基于堆叠自编码器神经网络的复合电磁检测铁磁性双层套管腐蚀缺陷分类识别方法

Classification of Subsurface Corrosion in Double-casing Pipes via Integrated Electromagnetic Testing with Stacked Auto-encoder Artificial Neural Network

铁磁性双层套管长期服役于恶劣的工作环境,极易出现腐蚀缺陷,定期为服役中的双层套管进行在线检测十分必要,而对管壁腐蚀缺陷位置的分类识别是管道定量检测与维修的前提和基础,实时准确的套管腐蚀缺陷分类识别能力是决定管道在线检测效率的重要因素。针对这一情况,将脉冲远场涡流和脉冲涡流技术相结合,提出了基于堆叠自编码器神经网络的分类方法。通过仿真和实验选取合适特征量作为输入层,实现了内管外壁腐蚀、外管内壁腐蚀和外管外壁腐蚀的分类,实验整体预判精度可达97.5%,结果表明该方法可对双层套管腐蚀缺陷缺陷实施高效、高精度分类识别。

The in-service Ferromagnetic Double-casing Pipe （FDP） is prone to Subsurface Corrosion （SSC） in the rigorous environments. It is necessary to evaluate SSC periodically. On the premise of defect classification in quantitative evaluation and maintenance, the real-time classification of SSC is of great importance. In light of this, this paper proposes a stacked Auto-Eneoder Artificial Neural Network （SAE-ANN） classification method for classification of SSC in FDP in conjunction with Pulsed Remote Field Eddy Current （PRFEC） and Pulsed Eddy Current （PEC）. By choosing appropriate eigenvalue as the input layer, 3 SSC scenarios （corrosion on external surfaces of inner and outer casing pipes; corrosion on the internal surface of the outer casing pipe） can be identified. The accuracy can reach 97.5% and the result shows that the proposed method is capable of identifying the localized SCC without much loss in accuracy.