基于DFT与ECA的滚动轴承故障诊断

Fault Diagnosis of Rolling Bearing Based on Discrete Fourier Transform and Efficient Channel Attention

针对滚动轴承故障诊断中传统卷积神经网络(convolutional neural networks,简称CNN)提取特征的感受野受限于卷积核大小的问题,提出了一种结合离散傅里叶变换(discrete Fourier transform,简称DFT)和高效通道注意力(efficient channel attention,简称ECA)的卷积神经网络模型(convolutional neural network combining discrete Fourier transform and efficient channel attention,简称DFT-ECANet)。首先,将原始振动信号通过DFT变换到频域,在频域上经卷积和离散傅里叶逆变换(inverse discrete Fourier transform,简称IDFT)转换到时域,使信号在时域上具有全局的感受野;其次,将该信号与经过卷积的数据在通道维度上进行拼接,通过ECA为各通道数据分配权重,并关注诊断性能高的特征;最后,通过多个卷积-池化层进一步提取模型深层特征,结合池化层和全连接层诊断轴承故障。实验结果表明:DFT-ECANet在原始振动数据集上具有较高的诊断精度和较好的泛化性能,通过T分布随机近邻嵌入(T-distributed stochastic neighbor embedding,简称T-SNE)可降维可视化模型的诊断过程;在强噪声干扰下仍能保持较高的精度,具备较强的鲁棒性和抗噪性能。

Aiming at the problem that the feature receptive field extracted by traditional convolutional neural network(CNN)in rolling bearing fault diagnosis is limited by the shape of convolution kernel,a network(DFT-ECANet)combining discrete Fourier transform(DFT)and efficient channel attention(ECA)is proposed.Firstly,convert the original vibration signal into the frequency domain through DFT,and transform it into the time domain through convolution in the frequency domain and make the signal has a global receptive field in the time domain through inverse discrete Fourier transform(IDFT);then,concatenate the signal with the convoluted data on the channel dimension,assign weight to each channel data through ECA,focusing on features with high diagnostic performance;finally,the deep features of the model are further extracted through several convolution-pooling pairs,and the fault diagnosis of rolling bearing is performed by linking the pooling layers and the fully connected layer.The experimental results show that DFT-ECANet has high diagnostic accuracy and good generalization performance on the original vibration datasets,and the diagnostic process of the model is visualized through T-SNE dimensionality reduction;it can still maintains high accuracy,robustness and anti-noise property under fierce noise interference.