基于SK-NLM包络的滚动轴承故障冲击特征增强

非局部均值算法(NLM)是活跃于图像信号处理领域的一种新方法,因其良好的去噪特性,近几年来在滚动轴承故障诊断领域也开始获得应用。NLM利用样本点邻域窗口包含的局部结构为基本单元,通过对相似成分加权运算后取其平均值以达到抑制噪声干扰、突出故障冲击特征的目的。但对于强噪声条件下的低信噪比信号而言,NLM滤波效果并不理想。提出一种结合谱峭度(SK)和NLM权重包络谱的故障诊断方法,首先对原始信号进行SK分析得到最优中心频率及带宽构成最优滤波器,初步消除环境干扰及测量噪声;其次对NLM算法进行改进,不再以滤波信号为分析对象,而是直接利用NLM加权运算得到的信号样本点权值分布曲线作为预处理信号的包络信号,从权重角度使故障冲击得到二次增强,消除SK带通滤波器的带内噪声;最后对权值分布曲线进行包络谱分析,进而得到诊断结果。通过仿真信号、实验室信号及工程实际信号分析对所提方法进行了验证,并与最小熵解卷积(MED)进行了对比。

Rolling bearing fault impact feature enhancement based on spectral kurtosis and non-local means

The non local means (NLM) algorithm is a new emerging method that attracts significant attention in the field of image signal processing; due to its good de-noising characteristic it has gained application in the field of rolling bearing fault diagnosis recent years. NLM takes the local structure contained in the neighborhood window of the sample points as basic unit, through the weighting operation of similar components and then obtaining its average, the noise interference is suppressed, and the goal of emphasizing the fault impact feature is achieved. However, for the signal with low SNR under strong noise condition, the filtering effects of the NLM algorithm is not satisfactory, a hybrid fault diagnosis approach combining spectral kurtosis (SK) and non local Means (NLM) weight envelop spectrum is proposed. The approach consists of the following three main steps. Firstly, the SK analysis is conducted on the original signal to obtain the optimal center frequency and bandwidth, and construct the optimal filter, which primarily eliminate the environmental disturbance and measurement noise. And then the NLM algorithm is improved, the filtered signal is no longer used as the analysis object, the sample point weight distribution curve obtained using NLM weighting operation is directly adopted as the envelop signal of the preprocessed signal. From the weight point of view the fault impact is enhanced secondarily, the in-band noise of the SK band pass filter is eliminated. Finally, the envelop spectrum analysis is performed on the weight distribution curve, and the diagnosis result is obtained. The simulation signal, laboratory signal and engineering application signal of rolling element bearings with different kind of faults were analyzed to verify the proposed approach, and the results were compared with those obtained using the minimum entropy deconvolution (MED) method.