基于LSTM的储罐底板缺陷识别  

Identification of Tank Bottom Plate Defects Based on LSTM

作　　者：潘勇 

机构地区：[1]河北省特种设备监督检验研究院邯郸分院,河北邯郸056011

出　　处：《工业控制计算机》2024年第4期29-31,34,共4页Industrial Control Computer

摘　　要：储罐底板因腐蚀穿孔或疲劳裂纹引起的泄漏是常压储罐的主要失效形式。储罐底板的检测通常采用电磁技术,利用永磁体磁化钢板产生的漏磁效应进行缺陷检测,但这种检测方法无法区分钢板中的裂纹缺陷与腐蚀缺陷,给后期处理带来不便。以包含裂纹缺陷及腐蚀缺陷的磁检测信号作为数据源,建立基于LSTM(长短期记忆)的一维时序神经网络模型,实现储罐底板缺陷分类识别。结果表明:基于LSTM的神经网络模型,可以快速实现储罐底板缺陷的分类识别,对钢板裂纹缺陷的识别准确率为96%,对钢板腐蚀缺陷的识别准确率为92%。Leakage caused by corrosion perforation or fatigue cracks on the bottom plate of storage tanks is the main failure form of atmospheric storage tanks.The detection of tank bottom plates usually adopts electromagnetic technology,which utilizes the leakage magnetic effect generated by permanent magnet magnetization of steel plates for defect detection.However,this detection method cannot distinguish between crack defects and corrosion defects in steel plates,which brings inconvenience to later processing.In this paper,the magnetic testing signals including crack defects and corrosion defects are used as data sources,and a one-dimensional time series neural network model based on LSTM(Long short-term memory)is established to realize the classification and recognition of tank floor defects.The results show that the neural network model based on LSTM can quickly classify and identify tank bottom plate defects,with an accuracy of 96%for identifying steel plate crack defects and 92%for identifying steel plate corrosion defects.

关 键 词：储罐底板 漏磁检测 缺陷识别 LSTM 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TE972[自动化与计算机技术—控制科学与工程]