融合CNN和ViT的声信号轴承故障诊断方法

Fault diagnosis method for bearing based on fusing CNN and ViT

针对轴承故障诊断任务数据量少、故障信号非平稳等特点,提出一种短时傅里叶变换、卷积神经网络和视觉转换器相结合的轴承故障诊断方法。首先,利用短时傅里叶变换将原始声信号转换为包含时序信息和频率信息的时频图像。其次,将时频图像作为卷积神经网络的输入,用于隐式提取图像的深层特征,其输出作为视觉转换器的输入。视觉转换器用于提取信号的时间序列信息。并在输出层利用Softmax函数实现故障模式的识别。试验结果表明,该方法对于轴承故障诊断准确率较高。为了更好解释和优化提出的轴承故障诊断方法,利用t-分布领域嵌入算法对分类特征进行了可视化展示。

Here,aiming at characteristics of low data volume and non-stationary fault signals in bearing fault diagnosis tasks,a bearing fault diagnosis method combining short-term Fourier transform(SFT),convolutional neural network(CNN)and vision transformer(ViT)was proposed.Firstly,the original acoustic signal was transformed into a time-frequency image containing timing information and frequency information using SFT.Secondly,the time-frequency image was taken as input of CNN to implicitly extract deep features of the image,and CNN output was taken as input of ViT.ViT was used to extract signal time series information.In ViT output layer,Softmax function was used to identify bearing fault modes.The experimental results showed that the proposed method has a higher accuracy in diagnosing bearing faults.In order to better explain and optimize the proposed bearing fault diagnosis method,the t-distributed stochastic neighbor embedding algorithm was used to visualize classification features.