基于CNN-ViT的滚动轴承故障类型识别方法

Rolling Bearing Fault Type Recognition Method Based on CNN-ViT

针对在滚动轴承的故障诊断中,一维信息无法充分挖掘数据特征的问题,提出一种基于卷积神经网络–视觉Transformer(Convolutional neural networks-vision transformer,CNN-ViT)的滚动轴承故障类型识别模型。首先将一维时域振动信号转化为二维灰度图以更好地表现数据特征,并在ViT模型基础上增加CNN对二维灰度图进行上采样,解决了挖掘数据特征不足以及ViT模型训练时的稳定性问题。通过所提模型对轴承不同故障类型及不同损伤程度进行识别。为了验证所提方法的有效性,采用某数据集进行实验验证,同时将所提方法与其他深度学习模型的诊断结果进行了对比。验证结果表明,该方法的准确率为99.4%,具有较高的精度。

Aiming at the problem that one-dimensional information can not fully mine data characteristics in rolling bearing fault diagnosis,a rolling bearing fault type recognition model based on Convolutional Neural Networks-Vision Transformer(CNN-ViT)is proposed.Firstly,one-dimensional time-domain vibration signals were transformed into two-dimensional grayscale maps to better display data characteristics;the CNN was added to the ViT model for up-sampling of two-dimensional grayscale maps,which solved the problem of insufficient mining data features and the stability of ViT model training.The proposed model was used for fault identification of different bearing fault types and different damage degrees.In order to verify the effectiveness of the proposed method,CWRU dataset is used for experimental verification,and the diagnostic results of the proposed method are compared with those of other deep learning models.The analysis shows that the accuracy of this method is 99.4%,and it has a high accuracy.