双通道特征融合CNN-GRU齿轮箱故障诊断  

Dual-channel feature fusion CNN-GRU gearbox fault diagnosis

作　　者：张龙 甄灿壮 易剑昱 蔡秉桓 徐天鹏 尹文豪 ZHANG Long;ZHEN Canzhuang;YI Jianyu;CAI Binghuan;XU Tianpeng;YIN Wenhao(School of Mechatronic&Vehicle Engineering,East China Jiaotong University,Nanchang 330013,China)

机构地区：[1]华东交通大学机电与车辆工程学院,南昌330013

出　　处：《振动与冲击》2021年第19期239-245,294,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金(51665013);江西省研究生创新资金(YC2019-S243)。

摘　　要：旋转部件是否发生局部故障,关键是判断其振动信号在空间上是否出现周期性冲击以及周期大小。卷积神经网络(CNN)善于挖掘数据空间上的局部重要的信息特征,具有“端对端”的优势,从而克服了人工提取特征的缺陷;由于振动信号在时间维度上也蕴含着丰富的信息,而长短时记忆网络(LSTM)善于从动态变化的序列数据中学习到时间上的关联性;门控递归单元(GRU)属于LSTM的变种,但相对于LSTM结构更加简洁,参数的数量更少,因此将CNN的空间处理能力和GRU时序处理能力的优势结合,提出一种双通道特征融合CNN-GRU齿轮箱故障诊断方法;即采用并列式结构令CNN与GRU双通道同时提取齿轮箱原始振动信号的故障特征,然后将双通道提取的特征向量合并成一个融合特征向量,输入到SoftMax进行故障分类。该方法可以直接从原始振动信号自适应提取到空间和时序的融合特征,实现了“端对端”的故障诊断。用齿轮实测数据和西储大学轴承数据进行验证,试验结果表明,所提方法识别准确率较高,具有实用性和可行性。Whether a rotating part has local fault or not,the key is to judge if its vibration signal has periodic impact in space and period size.Convolutional neural network(CNN)is good at mining local important information features in data space,and has the advantage of"end-to-end"to overcome the defect of manual feature extraction.As vibration signal also contains rich information in time domain,the long-short term memory(LSTM)network is good at learning temporal relevance from dynamic sequence data.The gate recurrent unit(GRU)is a variant of LSTM,but compared with LSTM,its structure is more concise and the number of parameters is less.Here,combining advantages of CNN’s spatial processing ability and GRU’s temporal processing ability,a dual-channel feature fusion CNN-GRU gearbox fault diagnosis method was proposed.A parallel structure was adopted to make CNN and GRU extract fault features of the original vibration signal of gearbox simultaneously,and then feature vectors extracted from the two channels were combined into a fused feature vector,it was input into the software SoftMax for fault classification.It was shown that the proposed method can adaptively extract fused features of space and time sequencies directly from the original vibration signal to realize the"end-to-end"fault diagnosis.The test results showed that the proposed method has a higher recognition accuracy,and a certain practicability and feasibility.

关 键 词：齿轮箱 卷积神经网络(CNN) 门控递归单元(GRU) 故障诊断 

分 类 号：TH133.3[机械工程—机械制造及自动化]