基于深度学习的液压监测系统蓄能装置故障诊断  

Fault Diagnosis of Accumulators in Hydraulic Monitoring System Based on Deep Learning

作　　者：姜保军[1,2] 王帅杰 董绍江 JIANG Bao-jun;WANG Shuai-jie;DONG Shao-jiang(School of Electromechanical and Vehicle Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Key Laboratory of Urban Rail Vehicle System Integration and Control,Chongqing Jiaotong University,Chongqing 400074,China)

机构地区：[1]重庆交通大学机电与车辆工程学院,重庆400074 [2]重庆交通大学城市轨道车辆系统集成与控制重点实验室,重庆400074

出　　处：《组合机床与自动化加工技术》2019年第9期89-92,共4页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金(51775072);重庆市教委科学技术研究项目(KJ1500525);重庆市基础与前沿研究计划项目(cstc2016jcyjA0526);重庆市社会事业与民生保障科技创新专项项目(cstc2017shmsA30016)

摘　　要：为了提高液压检测系统故障识别的准确率,提高对液压系统中的蓄能装置故障的诊断率,提出了一种基于堆栈稀疏自编码器的深度学习算法。该方法采用希尔伯特-黄变换和小波变换对压力信号的特征进行提取,然后对堆栈稀疏自编码器(SSAE)进行训练。将训练好的模型连接Softmax分类器,实现对蓄能装置的最佳压力、压力略微减轻、压力严重减轻、接近完全失效的4类压力状态进行诊断。实验结果表明,该深度学习神经网络比机器学习的准确率更高,可以达到98.3%,能够更好的识别液压系统蓄能装置的故障类型。In order to improve the accuracy of Fault diagnosis in hydraulic detection system,a deep learning algorithm was proposed based on stacked sparse autoencoders.By using Hilbert-Huang transform and wavelet transform,the features were extracted from the pressure signals.And the stacked sparse autoencoders(SSAE)were trained.The trained model was connected to the Softmax classifier to diagnose the four types of pressure states of the hydraulic accumulator device(the optimal pressure,the slightly reduced pressure,the severely reduced pressure,close to total failure).The experimental results show that the deep learning neural network has higher accuracy than machine learning and it’s accuracy reach 98.3%.Deep learning neural network model can better identify the fault type of the hydraulic accumulator device.

关 键 词：深度学习 蓄能装置 故障诊断 

分 类 号：TH137[机械工程—机械制造及自动化] TG506[金属学及工艺—金属切削加工及机床]