基于一维卷积子域适应对抗网络的变负荷轴承故障诊断

Rolling Bearing Fault Diagnosis with Variable Load Based on One-dimensional Convolutional Subdomain Adaptive Adversarial Network

在大型旋转机械滚动轴承故障诊断的建模中,由于设备运行负载不同,若训练数据与测试数据具有分布差异,则会使训练得到的深度神经网络诊断模型的准确率下降。针对此问题,基于迁移学习理论,提出了基于一维卷积子域适应对抗网络的故障诊断方法。该方法嵌入了融合样本级权重的局部最大均值差异来促进子域对齐,并引入域判别器与特征提取器进行对抗训练,辅助提取域共性特征。建立了一种有效的跨负载轴承故障诊断模型,实现了目标域的无监督故障诊断,提高了滚动轴承故障诊断的准确性。最后,在凯斯西储大学发布的轴承故障数据集上进行实验,实验结果验证了所提方法的有效性。

In the modeling of rolling bearing fault diagnosis for large rotating machinery,if the distribution of training data and test data is different due to the different operating loads of the equipment,the accuracy of the trained deep neural network diagnosis model will decrease.To solve the problem,based on transfer learning theory,a fault diagnosis method based on one-dimensional convolutional subdomain adaptive adversarial network is proposed.The local maximum mean discrepancy fused with sample-level weights is embedded to promote the alignment of subdomains,and a domain discriminator is introduced for adversarial training with the feature extractor to assist in extracting domain common features.An effective cross-load bearing fault diagnosis model is established to realize the unsupervised fault diagnosis in the target domain and improve the accuracy of the rolling bearing fault diagnosis.Finally,experiments are carried out based on the bearing fault dataset published by Case Western Reserve University,and the experimental results verify the effectiveness of the proposed method.