基于深度学习的电力设备红外图像故障诊断方法

Fault Diagnosis Method for Power Equipment Infrared I mages Based on Deep Learning

红外热成像检测技术在变电站中的广泛应用产生了大量红外图像。针对变电站人工故障诊断作业量大、效率低的问题,提出了一种基于MobileNet的设备红外图像自动故障诊断方法,并进行了软件封装。首先,考虑到检测速率需求,构建了基于MobileNet轻量化网络的深度学习电力设备分类模型,并通过迁移学习提高模型训练效率和准确率;其次,利用比色条和温度极值拟合出图像灰度与实际温度的函数关系,以此计算出红外图像中设备热点的温度;最后,对以上结果和故障诊断规范进行软件封装,实现了故障的自动诊断。实验结果表明:该软件设备分类准确率达到95.7%,计算所得热点温度与实际温度的误差均值为-0.29%,所提出方法和软件有效提高了电力设备红外图像故障诊断效率,为变电站智能巡检提供了新思路。

With the wide application of infrared thermal imaging technique in power substations,a large number of infrared images are produced.Aiming at the problems of a large amount of work and low efficiency of manual fault diagnosis,this paper proposed an automatic fault diagnosis method based on MobileNet for equipment using infrared images and carries out software package.Firstly,considering detection speed demand,the paper constructs a classification model of power equipment based on MobileNet using deep learning,and uses transfer learning to improve training efficiency and accuracy of the model.Secondly,it uses the colormap and extreme value of temperature to fit the function relationship between the gray-scale and actual temperature,which helps calculating the actual temperature of hot spots on the equipment.Finally,all the above and fault criteria are compiled in a software package so as to realize automatic fault diagnosis.The experimental results show the classification accuracy reaches 95.7%and the mean error between calculated temperature and actual temperature is-0.29%,which proves the proposed method and software can greatly improve the efficiency in infrared image fault diagnosis and provide a new idea for intelligent inspection in power substations.