Abstract:  Aiming at the problems that the magnetic memory detection of oil and gas pipelines was greatly affected by the direction and the accurate identification of defect edges was difficult, a magnetic gradient tensor combination invariant algorithm was proposed for accurate identification of pipeline defect edges based on MGD optimization. Taking L245N pipeline steel as the test material, the circular hole defects with different depths and diameters were prefabricated, and the magnetic gradient tensor measurement system was designed. Combined with TSC-5M-32 magnetic memory instrument, the magnetic gradient tensor matrix of pipeline was obtained. In order to overcome the influences of the detection direction on the magnetic memory signals, the second invariant I1 and the third invariant I2 of the magnetic gradient tensor were extracted respectively. Further considering that these two invariants were easy to present ambiguity at the edges of the defects, the two invariants were improved according to the Cardano formula, and the weights a and b were set separately for superposition to obtain the combined invariant I. The fractional derivative improved gradient descent algorithm was used to determine the optimal weight, and the magnetic memory representation model of the pipeline defect edges was established. The verification results show that the average relative error of the model for defect edge recognition is as 3.59%, and the maximum relative error is as 6%, which provides a feasible method for accurate identification of pipeline defect edges in practical engineering.

Key words: defect edge recognition, magnetic gradient tensor invariant, modified gradient descent（MGD）, metal magnetic memory

摘要： 针对油气管道磁记忆检测受方向影响较大且缺陷边缘精确识别困难的问题，提出了一种基于改进梯度下降算法（MGD）优化的磁梯度张量组合不变量算法模型，用于管道缺陷边缘的精确识别。以L245N管线钢为试验材料，预制不同深度、不同直径的圆孔状缺陷，设计磁梯度张量测量系统，结合TSC-5M-32型磁记忆仪进行检测实验，获得管道的磁梯度张量矩阵。为克服检测方向对磁记忆信号的影响，分别提取磁梯度张量第二、第三不变量I1、I2，进一步考虑这两种不变量在缺陷边缘处易出现模糊，根据Cardano公式对两种不变量进行改进，并分别设置权值a、b进行叠加获得组合不变量I，利用分数阶求导改进梯度下降算法确定最优权值，建立管道缺陷边缘磁记忆识别模型。研究结果表明：该模型对缺陷边缘识别平均相对误差为3.59%，最大相对误差为6%，为实际工程中管道缺陷边缘精准识别提供了可行办法。

关键词: 缺陷边缘识别, 磁梯度张量不变量, 改进梯度下降法, 金属磁记忆