摘要
针对风力发电企业在线风力发电机叶片表面损伤自动诊断难的实际问题,提出利用无人机机器视觉的基于L-AlexNet深度学习框架的风力机叶片表面损伤诊断方法。为验证该方法的有效性,选用经无人机采集的8270张像素为227×227的风力机叶片图像分别对传统BP神经网络、深度卷积网络AlexNet和L-AlexNet等分类器进行训练,再采用10次、每次350张图像进行测试。诊断类别包括:背景类、无损伤或伪损伤类、存在修复类、砂眼类、裂纹类和混合损伤类。测试结果表明:L-AlexNet深度卷积网络对表面损伤诊断的平均准确率达97.0286%,较AlexNet的平均准确率高1.9144%,较传统BP神经网络的平均准确率高26.9622%。所提出的基于优化深度学习框架的自动诊断方法可有效实现对风力机叶片表面损伤的准确诊断。
To solve the practical problems of wind turbine blade damage diagnosis,we proposed a L-AlexNet method,a type of deep learning algorithm,combined with a machine vision technology.The 8270 wind power generator blades images with a size of 227×227 pixel were captured by a UAV(Unmanned Aerial Vehicle) camera and taken as a training data set.The BP(Back Propagation) neural network,the deep(CNN) Convolutional Neural Network AlexNet,and another deep CNN L-AlexNet classifier were trained accordingly,and the newly added 10 turn of 350 images were used for classification tests.Diagnostic categories include:background,no damage or pseudo damage,repaired,sand holes,cracks,mixed damages.The test results show that the average accuracy rate of LAlexNet classifier is 97.0286%,which is 1.9144% higher than that of the AlexNet classifier,and 26.9622% higher than that of traditional BP network classifier.Therefore,the proposed method,based on the deep learning framework,is effective for the automatic damage diagnosis of wind power generator blade surfaces.
作者
赵肖懿
董朝轶
周鹏
朱美佳
任靖雯
陈晓艳
Zhao Xiaoyi;Dong Chaoyi;Zhou Peng;Zhu Meijia;Ren Jingwen;Chen Xiaoyan(College of Electric Power,Inner Mongolia University of Technology,Hohhot 010080,China;Inner Mongolia Key Laboratory,of Mecchanical and Electrical Control,Hohhot 010051,China)
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2021年第7期390-397,共8页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(61863029)
内蒙古科技成果转化项目(CGZH2018129)
内蒙古自治区科技计划项目(关键技术攻关计划项目)
内蒙古自然科学基金(2020MS06020)。
