摘要
针对风电机组轴承微弱故障信号的特征提取困难和故障诊断模型性能差等问题,提出一种并行卷积神经网络的故障诊断方法。首先,利用连续小波变换将一维信号转换成二维时频特性图;其次,构造一种并行卷积神经网络结构,该结构由大卷积层和并行卷积层组成,大卷积层快速提取输入层所有特征,并行卷积层识别特征中的有效故障信息,且并行卷积层为双层小卷积并行结构;然后,采用特征融合层,融合并行卷积层2次特征提取后的故障特征,实现诊断模型内部的特征增强,降低模型复杂度;最后,经实验验证,该模型诊断轴承故障的准确率为98.25%。
In view of the problems of difficulty in bearing weak fault signal feature extraction and poor performance of fault diagnosis model for wind turbine,a fault diagnosis method based on parallel convolutional neural network is proposed.Firstly,1-Dimensional signals are converted into 2-dimensional time-frequency feature maps using continuous wavelet transform.Secondly,a parallel convolutional neural network structure is constructed,which consists of large convolutional layer and parallel convolutional layer.The large convolutional layer can quickly extract all the features of the input layer.The parallel convolutional layer is a two-layer small convolution parallel structure,which can effectively identify fault information.Then,the feature fusion layer is adopted to achieve feature enhancement inside the diagnosis model and reduce the complexity of the model,which combines the fault features extracted by two parallel convolutional layers.Finally,experimental verification showes that the fault diagnosis accuracy of the proposed model for bearing is 98.25%.
作者
孟良
苏元浩
许同乐
孔晓佳
兰孝生
李云凤
Meng Liang;Su Yuanhao;Xu Tongle;Kong Xiaojia;Lan Xiaosheng;Li Yunfeng(Mechanical Engineering School,Shandong University of Technology,Zibo 255049,China)
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2023年第5期449-456,共8页
Acta Energiae Solaris Sinica
基金
山东省自然科学基金(ZR2021ME221)。
关键词
风电机组
轴承
故障诊断
卷积神经网络
特征增强
特征可视化
wind turbines
bearing
fault diagnosis
convolutional neural network
feature enhancement
feature visualzation