Aiming at the problem that the bearing fault signal is weak and usually interfered by the strong background noise, which makes the fault feature extraction very difficult, an enhanced variational mode decomposition (EVMD) technique is proposed. First, the autoregressive (AR) model was utilized to eliminate the stationary components in the signal in advance to reduce the noise interference and the maximum kurtosis of the residual signal was set as the target. Second, the maximum frequency-domain correlated kurtosis was adopted as the fitness value, and the decomposition modes K and quadratic penalty factor α in the VMD approach were adaptively selected by the whale optimization algorithm. Third, the reconstruction signal was acquired, then the enhanced envelope spectrum was employed to weaken the interference of irrelevant frequency components and the fault features of rolling element bearing could be extracted accurately. The results of simulation and experimental analysis show that the proposed algorithm can significantly reduce the noise interference and avoid the blindness selection of VMD parameters. The comparison with fix-parameter VMD and fast kurtogram approaches shows that the proposed technique can improve the effectiveness of defect signature extraction, which has a certain value for engineering application.

【摘要】：针对轴承故障信号微弱且常受强背景噪声干扰,使得故障特征提取非常困难的问题,提出了一种增强变分模态分解(EVMD)技术。首先，利用自回归（AR）模型预先消除信号中的平稳成分，以减少噪声干扰，并以残差信号的最大峰态为目标。其次，采用最大频域相关峰度作为适应度值，分解模式K和二次惩罚因子α在 VMD 方法中由鲸鱼优化算法自适应选择。第三，获取重建信号，然后采用增强包络谱减弱无关频率分量的干扰，准确提取滚动轴承的故障特征。仿真和实验分析结果表明，该算法能够显着降低噪声干扰，避免VMD参数选择的盲目性。与固定参数VMD和快速Kurtogram方法的比较表明，该技术可以提高缺陷特征提取的有效性，具有一定的工程应用价值。