基于GAN网络的双层管柱外层套管形变位置检测算法  

Deformation Position Detection Algorithm of Outer Casing in Double-Layer Pipe Base on GAN Network

作　　者：陈友荣[1] 金合丽 黄华 徐菲 任条娟[1] 王柯 CHEN Yourong;JIN Heli;HUANG Hua;XU Fei;REN Tiaojuan;WANG Ke(College of Information Science and Technology,Zhejiang Shuren University,Hangzhou Zhejiang 310015,China;School of Computer Science and Artificial Intelligence,Changzhou University,Changzhou Jiangsu 213164,China;Petroleum Engineering Institute,Zhongyuan Oilfield Company,SINOPEC,Puyang He’nan 457001,China)

机构地区：[1]浙江树人学院信息科技学院,浙江杭州310015 [2]常州大学计算机与人工智能学院,江苏常州213164 [3]中国石化中原油田分公司石油工程技术研究院,河南濮阳457001

出　　处：《传感技术学报》2023年第2期242-250,共9页Chinese Journal of Sensors and Actuators

基　　金：浙江省自然科学基金项目(LQ19F010012);国家科技重大专项项目(2016ZX05017-003)。

摘　　要：针对普光气田所使用的双层管柱外层套管存在形变位置难以检测问题,利用电磁探伤仪的电涡流检测信号,提出一种基于GAN网络的双层管柱外层套管形变位置检测算法(DPDA)。在DPDA中,首先对电涡流检测信号进行预处理将其转换为涡流图像,构建包括生成子网络和判别子网络的形变位置检测模型。然后提出权衡重构损失、隐性损失和对抗损失的两个子网络目标函数,并采用爬行动物搜索算法对目标函数的权重系数进行寻优。训练获得形变位置检测模型,并根据形变评分对形变像素进行初步检测。最后根据涡流图像特性,设计形变结果修正方法进行修正,获得形变位置检测结果。实验结果显示:DPDA能准确检测双层管柱外层套管的形变位置,提高检测准确率,降低漏检率和误检率,比自编码器(AE)、金字塔结构网络模型(PSNM)和猫群优化算法(CSOA)更优。It is difficult to detect the deformation position of outer casing in double-layer pipe in Puguang Gas field.To solve the problem,eddy current detection signal of electromagnetic flaw detector is used and a deformation position detection algorithm(DPDA)of outer casing in double-layer pipe based on GAN network is proposed.In DPDA,the eddy current detection signals are preprocessed and converted into eddy current images,and a deformation position detection model is constructed,including the generation sub-network and the discrimination sub-network.Then the loss functions of these two sub-networks with reconstruction loss,hidden loss and confrontation loss are proposed and the weight coefficients of the loss functions are optimized by using reptile search algorithm.The deformation position detection model is obtained by training,and the deformation pixels are preliminarily detected according to the deformation score.Finally,according to the characteristics of eddy current images,a correction method is designed to correct the abnormal results and obtain the deformation position detection results.The experimental results show that DPDA can accurately detect the deformation position of the outer casing in double-layer string.It improves the detection accuracy and reduces the missed detection rate and error detection rate.It is better than auto encoder(AE),Pyramid structure network model(PSNM)and cat swarm optimization algorithm(CSOA).

关 键 词：电涡流检测信号 形变位置检测 GAN网络 检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]