基于DR-VAE与自注意力机制的电机轴承故障诊断

Fault Diagnosis for Motor Bearings Based on DR-VAE and Self-Attention Mechanism

为提高电机轴承故障诊断的精度,提出一种深度残差变分自编码器与自注意力机制相结合(DR-VAE-SAM)的电机轴承故障诊断方法。首先,在数据预处理中使用矢量量化变分自编码(VQ-VAE)对轴承故障数据进行增强,训练完成后将生成的故障样本混入原始样本中以平衡和增强数据集;然后,将深度残差网络与变分自编码器结合,增加了迁移学习中最大均值差异(MMD)作为融合标准;最后,在编、解码过程中引入自注意力机制提取特征的关键信息。通过西储大学和东南大学的轴承数据进行试验验证,结果表明,在故障数据少、样本不平衡以及变工况情况下,DR-VAE-AM模型能够有效提高故障诊断精度,并有较好的泛化能力。

In order to improve the accuracy of fault diagnosis for motor bearings,an fault diagnosis method is proposed based on deep resnet-variational automatic encoder and self-attention mechanism(DR-VAE-SAM).Firstly,the vector quantised variational autoencoder(VQ-VAE)is used to enhance the bearing fault data during data preprocessing,and the generated fault samples are mixed into original samples to balance and enhance the data set after training.Then,the deep resnet is combined with variational self-encoder,and maximum mean discrepancy(MMD)in transfer learning is added as fusion criterion.Finally,the self-attention mechanism is introduced to extract the key information of features during process of encoding and decoding.The test verification is carried out by bearing data from Western Reserve University and Southeast University,the results show that DR-VAE-SAM method can effectively improve the accuracy of fault diagnosis and has good generalization ability under few fault data,unbalanced samples and variable working conditions.