摘要
风电机组主轴承是机组的关键部件,维修成本较高。为提供精准诊断支持,基于生成对抗网络(GAN)提出改进的故障诊断方法。表明了原始GAN模型和轴承数据集。在风电场极微弱负样本数据背景下,通过堆叠多个约束稀疏自动编码器(CSAE)形成深度学习网络框架,用于提取样本的深度特征,是一种构造具有更优表达能力函数的方法。将GAN与DCSAE相结合,以多层感知网络作为生成网络模型,DCSAE作为鉴别网络模型,形成轴承劣化模型的训练方法。更新优化劣化模型,得到高精度轴承劣化诊断模型。在辅助数据集与目标数据集间再次迁移学习,形成泛化能力和鲁棒性较强的诊断模型,进而实现轴承劣化状况的自适应诊断。经算例分析,所提出的改进方法可实现原始数据分布特征高效学习的目标,能够构建少数类故障数据分布模型,在不同小样本情境下,改进后的模型仍表现出较优的诊断能力。
The wind turbine main bearing is the key component of the unit,with high maintenance costs.In order to provide accurate diagnosis support,an improved fault diagnosis method is proposed based on the generative adversarial network(GAN).The original GAN model and bearing data set are described.In the background of extremely weak negative sample data of wind farm,a depth learning network framework is designed by stacking multiple constrained sparse automatic encoders(CSAE)to extract the depth features of samples,which is a method for constructing functions with better expressibility.Combining GAN with DCSAE,the multilayer perception network is used as generating network model,and DCSAE is used as discriminating network model to form the training method of bearing deterioration model.By updating the optimized deterioration model,a high-precision diagnosis model of bearing deterioration is obtained.The adaptive diagnosis of bearing deterioration can be realized by transferring the auxiliary data set and the target data set to form a diagnosis model with strong generalization ability and robustness.Through the example analysis,by using the proposed improved method,the goal of efficient learning of the distribution features of the original data can be realized,and the distribution model of a few kinds of fault data can be constructed,the improved model still shows better diagnostic ability in different small sample scenarios.
作者
颜毅斌
陈清化
吉天平
李晟方
Yan Yibin;Chen Qinghua;Ji Tianping;Li Shengfang(Hunan Railway Institute of Science and Technology,Zhuzhou,Hunan 412006,China)
出处
《机电工程技术》
2024年第2期103-106,共4页
Mechanical & Electrical Engineering Technology
基金
湖南省自然科学基金资助项目(2022JJ60074)
湖南省教育厅资助科研项目(20C1226)。
关键词
生成对抗网络
风电机组
主轴承
故障诊断
generative adversarial network(GAN)
wind turbine
main bearing
fault diagnosis
