基于多通道卷积神经网络的液压系统多故障诊断  

Multi-Fault Diagnosis of Hydraulic System Based on Multi-Channel Convolutional Neural Network

作　　者：李文华[1] 牛国波 刘羽佳 周性坤 林珊颖 葛杨元 LI Wen-hua;NIU Guo-bo;LIU Yu-jia;ZHOU Xing-kun;LIN Shan-ying;GE Yang-yuan(National Center for International Research of Subsea Engineering Technology and Equipment,College of Marine Engineering,Dalian Maritime University,Dalian 116026,China;Nantong Li Wei Machinery Co.,Ltd.,Nantong 226500,China)

机构地区：[1]大连海事大学轮机工程学院海底工程技术与装备国际联合研究中心,大连116026 [2]南通力威机械有限公司,南通226500

出　　处：《组合机床与自动化加工技术》2023年第3期91-94,98,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：辽宁省“兴辽英才计划”项目资助(XLYC2007092);辽宁省科学技术计划(2020-HYLH-35);国家自然科学基金面上项目(51779026);工信部高技术船舶科研计划(工信部装函﹝2018﹞473号);国家重点研发计划资助(2018YFC0309003);高等学校学科创新引智计划(111计划)(B18009)。

摘　　要：针对现有研究液压系统单一故障较多而复杂多故障较少的情况,结合液压系统实际采集的数据往往具有多采样率的特点,提出一种多通道卷积神经网络模型,将不同采样频率下的传感器数据分别作为一个通道输入,利用卷积神经网络自动进行提取特征,全连接层将提取的多传感器特征信息进行融合,实现液压系统的混合多故障诊断。利用多级评价指标将提出的多通道卷积神经网络模型与传统的单通道卷积神经网络算法对比,结果表明,相较于单通道卷积神经网络不能很好地识别液压系统多故障发生时的液压泵和蓄能器故障,提出的算法模型诊断准确率达到了99%,并且具有很好的鲁棒性。Aiming at the fact that there are many single faults and few complex multi-faults in the existing hydraulic system,combined with the fact that the actual data collected by the hydraulic system often has the characteristics of multiple sampling rates,a multi-channel convolutional neural network model is proposed.The sensor data of the sensor is input as a channel,and the convolutional neural network is used to automatically extract features.The fully connected layer fuses the extracted multi-sensor feature information to realize the hybrid multi-fault diagnosis of the hydraulic system.Compared with the single-channel convolutional neural network,which cannot well identify the hydraulic pump and accumulator faults when multiple faults of the hydraulic system occur,the diagnostic accuracy of the proposed algorithm model reaches 99%,and it has good robustness.

关 键 词：液压系统 采样频率 多故障诊断 多通道卷积神经网络 

分 类 号：TH137[机械工程—机械制造及自动化] TG502[金属学及工艺—金属切削加工及机床]