欢迎访问《太阳能学报》官方网站,今天是 分享到:
ISSN 0254-0096 CN 11-2082/K

太阳能学报 ›› 2022, Vol. 43 ›› Issue (8): 281-291.DOI: 10.19912/j.0254-0096.tynxb.2020-1321

• • 上一篇    下一篇

基于AEWT-KELM的风电机组轴承故障诊断策略

齐咏生1,2, 单成成1,2, 高胜利3, 刘利强1,2, 董朝轶1,2   

  1. 1.内蒙古工业大学电力学院,呼和浩特 010080;
    2.内蒙古自治区机电控制重点实验室,呼和浩特 010051;
    3.内蒙古北方龙源风力发电有限责任公司,呼和浩特 010050
  • 收稿日期:2020-12-07 出版日期:2022-08-28 发布日期:2023-02-28
  • 通讯作者: 齐咏生(1975—),男,博士、教授、硕士生导师,主要从事风电机组状态监测与故障诊断方面的研究。qyslyt@163.com
  • 基金资助:
    国家自然科学基金(61763037,61863029); 内蒙古自然科学基金(2019LH6007); 内蒙古科技成果转化项(CGZH2018129)

FAULT DIAGNOSIS STRATEGY OF WIND TURBINES BEARING BASED ON AEWT-KELM

Qi Yongsheng1,2, Shan Chengcheng1,2, Gao Shengli3, Liu Liqiang1,2, Dong Chaoyi1,2   

  1. 1. Institute of Electric Power, Inner Mongolia University of Technology, Hohhot 010080, China;
    2. Inner Mongolia Key laboratory of Electrical and Mechanical Control, Hohhot 010051, China;
    3. Inner Mongolia North Long yuan Wind Power Co., Ltd., Hohhot 010050, China
  • Received:2020-12-07 Online:2022-08-28 Published:2023-02-28

摘要: 针对风力发电机组轴承故障振动信号传递路径复杂多变,且故障信号易受到背景噪声的严重干扰,传统方法对故障特征难以准确提取的问题,提出一种自适应经验小波变换(AEWT)与奇异值分解(SVD)的特征提取方法,并结合核极限学习机(KELM)实现风电机组轴承的故障诊断,该方法同时考虑轴承不同故障类型及不同损伤等级的情况。其中,自适应EWT为两阶段调整过程:基于尺度空间法固有模态函数(IMF)分解-确保EWT分解的有效性、基于相关系数最大的敏感分量提取-实现相关特征最大化和冗余信息的消除。通过相关实验结果可明显发现,所提AEWT的分解效果优于EMD、EEMD、CEEMDAN、LMD等方法。对提取敏感分量利用SVD计算奇异值,构建故障特征向量;最后将特征向量作为KELM的输入,建立KELM轴承状态识别模型。通过西储大学平台轴承振动信号和实际风场采集的轴承振动信号对算法进行验证,结果表明,相比SVM、ELM、KNN等识别模型,该方法能有效识别出不同故障类型及不同损伤等级下的轴承故障,整体识别率达99%。

关键词: 风电机组, 故障诊断, 轴承, 特征提取, 信号处理, 经验小波变换, 核极限学习机

Abstract: Aiming at the problem that the transmission path of the vibration signal of wind turbine bearing fault is complex and changeable, and the fault signal is susceptible to severe interference from background noise, which makes it is difficult to accurately extract fault features with traditional methods, an adaptive empirical wavelet transform (AEWT) and the singular value decomposition (SVD) feature extraction method is proposed in this paper, by which and combined with the kernel extreme learning machine (KELM), the fault diagnosis of the wind turbine bearing is realized. This method also considers the different fault types and different damage levels of the bearing. Among them, the adaptive EWT is a two-stage adjustment process: that is, based on the intrinsic mode function (IMF) decomposition of the scale space method-to ensure the effectiveness of the EWT decomposition; based on the extraction of the sensitive component with the largest correlation coefficient to maximize the relevant features and redundant information elimination. Through relevant experimental results, it can be clearly found that the decomposition effect of the proposed AEWT is better than EMD, EEMD, CEEMDAN, LMD and other methods. After that, SVD is used to calculate the singular values of the extracted sensitive components to construct the fault feature vector; finally, the feature vector is used as the input of KELM to establish the KELM bearing state recognition model. The algorithm is verified by bearing data collected on the platform of Western Reserve University and bearing data collected from actual wind farms. The results show that compared with recognition models such as SVM, ELM, and KNN, this method can effectively identify bearing faults under different fault types and different damage levels. The overall recognition ratches 99%.

Key words: wind turbines, fault detection, bearing, feature extraction, signal processing, Empirical wavelet transform, Kernel extreme learning machine

中图分类号: