计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (7): 269-278.DOI: 10.3778/j.issn.1002-8331.2006-0030

• 工程与应用 • 上一篇    

适用于轴承故障诊断的数据增强算法

林荣来,汤冰影,陈明   

  1. 同济大学 机械与能源工程学院,上海 201804
  • 出版日期:2021-04-01 发布日期:2021-04-02

Data Augmentation Algorithm for Bearings Faults Diagnosis

LIN Ronglai, TANG Bingying, CHEN Ming   

  1. School of Mechanical Engineering, Tongji University, Shanghai 201804, China
  • Online:2021-04-01 Published:2021-04-02

摘要:

针对在轴承故障诊断中存在的故障数据较少、数据所属工况较多的问题,提出了一种基于阶次跟踪的数据增强算法。该算法利用阶次跟踪中的角域不变性,对原始振动信号进行时域重采样从而生成模拟信号,随后重新计算信号的幅值来抵消时域重采样以及环境噪声对原始信号能量的影响,最后使用随机零填充来保证信号在变化过程中采样长度不变。对比实验表明,该算法既可以增加样本多样性,又可以增加数据集样本的数量,改善原始数据集中存在的问题,有效提高故障诊断模型的分类准确率和泛化性能。

关键词: 数据增强, 信号处理, 故障诊断, 阶次跟踪

Abstract:

To address the problem of the inadequate faulty examples and various working conditions of industrial data in bearings faults diagnosis, a practical data augmentation based on order tracking is proposed. With the use of the angular invariance of order tracking, original signals are resampled in time domain in order to obtain the simulated signal that having the sharing pattern in angular domain. Then, the amplitude of the signal is recalculated to offset the energy change causing by resampling and environment noise. Finally, random zero padding is utilized to make the length of the signal consistently. According to experimental results, the newly proposed algorithm can increase the sample diversity and the quantity of dataset. Besides, it can alleviate existing problems in original dataset and effectively improve the classification accuracy and the generalization performance of diagnosis models.

Key words: data augmentation, signal processing, fault diagnosis, order tracking