基于信息融合及堆栈降噪自编码的齿轮故障诊断  

Gear fault diagnosis based on information fusion and stacked de-noising auto-encoder

作　　者：李松柏[1] 康子剑 陶洁[1] LI Songbai;KANG Zijian;TAO Jie(School of Mechanical and Electrical Engineering,Central South University,Changsha 410083,China)

机构地区：[1]中南大学机电工程学院,长沙410083

出　　处：《振动与冲击》2019年第5期216-221,共6页Journal of Vibration and Shock

基　　金：国家重点基础研究发展计划(973计划)项目(2014CB046300);湖南省科技计划项目(2016GK2005)

摘　　要：针对传统分类器在齿轮故障诊断时易受噪声干扰,以及单传感器可靠性和容错性不佳的问题,提出了基于信息融合及堆栈降噪自编码(SDAE)的齿轮故障诊断方法。该方法提取多传感器振动时域信号进行数据级融合;利用SDAE进行逐层特征提取;通过有标签数据对网络进行整体微调,建立齿轮状态监测模型。对不同故障齿轮进行故障诊断,对比SDAE、支持向量机(SVM)、神经网络(BPNN)的诊断准确性和鲁棒性。试验结果表明:基于信息融合的SDAE的齿轮故障诊断率为95.17%,高于单一信号分类器的准确率,鲁棒性优于对比方法。Aiming at problems of traditional classifiers being susceptible to noise interference and a single sensor’s reliability and fault tolerance being not good,a gear fault diagnosis method based on multi-sensor information fusion and stacked de-noising auto-encoder (SDAE) was proposed.Firstly,multi-sensor vibration time domain signals were extracted to do data level fusion.Then SDAE was used to extract features layer by layer.Finally,labeled data was used to do the overall fine-tuning of the deep learning network and establish a gear state monitoring model.The fault diagnoses were conducted for different faulty gears,and diagnosis correctness and robustness of SDAE,SVM and BPNN were compared.The results showed that the gear fault diagnosis accuracy rate of SDAE based on information fusion reaches 95.17%,it is higher than that of the single signal classifier;the proposed method’s robustness is superior to those of other methods compared with the former.

关 键 词：信息融合 堆栈降噪自编码 深度学习 齿轮 故障诊断 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]