基于联合对抗深度迁移的轴承故障诊断方法

Rolling Bearing Fault Diagnosis Based on Joint Adversarial Deep Transfer Learning

长期现场监测的滚动轴承大数据往往故障样本较少且受噪声和监测误差的影响,有用故障信息容易淹没在正常样本数据中,特别对于变工况和不同场景下的数据集服从不同分布的问题,如果直接利用基于深度学习的故障诊断方法进行识别,容易造成误判或漏判。为此,提出一种基于联合对抗深度迁移的滚动轴承故障诊断方法。该方法融合了深度学习和迁移学习的各自优点。首先利用具有批量归一化层的卷积神经网络,提取实验室有标签源域数据和现场无标签目标域数据的不变特征,然后通过广义切片Wasserstein距离来度量域间数据的联合分布差异性,并采用softmax激活函数作为域判别器和分类器实现识别。最后通过凯斯西储大学滚动轴承数据集、西安交通大学滚动轴承数据集和现场滚动轴承数据集进行不同诊断方法之间的对比、验证和应用研究。研究结果表明,所提方法在滚动轴承数据集的平均分类准确率能达到99%,相比于其他方法具有更高的分类精度和更好的迁移性能,可为滚动轴承故障诊断提供一种新的有效方法。

The big data of rolling bearing monitored on site for a long time are often limited in fault samples and affected by noise and monitoring errors,so the useful fault information is easily overlooked in the normal sample data.Especially,under variable working conditions and different scenarios,data set obey different distributions.Therefore,if the deep learning based fault diagnosis method is directly used for identification,misjudgement or missing judgment easily occurs.This paper presents a method for rolling bearing fault diagnosis based on joint adversariel deep transfer learning,which combines the advantages of deep learning and transfer learning.Firstly,the convolutional neural networks(CNN)with batch normalization layer were used to extract the invariant features of laboratory labeled source domain data and field unlabeled target domain data.Then,the generalized slice Wasserstein distance was used to measure the joint distribution difference of data between domains,and the softmax activation function was used as the domain discriminator and classifier to realize recognition.Finally,the rolling bearing data set of Case Western Reserve University(CWRU),Xi'an Jiaotong University(XJTU)and field were used to conduct comparison,verification and application research of different diagnostic methods.The results show that the average classification accuracy of the proposed method in the rolling bearing data set reaches 99%,which has higher classification accuracy and better transfer performance than other methods,and provides a new and effective method for rolling bearing fault diagnosis.