基于多源异构信息迁移学习的融合故障诊断方法

Fusion Fault Diagnosis Method Based on Multi-Source Heterogeneous Information Transfer Learning

基于深度学习的故障诊断模型的精确度依赖于带标签的样本数量和信息使用方式。实际的工业控制获取的数据往往既有一维的信号序列又有二维的图像。基于深度学习的故障诊断方法仅利用一类数据进行故障诊断会造成信息的浪费,需要将多源异构信息进行融合。但工业控制中带标签的故障样本量很少,仅利用故障样本不能获得精度较高的故障诊断模型。迁移学习是运用已有的知识对不同但相关领域问题进行求解的一种新方法。通过迁移学习,利用在ImageNet数据集中训练好的VGG16网络作为特征抽取器,提取故障图像的特征,然后将故障图像特征和一维信号特征进行融合,以获得一个精确率较高的故障诊断模型。使用凯斯西储大学轴承数据集证明了该方法的有效性。

The accuracy of the fault diagnosis model based on deep learning depends on the number of labeled samples and the way of information using.The data obtained in industrial control often have both one-dimensional signal sequences and two-dimensional images.The fault diagnosis method based on deep learning only uses one kind of data for fault diagnosis,which will cause the waste of information.Therefore,multi-source heterogeneous information needs to be fused.However,the number of labeled fault samples in industrial control is very small,by which a high-precision fault diagnosis model could not be obtained.Transfer learning is a new method to solve problems in different but related fields using existing knowledge.Through transfer learning,the VGG16 network trained in the ImageNet data set is used as a feature extractor to extract the feature of the fault image,and then the fault image feature and one-dimensional signal feature are fused to obtain a fault diagnosis model with high accuracy.In this paper,the effectiveness of this method is proved by using the Case Western Reserve University bearing data set.