摘　　要：提出一种针对液压机械无级变速箱离合器液压系统的故障诊断方法。首先,试验得到87组压力数据并进行数据预处理;而后,通过随机抽取的方法得到训练样本集与测试样本集,基于BP神经网络对训练样本集进行训练并得到故障判别模型,使用该模型对随机样本集进行分类测试;最后,基于神经网络特征选择方法对所得分类模型进行属性约简,并对BP神经网络的多值分类作了比较研究。结果显示:BP二值分类方法对正常及4种典型故障模式的平均识别率分别为98.89%、99.78%、99.78%、99.11%、99.78%;BP多值分类方法对5种典型油路状态模式的平均识别率分别为97.11%、99.78%、99.33%、99.56%、100%;正常模式、活塞卡住、密封圈损坏、油道阻塞与密封不严分别可约简3、4、3、3、3个样本属性。该结果表明:换段期间的油压波动与各故障模式之间存在特异性关联,可通过BP网络方法进行模式分类,但该方法对于组合故障判别无效,还有待进一步研究。Diagnosis method for hydro-mechanical CVT clutch hydraulic system fault was proposed Firstly,87 pressure data were obtained and preprocessed.Secondly,the training and testing samples were gained through randomly select methods and the fault discriminate model was gotten by training the training samples,then random samples were classified to test by using the model.Finally,attribute of classification models were simplified based on the neural network feature selection methods and multi-valued BP neural network categories were compared.The results showed that five typical average oil state pattern recognition rates of BP binary classification method were 98.89%,99.78%,99.78%,99.11%,99.78%;five typical oil-state model average recognition rates of BP multi-value classification were 97.11%,99.78%,99.33%,99.56% and 100%;five failure modes could be simplified to 3,4,3,3,3property.The results above indicated that oil pressure fluctuation was associated with the various failure modes during the transducer period and pattern classification could be carried out by BP Network.

关 键 词：湿式离合器 故障诊断 无级变速 特征选择 神经网络 

分 类 号：S219.07[农业科学—农业机械化工程]