摘要
为提高卷积神经网络在风力发电机组轴承故障诊断上的准确率,本文对某2MW风力发电机组轴承故障数据,进行单通道及多通道、多种诊断网络模型、不同优化算法的故障诊断分析对比,提出将多个振动传感器的数据整合为多通道一维数据集,再使用一维残差卷积神经网络进行故障诊断。得出基于Adam优化算法的多通道一维残差卷积神经网络诊断准确率最高。因此,多通道一维残差卷积神经网络在风力发电机组轴承故障诊断中应用效果良好,能够准确的识别各类故障模式,为机组的安全、稳定运行提供了保障。
To improve the accuracy of convolution neural network on fault diagnosis of wind turbine bearing,the bearing fault data of a 2 MW wind turbine generator unit are analyzed and compared with single channel and multiple channels,multiple diagnosis network models and different optimization algorithm.It is proposed to integrate multiple vibration sensor data for multi-channel one-dimension data set,and then one-dimension residual convolution neural network is used for fault diagnosis.It is concluded that the multi-channel one-dimension residual convolution neural network based on Adam optimization algorithm has the highest diagnostic accuracy.Therefore,multi-channel one-dimension residual convolution neural network has good application effect on failure diagnosis of wind turbine bearing,which can accurately identify various fault modes and provide guarantee for safe and stable operation of the wind turbine.
作者
郑梁
刘桂然
朱孝晗
ZHENG Liang;LIU Guiran;ZHU Xiaohan(Guodian United Power Technology Co.,Ltd.,Beijing 100039,China;State Key Laboratory of Wind Power Equipment and Control,Baoding 071000,China)
出处
《机械》
2023年第3期1-7,共7页
Machinery
基金
国家重点研发计划(2019YFB2005005-02)。
关键词
风力发电机组
智能故障诊断
多通道数据
一维残差卷积神经网络
wind turbine
intelligent fault diagnosis
multichannel data
one-dimension residual convolution neural network
