北京大学学报自然科学版 ›› 2023, Vol. 59 ›› Issue (2): 251-260.DOI: 10.13209/j.0479-8023.2022.098

上一篇    下一篇

基于频域降采样和CNN的轴承故障诊断方法

周翔宇, 毛善君, 李梅   

  1. 北京大学遥感与地理信息系统研究所, 北京 100871
  • 收稿日期:2022-03-13 修回日期:2022-04-06 出版日期:2023-03-20 发布日期:2023-03-20
  • 通讯作者: 毛善君, E-mail: sjmao(at)pku.edu.cn
  • 基金资助:
    国家重点研发计划(2020YFB1314000)资助

Bearing Fault Diagnosis Method Based on Down-Sampling in Frequency Domain and CNN

ZHOU Xiangyu, MAO Shanjun, LI Mei   

  1. Institute of Remote Sensing and Geographical Information System, Peking University, Beijing 100871
  • Received:2022-03-13 Revised:2022-04-06 Online:2023-03-20 Published:2023-03-20
  • Contact: MAO Shanjun, E-mail: sjmao(at)pku.edu.cn

摘要: 在工业领域, 设备运行过程中采集的原始故障信号具有强噪声以及多工况的特点, 现有的基于数据的轴承故障诊断模型的抗噪能力与泛 化能力相对较弱 。针对以上问题, 提出一种基于频域降采样(down-sampling)和卷积神经网络(CNN)的轴承故障诊断方法Ds-CNN。频域降采样包含最大偏移降采样和噪声横截断两个部分, 可以实现样本增强, 降低样本在频域的差异性, 同时减弱噪声对频域信号的影响。基于频域信号建立的CNN模型能够自动提取降采样后频域信号的故障特征, 并完成对轴承故障的识别分类。实验结果表明, 在强噪声环境和多工况条件下, 与目前常用模型相比, Ds-CNN具有更高的识别准确率。

关键词: 轴承故障诊断, 深度学习, 卷积神经网络(CNN), 强噪声, 多工况

Abstract:

In the industrial field, the original fault signals collected during the operation of the equipment have the characteristics of strong noise and multiple working conditions. Most of previous data-driven fault diagnosis methods for bearings have relatively weak anti-noise ability and generalization ability. To solve these problems, a novel bearing fault diagnosis method based on down-sampling in frequency domain and convolutional neural network (CNN), called Ds-CNN, is proposed. Down-sampling in frequency domain consists of maximum down-sampling with bias and noise transverse truncation, which can realize data augmentation, reduce the difference between samples in frequency domain, and reduce the influence of noise on signals in frequency domain. The CNN model based on frequency domain signals can automatically extract fault features from signals after down-sampling and complete the identification and classification of bearing faults. The results of the experiment show that Ds-CNN has higher recognition accuracy than common models under strong noise environment and multiple working conditions. 

Key words: bearing fault diagnosis, deep learning, convolutional neural network (CNN), strong noise, multiple working conditions