IEWT-CS和LCNN在轴承故障诊断中的应用

Application of IEWT-CS and LCNN in bearing fault diagnosis

针对传统轴承故障诊断方法易受噪声干扰、过度依赖专家经验和故障信号特征提取与优化选择困难的问题,本文提出了一种基于改进经验小波变换与压缩感知联合降噪结合导联卷积神经网络的轴承故障诊断方法。采用压缩感知方法减弱轴承信号强背景噪声干扰;采用改进经验小波变换算法将信号分解为若干本征模态函数,并通过相关系数-峭度准则选出故障特征较为明显的分量并重构;将重构信号输入导联卷积神经网络中进行自动特征提取与故障识别。轴承故障诊断实验表明:提出方法受先验知识和主观影响较小,避免了复杂的特征提取与分类过程,相较于其他方法具有更高的泛化能力、特征提取能力和故障识别能力。

Given that traditional methods for rolling bearing fault diagnosis largely depend on prior knowledge of experts,are difficult to apply to fault feature extraction and selection,and can be easily disrupted by noise,on the basis of the improved empirical wavelet transform and compressive sensing(IEWT-CS)joint denoising and lead convolution neural network(LCNN),we propose a method for bearing fault diagnosis.First,CS is used to restrain the interference of strong background noise of bearing signals.Subsequently,the denoised signal is decomposed into intrinsic mode functions using the IEWT,and the most qualified components are selected to reconstruct signals using the correlation coefficient and kurtosis criteria.Finally,the reconstructed signal is inputted into the LCNN for automatic feature extraction and fault identification.The experimental results indicate that the proposed method is less influenced by prior and subjective knowledge and avoids the complicated feature extraction and classification process.Thus,the proposed method has better generalization,feature extraction,and fault recognition capabilities than other methods.