基于一维卷积神经网络的齿轮箱故障诊断  

Faults diagnosis method for gearboxes based on a 1-D convolutional neural network

作　　者：吴春志 江鹏程 冯辅周 陈汤 陈祥龙 WU Chunzhi;JIANG Pengcheng;FENG fuzhou;CHEN Tang;CHEN Xianglong(Department of Vehicle Engineering,Academy of Army Armored Forces,Beijing 100072,China)

机构地区：[1]陆军装甲兵学院车辆工程系,北京100072

出　　处：《振动与冲击》2018年第22期51-56,共6页Journal of Vibration and Shock

基　　金：装备预研基金重点项目(9140A27020115JB35071)

摘　　要：传统故障诊断方法通常需要先人工提取特征再用模式识别方法进行分类,难以解决端到端故障诊断的问题,为此,提出了一种利用一维卷积神经网络的齿轮箱故障诊断模型。其特点是可以直接从原始振动信号中学习特征并完成故障诊断。采用PHM 2009 Challenge Data和某型坦克变速箱的复合故障数据对三种传统模型和一维卷积神经网络模型进行测试,结果表明,1-DCNN模型对单一和复合故障诊断准确率均高于传统诊断方法。Traditional diagnosis methods need to extract features manually and classify faults by pattern recognition methods. It’s difficult to solve the problem of end-to-end fault diagnosis. Therefore, a one-dimensional convolution neural network （1-DCNN） model suitable for analyzing vibration data was established. The model can learn features directly from raw vibration data and complete fault diagnosis in succession. Three traditional models and the 1-DCNN model were tested with the compound fault data collected from PHM 2009 Challenge Data and a tank gearbox. The results show that the precision of the 1-DCNN model for single and complex fault diagnosis is higher than that of traditional diagnostic methods.

关 键 词：卷积神经网络 故障诊断 齿轮箱 特征学习 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]