机电工程技术2024，Vol.53Issue(2)：1-7,7.DOI:10.3969/j.issn.1009-9492.2024.02.001
基于卷积对抗自编码网络的轴承早期故障检测方法
Early Fault Detection of Bearings Based on Convolutional Adversarial Autoencoder Network
摘要
Abstract
Traditional statistical learning modeling methods often rely on manual feature extraction,making them less effective in accurately capturing subtle early fault signals under varying operating conditions.Aiming to address critical issues related to enhancing equipment reliability and reducing maintenance costs through early fault detection,a bearing early fault detection method based on convolutional adversarial autoencoder networks is proposed.By constructing a convolutional autoencoder network,the learning of fault information from raw signals,enhancing feature learning capabilities and training efficiency are achieved through the encoding-decoding structure.Further improvements to the convolutional autoencoder network involve the introduction of skip connections,enhancing its ability to learn complex features.The incorporation of adversarial training strategies enhances sample reconstruction capabilities and strengthens fault information.The reconstruction loss between input data and reconstructed samples is defined as a health indicator,coupled with spectrum analysis to achieve early fault detection in bearings.The proposed method is validated using data from a full-life bearing degradation test rig.Through unsupervised training with a limited number of normal samples,the method successfully applies to bearing fault detection post-training.The results show that the proposed health indicator and detection method during the healthy phase of bearings,and a good monotonicity during the degradation phase.By using this method,subtle early fault information in bearings can be effectively captured,providing a reliable means for improving equipment reliability.关键词
自编码网络/对抗学习/故障诊断/健康指标/轴承Key words
autoencoder network/adversarial learning/fault diagnosis/health index/bearings分类
信息技术与安全科学引用本文复制引用
陈祝云,焦健,纪传鹏,许维冬,贺毅,万海洋..基于卷积对抗自编码网络的轴承早期故障检测方法[J].机电工程技术,2024,53(2):1-7,7.基金项目
国家自然科学基金资助项目(52205101) （52205101）
广东省基础与应用基础研究基金(2021A1515110708) （2021A1515110708）
广州市基础研究计划基础与应用基础研究基金(202201010615) （202201010615）
鹏城实验室重大攻关项目(PCL2023A07 （）
PCL2023A09 （）
PCL2023A10) （）
