基于机器学习的管道金属损失缺陷识别方法

Recognition Method of Pipeline Metal Loss Defects Based on Machine Learning

针对石化管道漏磁检测时缺陷较难识别的问题,首先对管道金属损失缺陷类型进行分类量化,建立石化管道金属损失缺陷漏磁检测三维有限元模型,采用Maxwell对1000组缺陷进行了漏磁仿真,得到了漏磁信号数据;然后分析了漏磁信号与缺陷类型及尺寸之间的关系,提取了漏磁检测信号的4个特征值,并验证了特征值对于识别缺陷类型的有效性;最后采用支持向量机、随机森林以及梯度提升决策树(GBDT)3种机器学习算法对缺陷信号特征量进行了分类识别。研究结果表明,3种算法对于缺陷的分类识别效果均较好,特别是GBDT算法在现有的数据范围内达到了100%的识别率。研究结果对石化管道完整性评价具有一定的指导意义。

In view of the difficulty of identifying defects in petrochemical pipeline magnetic flux leakage(MFL)detection,the types of metal loss defects in pipelines are classified and quantified,and the three-dimensional finite element model of MFL detection of petrochemical pipeline metal loss defects is established.Then,the relationship between MFL signal and defect type and size is analyzed,and four eigenvalues of MFL signal are extracted,and the effectiveness of the eigenvalues for identifying defect types is verified.Finally,three machine learning algorithms,support vector machine,random forest and gradient boosting decision trees(GBDT),are used to classify and identify the feature quantity of defect signal.The results show that the three algorithms are effective for defect classification and recognition,especially GBDT algorithm achieves 100%recognition rate in the existing data range.The study has a certain guiding significance for the integrity evaluation of petrochemical pipeline.