针对滚动轴承故障诊断的新时频特征提取方法

A New Time-frequency Domain Feature Extraction Method for Rolling Bearing Fault Diagnosis

针对滚动轴承的故障诊断,提出了一种基于词包模型和短时傅里叶变换的特征提取方法。根据轴承故障的产生机理,不同轴承的振动信号在频域上会有相应的能量分布规律,然而在实际现场中,信号干扰或者生产环境等因素会弱化这种规律性,使得在频谱上难以准确看到相应分布特征。当采用词包模型时,把每一时间帧下能量在频率维度上的分布看成一个单词,则每段信号就表示成了由各个单词组成的一篇篇文档,这就可以直接从数据的角度去揭示能量分布的这种规律性。然后,以词包模型处理后的结果作为特征向量,用SVM分类算法诊断出结果。最后用无锡某汽车生产线SQI-MFS实验平台和美国凯斯西储大学的轴承振动数据进行了实验,实验验证了该方法比时域特征(RMS)和时频域特征(WE&WEE)的诊断结果精确,可以在滚动轴承故障诊断领域展开应用。

For fault diagnosis of rolling bearings,a new feature extraction strategy based on short time Fourier transform（ STFT） and bag of wordss（ BOW） is proposed. Based on the generate mechanism of bearing fault,the different bearing vibration signals have relevant energy distribution. But in the factory,some factors like signal interference or environment noise will destroy the energy distribution. When using BOW,it regards the distribution of energy in frequency domain each time frame as a word,so segments of signal will be documents which are made up of many words. It shows the energy distribution directly in data perspective. Then,with the new features and SVM classifier,the results of fault diagnosis can be known. At last,effectiveness of the proposed method is verified,vibration from SQI- MFS platform and CWRU platform are analyzed. The results in experiments shows that this method is better than RMS and WEWEE. So the new feature can be used in fault diagnosis area.