摘要
风电机组滚动轴承的振动信号存在非线性、非平稳的特性,且其特征不易被提取,针对这一问题,提出了一种基于S变换、卷积神经网络、双向门控循环单元的滚动轴承故障诊断方法(即基于S-CBiGRU的诊断方法)。首先,利用S变换对风场采集的振动信号进行了多分辨率时频分析,将一维振动信号转化为包含时间与空间特征信息的二维时频图像;然后,将经S变化所得到的时频图输入到CBiGRU网络模型中,采用CNN卷积池化层提取了振动信号的空间特征;其次,采用BiGRU结构提取了振动信号中的时间序列特征;最后,为了对上述诊断方法的有效性进行验证,采集了风电机组轴承实验数据,并将其输入到该模型中进行诊断实验。实验结果表明:在风电机组轴承故障诊断中,采用S-CBiGRU方法准确率达到93.17%,分类效果优于其他深度学习算法。研究结果表明:S-CBiGRU故障诊断方法具有可行性,可以为风电机组滚动轴承的故障诊断提供一种新途径。
For wind turbine rolling bearing vibration signal,there were problems with non-linear,non-smooth characteristics,which led to features that were not easy to be extracted.The S-CBiGRU fault diagnosis method based on S-transform,convolutional neural network(CNN),and bidirectional gated recurrent unit(BiGRU)was proposed.Firstly,in order to improve the accuracy of wind turbine rolling bearing fault diagnosis,the S-transform was used to carry out multi-resolution time-frequency analysis on the vibration signals collected from wind farms.The one-dimensional vibration signals were transformed into two-dimensional time-frequency images containing both temporal and spatial feature information.Secondly,the time-frequency maps obtained from the S-transform were input into the CBiGRU network model,and the spatial features of the vibration signals were extracted by the CNN convolutional pooling layer.Followed by the BiGRU structure,the time series features of the vibration signal were extracted.Finally,in order to verify the effectiveness of the above bearing fault diagnosis method,the experimental data of the wind turbine bearing was collected and input into the model for diagnosis experiments.The diagnostic results show that the accuracy of the S-CBiGRU method in wind turbine bearing fault diagnosis reaches 93.17%,the classification effect is better than other deep learning algorithms.The results show that the S-CBiGRU fault diagnosis method is feasible and provides a new way to diagnose rolling bearing faults in wind turbines.
作者
史宗辉
陈长征
田淼
安文杰
孙鲜明
SHI Zong-hui;CHEN Chang-zheng;TIAN Miao;AN Wen-jie;SUN Xian-ming(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China;Liaoning Vibration and Noise Control Engineering Research Center,Shenyang 110870,China;Ningbo Kunbo Measurement and Control Technology Co.,Ltd,Ningbo 315200,China)
出处
《机电工程》
CAS
北大核心
2023年第2期232-238,共7页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(51675350)。
关键词
时频分析
空间特征
时间序列特征
S变换
卷积神经网络
双向门控循环单元
time-frequency analysis
spatial features
time series features
S transform
convolutional neural network(CNN)
bidirectional gated recurrent unit(BiGRU)
