• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2020, Vol. 56 ›› Issue (1): 30-36.doi: 10.3901/JME.2020.01.030

• 机械动力学 • 上一篇    下一篇

扫码分享

噪声干扰下的RV减速器故障诊断

彭鹏, 柯梁亮, 汪久根   

  1. 浙江大学机械工程学院 杭州 310027
  • 收稿日期:2019-02-26 修回日期:2019-08-07 出版日期:2020-01-05 发布日期:2020-03-09
  • 通讯作者: 汪久根(通信作者),男,1963年出生,博士,教授,博士研究生导师。主要研究方向为摩擦学与仿生设计。E-mail:me_jg@zju.edu.cn
  • 作者简介:彭鹏,男,1994年出生,博士研究生。主要研究方向为磨损检测技术和故障诊断算法。E-mail:1498714385@qq.com;柯梁亮,男,1995年出生,硕士研究生。主要研究方向为RV减速器故障诊断和寿命预测。E-mail:2691047652@qq.com
  • 基金资助:
    国家高技术研究发展计划(863计划,2015AA043002)、国家自然科学基金(51375436)和浙江省重大科技专项(2017C01047)资助项目。

Fault Diagnosis of RV Reducer with Noise Interference

PENG Peng, KE Liangliang, WANG Jiugen   

  1. School of Mechanical Engineering, Zhejiang University, Hangzhou 310027
  • Received:2019-02-26 Revised:2019-08-07 Online:2020-01-05 Published:2020-03-09

摘要: 在实际工况下,旋转矢量(Rotate vector,RV)减速器的振动信号往往掺杂噪声。被噪声污染后的振动信号给RV减速器的故障诊断带来挑战。为此,提出一种噪声干扰下的卷积神经网络模型(Anti-noise network,ANNet)以实现RV减速器的故障模式识别。该模型首先将一维振动信号通过信号堆叠的方式转化成二维灰度图像,然后采用Dropout操作直接对原始输入信号进行随机干扰,并同时利用多个不同尺度的卷积核对输入信号的不同特征进行自动提取和融合。进一步将ANNet算法与其他算法进行了对比分析,结果表明在不同强度的噪声干扰下,ANNet算法相比于其他算法具有更强的抗噪干扰能力;尤其是在强噪干扰环境下,ANNet模型的诊断准确率比现有算法高出10%~20%。最后探讨和解释了模型的特殊结构设计以及模型具备抗噪能力的原因。

关键词: RV减速器, 噪声干扰, 输入Dropout, 多尺度卷积核, 故障诊断

Abstract: The vibration signal of rotate vector (RV) reducer is often covered with noise, which poses challenges to the fault diagnosis of RV reducer. To this end, a novel model of convolution neural network (Anti-noise network, ANNet) is proposed to achieve the fault mode identification of RV reducers under noise interference. The one-dimensional vibration signal is first converted into a two-dimensional gray image with the method of signal stacking, and then the dropout operation is utilized to directly interfere with the original input signal. Additionally, different sizes of kernels are applied to extract and fuse the different features of the input signal. Furthermore, the proposed method is compared with other algorithms. The results demonstrate that the proposed algorithm has a stronger anti-noise performance than that of others. Particularly, under the condition of intense noise (3 dB), the accuracy of the model is 10%-20% higher than the existing approaches. Finally, the special structural design of the proposed model and the reason for anti-noise performance of the model are discussed and explained.

Key words: RV reducer, random noise, Dropout of input signal, multi-scale kernels, fault diagnosis

中图分类号: