Abstract: The feature signal of rolling bearing is weak and affected by transfer path attenuation and environmental noise in early failure period，so it is difficult to identify fault. In order to solve this problem，a diagnosis method based on continuous wavelet transform for incipient fault of bearing was proposed. Firstly，the original signal was processed by continuous wavelet transform and different scale of wavelet coefficients were used to reconstruct the signal，then the corresponding scale signal components could be obtained，in order to acquire signal component which contains fault information as much as possible，merging process，which was guided by kurtosis criterion，was performed on reconstructed signals and correlation coefficent criterion was used to eliminate redundant signals. The signal component whose kurtosis was maximum was selected from the reserved signal components and was regarded as the best component. Envelope demodulation operation was performed on the best component further. Finally，the fault type of bearing could be judged by analyzing the envelope spectrum. Both simulated and measured signals were processed by proposed method and weak feature information was extracted successfully. The results show the proposed method could diagnose the incipient fault of rolling bearing precisely.

Key words: Rolling bearing；Incipient fault；Continuous wavelet transform；Kurtosis；Correlation coefficient

摘要： 滚动轴承早期失效阶段，特征信号微弱，并且受传递路径衰减及环境噪声影响，故障识别相对困难。针对这一问题，提出一种基于连续小波变换的轴承早期故障诊断方法。对原始信号进行连续小波变换，利用不同尺度小波系数进行信号重构，从而得到相应尺度下的信号分量，为了获取包含尽可能多的故障信息的信号分量，以峭度为指导标准对重构信号分量做合并处理，并利用相关系数准则剔除冗余信号分量，从保留信号分量中筛选出峭度值最大的分量，将其作为最佳分量用于进一步包络解调运算，通过分析包络谱判断轴承的故障类型。利用所述方法处理轴承早期故障仿真及实测信号，均成功提取出微弱特征信息，由此表明该方法可实现滚动轴承早期故障的精确诊断。

关键词: 滚动轴承；早期故障；连续小波变换；峭度；相关系数