基于迭代K均值聚类的改进谱峭度方法及在滚动轴承故障诊断中的应用

Improved spectral kurtosis method based on iterative K-means clustering and its application in rolling bearing fault diagnosis

针对快速谱峭度指标对偶发性冲击过于敏感,从而导致故障诊断错误的问题,提出一种基于迭代二均值聚类的改进谱峭度算法。将信号样本在时域分段后,根据数据片段的峭度特征,使用二均值聚类算法识别被偶发性冲击污染的片段并剔除,最后基于正常片段统计信号样本的改进峭度。为解决聚类簇数难以预先确定的问题,提出一种迭代二均值聚类算法,以正常片段的峭度相对极差为迭代终止判据,逐次剔除被污染的数据片段。对原信号进行双树复小波包分解,计算各重构子信号的改进峭度生成谱峭度图,检测最大改进峭度值对应频段进行包络解调分析诊断轴承故障。仿真和实验结果表明,该方法能够在多个不同强度的偶发性冲击干扰下准确识别循环瞬态冲击,识别效果优于其他算法。

The kurtosis index of Fast Kurtogram is too sensitive to sporadic impulse,which causes FK fault diagnosis to fail.To overcome this problem,this paper presents an improved spectral kurtosis algorithm based on iterative two-mean clustering.After segmenting the signal samples in the time domain,according to the kurtosis characteristics of the data fragments,the two-mean clustering algorithm is used to identify and eliminate the fragments contaminated by sporadic impulse.Finally,the improved kurtosis of the signal samples is calculated based on the normal fragments.In order to solve the problem that the number of clusters is difficult to determine in advance,it is proposes an iterative two-mean clustering algorithm,which uses the ratio of the maximum kurtosis divided by the minimum kurtosis in the normal fragments as the iterative termination criterion,and successively eliminates anomalous fragments containing sporadic impulse.Perform dual-tree complex wavelet packet decomposition on the original signal,calculate the improved kurtosis of each reconstructed sub-signal,generate a spectral kurtosis map,and detect the frequency band corresponding to the maximum improved kurtosis value to perform envelope demodulation analysis to diagnose bearing faults.Simulation and experimental results show that this method can accurately identify cyclic transient impulse under the interference of multiple sporadic impulse of different intensities,and the identification effect is better than other algorithms.