结合频谱聚类与经验小波的轴承故障诊断方法

Bearing Fault Diagnosis Method Using Spectral Clustering and Empirical Wavelet

实测轴承振动信号就有非平稳、非线性特征,因此,对该类信号的分析需要进行解调得到特征频率,在众多解调法中包络分析是最为常用的方法;为了使解调结果更加清晰,常在解调前进行滤波,达到滤除干扰成分可有效提升解调的效果。经验小波变换提供了基于频带划分的小波滤波框架,划分后频带可滤除部分干扰信号,突出故障信号。对此,受“箱型图”和层次聚类法的启发,对“突出值”聚类法进行频带划分,通过平方包络互相关系数选取合理的频带划分个数。最后选取平方包络峭度值最大的滤波子信号进行Teager能量算子解调,获取特征频率。文章针对不同工况下的不同故障类型轴承运行数据进行分析,验证算法的有效性。特别地,在复合故障分析中,利用动态阈值法到达分别突出不同轴承故障频率的效果。

The measured bearing vibration signals are usually non-stationary and non-linear,so the demodulation is necessary to obtain the frequency characteristic frequency.Among lots of demodulation methods,envelope analysis is the most popular one.When using the envelope analysis demodulation method,filtering is necessary to wipe out irrelevant signal components which can effectively improve the demodulation effect.Empirical wavelet transform provides a wavelet filter framework based on frequency band division and it can achieve the purpose of filtering out the interfering signals and highlight fault signals.Inspired by box-plot andhierarchical clustering,the method of"outliers"clustering is proposed for frequency band division,and reasonable number of frequency band division is selected by means of cross correlation coefficient.Finally,the filter signal with the maximum square envelope kurtosis value is selected for the square envelope demodulation to obtain the characteristic frequency employing the Teager energy operator.The validity of the algorithm is verified by analyzing the measured data of the failure bearingsof different kinds under different working conditions collected from a test bed.Specially,dynamic threshold is used to highlight the characteristic frequencies of different bearing faults.