基于对抗学习的机械故障迁移诊断方法及系统

Translated title of the contribution: A novel mechanical fault transfer diagnosis method based on adversarial learning

Ming Zhang (Inventor), Jun Yang (Inventor), Weining Lu (Inventor), Zhang Chen (Inventor), Bin Liang (Inventor)

Research output: Patent

Abstract

本发明公开了一种基于对抗学习的机械故障迁移诊断方法及系统,其中,该方法包括:获取不同工况下机械故障的原始信号进行分析生成不同工况下带标签的源域训练数据集、不带标签的源域训练数据集和目标域测试数据集;根据带标签的源域训练数据集和反向传播算法训练深度卷积神经网络模型生成故障诊断模型;根据不带标签的源域训练数据集和目标域测试数据集对故障诊断模型进行训练;根据带标签的源域训练数据集和反向传播算法对训练后的故障诊断模型进行微调;将不带标签的目标域测试数据集输入微调后的故障诊断模型,输出待测试样本的故障类别。该方法通过对抗学习方法获得域不变特征,实现不同域之前的迁移,实现了对变工况机械故障的智能诊断。
Translated title of the contributionA novel mechanical fault transfer diagnosis method based on adversarial learning
Original languageChinese
Patent number201910289486.8
Publication statusPublished - 28 Jun 2019

Fingerprint Dive into the research topics of 'A novel mechanical fault transfer diagnosis method based on adversarial learning'. Together they form a unique fingerprint.

Cite this