摘要
针对核电厂应急柴油机的指针式仪表无法实现数字化记录、趋势分析的问题,该文研究采用机器视觉技术和卷积神经网络方法对指针式仪表进行采集和数字化识别。该技术通过摄像头实时采集指针式仪表盘图像,利用卷积神经网络图像识别模型进行仪表值数字化,并解决了识别模型适用范围窄、预处理过程繁琐、难以识别刻度非均匀的仪表等问题。研发的识别算法和样机在核电厂应急柴油机再鉴定试验期间进行试用,取得了很好识别准确率和识别速率,可以完全替代人工读数的方法,为核电厂应急柴油机的运行状态监测和巡检工作提供了高效、便捷、可靠的手段。
Aiming at the problem that the digital recording and trend analysis can not be realized by the pointer in- strument of emergency diesel engine in nuclear power plant,this paper studies the acquisition and digital recognition of the pointer instrument by using machine vision technology and convolution neural network method.The technology collects real-time image of pointer dashboard by camera,digitizes the instrument value by using convolution neural network image recognition model,and solves the problems of narrow application range of recognition model,cumbersome pretreatment process and difficulty in identifying instruments with non-uniform scale.The recognition algorithm and prototype developed were tested during the re-qualification test of emergency diesel engine in nuclear power plant,and good recognition accuracy and recognition rate were obtained.It can completely replace the manual reading method.It provides an efficient,convenient and reliable means for the operation condition monitoring and patrol inspection of emergency diesel engine in nuclear power plant.
作者
周勇
陈星
朱鹏树
梁永飞
ZHOU Yong;CHEN Xing;ZHU Peng-shu;LIANG Yong-fei(Daya Bay Nuclear Power Operations and Management Co.,Ltd.,Shenzhen 518124,China)
出处
《自动化与仪表》
2018年第12期78-82,共5页
Automation & Instrumentation
关键词
指针式仪表
图像识别
卷积神经网络
遮挡检测
dail instrument
image identification
convolutional neural networks
occlusion detection
