一种基于EMDFICA-CNN的滚动轴承故障诊断方法

A Fault Diagnosis Method of Rolling Bearings Based on EMDFICA-CNN

针对滚动轴承故障信号信噪比低、特征学习效率低、诊断模型结构复杂等问题,提出一种基于经验模态分解快速独立成分分析(Empirical mode decomposition fast independent component analysis,EMDFICA)与卷积神经网络(Convolution neural network,CNN)相结合的滚动轴承故障诊断方法。首先,通过经验模态分解预处理得到原始振动信号的内禀模式函数(Intrinsic mode function,IMF);然后,对提取的IMF分量进行快速独立成分分析(Fast independent component analysis,FICA)并获得特征ICA分量;最后,将得到的ICA分量基于CNN模型进行训练、测试,实现滚动轴承智能诊断分类识别。实验结果表明:利用EMDFICA对振动信号进行预处理,能够准确高效提取故障特征。通过与其他模型的诊断结果对比发现,所提出CNN模型能实现高精度故障诊断。

Aiming at the problems of low signal-to-noise ratio of rolling element bearing fault signals,low efficiency of feature learning and complex structure of diagnosis models,a rolling element bearing fault diagnosis method based on empirical mode decomposition fast independent component analysis(EMDFICA)and convolutional neural network(CNN)is proposed.Firstly,the intrinsic mode functions(IMF)of the original vibration signals are obtained through empirical mode decomposition(EMD)preprocessing method.Then,the extracted IMF components are analyzed by fast independent component analysis method and the characteristic ICA components are obtained.Finally,the ICA components are trained and tested by CNN model,and the intelligent diagnosis and classifications of rolling bearing are realized.The experimental results show that the EMDFICA method can extract the fault features accurately and effectively in the pre-processing of vibration signals.Compared with the diagnosis results of other models,the proposed CNN model can realize high precision fault diagnosis.