基于迁移学习的轴承故障诊断
Bearing fault diagnosis of cross-working grinding mill based on transfer learning
傅锦涛 1张弓 2张树忠2
作者信息
- 1. 福建理工大学 机械与汽车工程学院,福建 福州 350118
- 2. 福建理工大学 机械与汽车工程学院,福建 福州 350118;广州先进技术研究所 前瞻科学与技术研究中心,广东 广州 511458
- 折叠
摘要
提出一种基于多核最大均值差异的一维卷积迁移学习方法.利用一维卷积网络直接从原始振动信号中提取故障特征信息;应用对抗策略迁移技术辅助网络提取两个域之间的共同特征;以多核最大均值差异作为评价源域和目标域的距离指标,实现域不变特征提取并在凯斯西储大学轴承数据集的 4 种工况下进行迁移学习.研究表明,相比于传统方法,该文所提方法在故障分类精度上提高了6%,具有良好的应用前景.
Abstract
A one-dimensional convolution transfer learning method based on multi-core maximum mean diffe-rence is proposed.Firstly,one-dimensional convolutional networks are utilized to directly extract fault feature information from the original vibration signals.Secondly,an adversarial strategy migration technique is em-ployed to assist the network in extracting common features between the two domains.Finally,the multi-core maximum mean difference is used to evaluate the distance between the source domain and target domain,ena-bling extraction of domain invariant features and facilitating transfer learning under four working conditions of the bearing dataset from Case Western Reserve University.Compared with traditional methods,the proposed approach can enhance fault classification accuracy by 6%,which has a good application prospect.
关键词
轴承故障/迁移学习/多核最大均值差异/故障诊断Key words
bearing fault/transfer learning/multi-core maximum mean difference/fault diagnosis引用本文复制引用
基金项目
福建省智能加工技术及装备重点实验室开放基金(KF-01-22005)
福建省2022年中央引导地方科技发展资金项目(2022L3014)
出版年
2024
NETL
NSTL
万方数据