Abstract: Fault diagnosis is very important for the health management of mechanical equipment.At present,data-driven fault diagnosis methods have become a research hotspot in this field.However,the working status and conditions of mechanical equipment are constantly changing,which leads to different distributions of fault data and brings challenges to fault diagnosis.To solve this problem,an unsupervised fault diagnosis method was proposed based on domain adaptive neural network and joint distributed adaptive.The fault diagnosis data of different data distributions were preprocessed by the method of signal to image.Then,the domain adaptive neural network was used to generate features with similar data distribution,and finally the joint distribution adaptive method was used to process the generated features.The proposed method could effectively solve the problem of different data distribution caused by changes in working status and conditions.The generated model could more accurately diagnose the fault data sampled in another working state without a label.Using a classic case in Case Western Reserve University bearing data set,the method was tested and verified,and the experimental results proved the feasibility and effectiveness of the method.

Key words: fault diagnosis, domain adaptation neural network, joint distribution adaptive method, unsupervised learning, transfer learning

摘要： 故障诊断对于机械设备的健康管理十分重要,当前,数据驱动的故障诊断方法已成为了本领域研究热点。然而,机械设备的工作状态与条件是不断变化的,这导致故障数据分布不同,故障诊断带来了挑战。针对该问题,提出一种基于域适应神经网络与联合分布自适应的无监督故障诊断方法。首先,将不同数据分布的故障诊断数据通过信号转图像的方法进行数据预处理;然后,使用域适应神经网络生成数据分布相似的特征;最后使用联合分布自适应方法处理所生成的特征。该方法可以有效地解决工作状态与条件发生变化所带来的数据分布不同的问题。所生成的模型可以在无标签的情况下,较为准确地诊断在另一个工作状态下采样的故障数据。最后,利用本领域的经典案例——凯斯西储大学轴承数据集,对所提方法进行了测试验证,实验结果证明了该方法的可行性与有效性。

关键词: 故障诊断, 域适应神经网络, 联合分布自适应方法, 无监督学习, 迁移学习