摘要
针对风力发电机轴承实际运行工况下故障类别随时间的推移逐步积累的问题,提出一种改进的具有增量学习能力的宽度卷积注意网络(IBCAN)的故障诊断方法,可在不重新训练模型的基础上诊断新增故障类别。首先,将风力发电机轴承振动信号利用连续小波变换(CWT)提取时频特征;其次,针对历史故障类别数据集,利用卷积注意网络(CAN)获得风力发电机轴承振动信号小波变换图的深度特征表示;然后,利用弹性网回归改进宽度学习系统(IBLS)将CAN所获特征和相应标签传输到IBLS中进行分类;最后,针对新增故障类别数据集,通过IBLS的扩展节点进行增量学习,进而实现新增故障类别诊断。通过实际采集的风力发电机轴承数据对所提方法进行试验验证,并与其他方法进行对比,结果表明,该方法能有效地更新风力发电机轴承故障诊断模型,增量学习新故障类别,对实际工程中风力发电机轴承故障诊断研究具有重要意义。
Aiming at the problem that the fault categories of wind turbine bearings accumulate gradually over time under actual operating conditions,this paper proposes an improved fault diagnosis method with incremental learning capability of Width Convolutional Attention Network(IBCAN),which can diagnose new fault categories without retraining the model.Firstly,the time-frequency characteristics of wind turbine bearing vibration signals are extracted using continuous wavelet transform(CWT).Secondly,the convolutional attention network(CAN)will be used to obtain the deep feature representation of the wavelet transform map of the bearing vibration signal for historical fault category datasets.Then,using elastic network regression to improve the width learning system(IBLS)and transfer the features and corresponding labels obtained from CAN to IBLS for classification.This paper conducts experimental verification on the proposed method through the collected bearing data of wind turbines under actual operating conditions and compares it with other methods.The results show that this method can effectively update the fault diagnosis model of wind turbine bearings,incrementally learn new fault categories,and is of great significance for the research of wind turbine bearing fault diagnosis in practical engineering.
作者
和林芳
王道涵
田淼
安文杰
孙鲜明
He Linfang;Wang Daohan;Tian Miao;An Wenjie;Sun Xianming(School of Safety Science and Engineering,Liaoning Technical University,Huludao 123032,China;School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China;Ningbo Kunbo Measurement and Control Technology Co.,Ltd.,Ningbo 315200,China)
出处
《太阳能学报》
北大核心
2025年第1期97-104,共8页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(51675350
51575361)。
关键词
风力发电机
轴承
故障诊断
增量学习
卷积神经网络
wind turbines
bearings
fault diagnosis
incremental learning
convolutional neural network
作者简介
通信作者:王道涵(1971-),男,博士、教授,主要从事生态经济学和城市生态学方面的研究。wangdaohan@sina.com。
