基于AM和CNN的多级特征融合的风力发电机轴承故障诊断方法
FAULT DIAGNOSIS METHOD OF WIND TURBINE BEARING BASED ON AM AND CNN MULTI-LEVEL FEATURE FUSION
王进花 1韩金玉 2曹洁 3王亚丽2
作者信息
- 1. 兰州理工大学电气工程与信息工程学院,兰州 730050;甘肃省工业过程先进控制重点实验,兰州 730050;兰州理工大学电气与控制工程国家实验教学中心,兰州 730050
- 2. 兰州理工大学电气工程与信息工程学院,兰州 730050
- 3. 兰州理工大学电气工程与信息工程学院,兰州 730050;甘肃省制造信息工程研究中心,兰州 730050
- 折叠
摘要
提出一种基于注意力机制的多级特征融合卷积神经网络(A2ML2F-CNN)故障诊断方法.该方法将原始电流和振动信号作为输入,首先使用基于注意力卷积神经网络(AMCNN)模块分别进行数据信号特征提取,并进行一级特征融合连接.在此基础上,再次分别采用注意力机制一维卷积神经网(AM1DCNN)和二维卷积神经网络(2DCNN)提取相关信息,并进行二级特征融合,以此来解决单传感器数据故障信息不足及互补特征难以提取的问题,最后采用全连接层和Softmax层进行分类,得到诊断结果.为验证所提方法的故障诊断效果,通过帕德伯恩数据集进行实验验证,并将其与CNN、LSTM、SVM等方法的诊断精度进行对比,相较于上述方法,该文方法的诊断准确率分别提高1.8、3.2和4.8个百分点,验证了所提方法的有效性.
Abstract
In this paper,we proposed a attention mechanism multi-level feature fusion convolutional neural network(A2ML2F-CNN)fault diagnosis method.The method takes the original current and vibration signals as inputs,firstly uses the attention mechanism convolutional neural network(AMCNN)module to extract the data signal features separately,and perform a first-level feature fusion connection.On this basis,the attention mechanism one-dimensional convolutional neural network(AM1DCNN)and the two-dimensional convolutional neural network(2DCNN)are used to extract relevant information,and perform a secondary feature fusion,to solves the problem of insufficient fault information of single-sensor data and the difficulty of extracting complementary features.Finally,the fully connected layer and the Softmax layer are used to classify and the diagnostic results are obtained.In order to verify the fault diagnosis effect of the method proposed in this paper,the Paderborn data set is used for experimental verification,and its diagnosis effect is compared with CNN,LSTM,SVM.The results showed that the diagnostic accuracy of the method in this paper increased by 1.8,3.2 and 4.8 percentage points respectively compared to the above methods,which shows the effectiveness of the method in this paper.
关键词
风力机/故障诊断/特征融合/注意力机制/卷积神经网络/风力发电机轴承Key words
wind turbines/fault diagnosis/feature fusion/attention mechanism/convolutional neural network(CNN)/wind turbine bearings引用本文复制引用
基金项目
国家自然科学基金(62063020)
国家自然科学基金(61763028)
国家重点研发计划(2020YFB1713600)
甘肃省自然科学基金(20JR5RA463)
出版年
2024
NETL
NSTL
万方数据