期刊文献+

采用非平衡小样本数据的风机主轴承故障深度对抗诊断 被引量:29

A Deep Adversarial Diagnosis Method for Wind Turbine Main Bearing Fault With Imbalanced Small Sample Scenarios
下载PDF
导出
摘要 风机主轴承振动信号中存在噪声干扰,且实验环境下获取众多故障类型与故障程度数据难度大、成本高。为提高高噪声环境下基于小样本非平衡振动数据的风机主轴承故障诊断准确率,提出采用改进辅助分类生成对抗网络(auxiliary classifier generative adversarial networks,AC-GAN)的风机主轴承故障诊断新方法。首先,在AC-GAN生成器中加入Dropout层,防止过拟合导致生成重复的样本数据,保证样本生成质量。之后,在AC-GAN判别器加入卷积层,提取更多细节特征,并引入噪声过渡模型、重定义损失函数,提高判别器抗噪能力。然后,为训练样本添加标签约束,使生成器针对性生成大量符合真实样本概率分布特性的非平衡场景下小样本故障类型数据,由此,实现判别器增强。最后,通过判别器与生成器博弈达到平衡,提高小样本非平衡场景下故障识别准确率。实验表明,在高噪声干扰、样本数量不足及不同类型样本训练集规模非平衡等复杂场景下,新方法依然能够保持良好的主轴承故障识别准确率。 Wind turbine main bearing vibration signal has noise interference in the acquisition process,and it is difficult and costly to obtain data of many fault types and different fault degrees in the experimental environment.In order to improve the accuracy of wind turbine main bearing fault diagnosis based on small and imbalanced sample vibration data under high noise environment,a new fault diagnosis method of fan bearing based on improved auxiliary classifier generative adversarial networks(AC-GAN)was proposed.Firstly,Dropout layers were added to the AC-GAN generator to prevent the generation of duplicate sample data caused by over-fitting and ensure the quality of sample generation.After that,convolutional layers were added to AC-GAN discriminator to extract more detailed features and introduce noise transition model,redefine loss function to improve the anti-noise ability of discriminator.Then,label constraints were added to the training samples to enable the generator to generate a large number of targeted small-sample fault type data in non-equilibrium scenarios consistent with the probability distribution characteristics of real samples,so as to enhance the discriminator.Finally,the game between discriminator and generator was balanced to improve the fault identification accuracy in the imbalance scenario with small samples.Experiments show that the new method can still maintain good main bearing fault identification accuracy in complex scenarios such as high noise interference,insufficient samples and imbalanced training sets of different types of samples.
作者 黄南天 杨学航 蔡国伟 宋星 陈庆珠 赵文广 HUANG Nantian;YANG Xuehang;CAI Guowei;SONG Xing;CHEN Qingzhu;ZHAO Wenguang(Key Laboratory of Modern Power System Simulation and Control&Renewable Energy Technology,Ministry of Education(Northeast Electric Power University),Jilin 132012,Jilin Province,China)
出处 《中国电机工程学报》 EI CSCD 北大核心 2020年第2期563-574,共12页 Proceedings of the CSEE
基金 国家重点研发计划项目(2016YFB0900104) 吉林省产业技术开发专项(2019C058-8).
关键词 风机主轴承故障 小样本 非平衡 生成对抗性网络 辅助分类器 wind turbine main bearing fault small sample imbalanced generative adversarial networks auxiliary classifier
  • 相关文献

参考文献12

二级参考文献208

共引文献464

同被引文献406

引证文献29

二级引证文献196

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部