管道漏磁检测缺陷识别技术  

Defect recognition technology of magnetic flux leakage detection for pipeline

作　　者：杨理践[1] 余文来[1] 高松巍[1] 孔丽新[1] 

机构地区：[1]沈阳工业大学信息科学与工程学院,沈阳110870

出　　处：《沈阳工业大学学报》2010年第1期65-69,共5页Journal of Shenyang University of Technology

基　　金：国家自然科学基金仪表专项基金资助项目(60927004)

摘　　要：针对神经网络法对管道缺陷进行识别存在所需样本数据量大、容易陷入局部极值、泛化能力没有保证等问题,提出一种新的机器学习方法,即支持向量机法.介绍了支持向量机的分析原理,以碗状缺陷为例,采用ANSYS有限元分析软件对40组不同尺寸的缺陷进行仿真,将得到的40组漏磁数据和10组实测数据作为学习样本,另外10组新数据作为验证数据,以MATLAB软件为平台进行了识别实验.实验结果表明,识别误差均在5%以下,且泛化能力强,识别方法简单,在有限样本情况下比神经网络法更具优势.With performing the pipeline defect recognition using the previous neural network method,a great amount of samples are required,the local extremum is easily attained and the generalizable ability can not be ensured. Thus,a novel machine learning method,support vector machine （SVM） method,was proposed. The analysis principle of the SVM was introduced. With taking the bowl shape defect as the example,the simulation for 40 groups of defects with different dimension was conducted using ANSYS finite element analysis software. With taking 40 groups of magnetic flux leakage data and 10 groups of measured data as the learning samples as well as 10 groups of additional data as the verifying data,the recognition experiment was carried out based on MATLAB software. The experimental results show that the recognition error is below 5%. The present recognition method is simple and more advantagous than the neural network method especiallly for finite samples,and its generalizable ability is strong.

关 键 词：管道 漏磁检测 缺陷识别 支持向量机 泛化能力 漏磁场 有限元分析 样本数据 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]