摘要
针对风电机组轴承故障诊断中特征提取困难,模型迭代速度慢,精度低的问题,该文提出一种基于改进二值化神经网络(BNN)的风电机组轴承故障诊断方法。首先采用格拉姆角场(GAF)将轴承振动信号转换为二维图像,以提高特征提取精度,然后结合深度残差网络和注意力机制构建BNN-RA(BNN+Residual Network+Spatial attention network structure)故障诊断模型,实现轴承的高效故障诊断,最终通过美国凯斯西储大学(CWRU)与江南大学(JNU)公开的轴承数据集进行方法有效性验证。结果表明,该方法可有效提高网络迭代速度和诊断精度,模型在CWRU轴承数据集单一工况下迭代11次可达到收敛,故障诊断准确率达到99.20%,在两数据集的不同工况下平均准确率可达98.46%与97.6%。
To address the challenges of feature extraction,slow model iteration,and low accuracy in diagnosing faults in wind turbine bearings,this paper introduces a diagnostic approach based on an enhanced version of the Binarized Neural Network(BNN)methodology.Firstly,the Gramian Angular Field(GAF)is utilized to transform the bearing vibration signal into a two-dimensional image,improving the accuracy of feature extraction.Next,the BNN-RA model(BNN+Residual Network+Spatial Attention Network)is constructed by integrating a deep residual network with an attention mechanism,enabling efficient fault diagnosis for bearings.The results demonstrate that the proposed method significantly enhances both network iteration speed and diagnostic accuracy.Specifically,the model achieves convergence after only 11 iterations under a single working condition of the CWRU bearing dataset,with fault diagnosis accuracy reaching 99.20%.Furthermore,the average accuracy across the two datasets are 98.46%and 97.60%under different operating conditions,respectively.
作者
余萍
宋紫琼
曹洁
陈息良
Yu Ping;Song Ziqiong;Cao Jie;Chen Xiliang(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Key Laboratory of Gansu Advanced Control for Industrial Processes,Lanzhou University of Technology,Lanzhou 730050,China;National Demonstration Center for Experimental Electrical and Control Engineering Education,Lanzhou University of Technology,Lanzhou 730050,China)
出处
《太阳能学报》
北大核心
2025年第3期643-651,共9页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(62241307)
甘肃省科技计划项目(22YF7FA166)
兰州市科技计划项目(2023-RC-26)。
关键词
风电机组
故障诊断
轴承
二值化神经网络
格拉姆角场
wind turbines
fault diagnosis
bearing
binarized neural networks
Gramian angular field
