噪声背景下的MSECAE轴承故障诊断方法研究

Research on MSECAE Bearing Fault Diagnosis Method Under Noise Background

针对传统基于深度学习的故障诊断方法存在特征提取尺度单一、抗噪声能力差的问题,提出一种多尺度卷积自编码器融合高效通道注意力机制的方法(MSECAE)用于轴承故障诊断。首先,使用傅里叶变换对原始数据进行归一化处理,将原始的一维振动信号转换到频域进行表示,有利于模型进行提取特征。其次,构造MSECAE结构,利用多尺度卷积(MSCNN)提取原始信息中的多尺度特征,使用高效通道注意力机制(ECA)动态选择卷积核大小,根据各个通道中特征的重要程度赋予不同的权重。最后通过卷积解码器对融合后的信息进行特征重构,利用Softmax分类器进行故障类别分类。为了验证所提模型的性能,使用2种不同采样频率下的数据集在4种不同噪声条件下进行多次实验,并和其他模型进行对比。实验结果表明,所提模型与其他模型相比,分类精度达到99%以上,具有更好的泛化能力和更强的鲁棒性。

In response to the problems of single scale feature extraction,poor noise resistance in traditional fault diagnosis based deep learning,a multi-scale convolutional autoencoder fusion efficient channel attention mechanism(MSECAE) method for fault diagnosis of bearings is proposed.Firstly,the Fourier transform is used to normalize the raw data,the original one-dimensional vibration signal is converted to the frequency domain for representation,which is beneficial for the model to extract features.Secondly,the MSECAE structure is constructed,and multi-scale convolution(MSCNN) is used to extract multi-scale features from the original information.An efficient channel attention mechanism(ECA) is used to dynamically select the size of the convolution kernel,and different weights are assigned based on the importance of features in each channel.To verify the performance of the proposed model,multiple experiments are conducted using datasets with two different sampling frequencies under four different noise conditions,and compared with other models.The experimental results show that compared with other models,the proposed model has a classification accuracy of over 99%,better generalization ability,and stronger robustness.