东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

模糊双曲正切模型研究综述

ISSN号：0254-4156
期刊名称：自动化学报
时间：2013
页码：1849-1857
分类：TE88[石油与天然气工程—油气储运工程] TP277[自动化与计算机技术—控制科学与工程;自动化与计算机技术—检测技术与自动化装置]
作者机构：[1]东北大学信息科学与工程学院,沈阳110004, [2]中海油能源发展装备技术有限公司,天津300451
相关基金：国家自然科学基金（61473069,61374124）; 中央高校基本科研业务费专项资金（N140405002,N130104001）项目资助
相关项目：基于数据驱动的复杂控制过程故障诊断和监控理论与关键技术研究

关键词：特征提取, 目标识别, 随机森林, 漏磁信号, feature extraction, object recognition, random forest, magnetic flux leakage（MFL） signal

中文摘要：

基于漏磁原理的管道内检测信号和管壁缺陷之间存在强非线性关系,导致缺陷（特别是小缺陷）识别和分析困难。针对内检测器漏磁信号中的缺陷识别问题,设计一种新的缺陷识别方法,可以在漏磁内检测数据中精确识别缺陷。该方法包括3个部分：针对普通基值校准算法在多通道数据对齐中精度低的问题,提出一种基值二次校准算法,为后续数据分析提供精确数据源;区别于常规方法中用到的峰谷值等显性特征,提出6种更能反映缺陷信号的本质特征,并设计了相应的特征提取方法,为缺陷识别提供精确的分类依据;设计了基于随机森林的缺陷识别算法,可以在漏磁内检测数据中精确识别各种缺陷。最后,对所提出的方法进行了性能对比分析和试验验证。结果表明：所设计的缺陷识别方法准确率为99.59%,其中对小缺陷的识别灵敏度为98.66%,证明所提出的缺陷识别方法可有效地完成目标缺陷的识别。

英文摘要：

There exists strong nonlinear relationship between the pipeline inner detection signal based on magnetic flux leakage（ MFL）principle and wall defects,which leads to the difficulty in the recognition and analysis of defects,especially small defects. Aiming at the defect recognition problem in MFL inner detection signal,a new defect recognition method is proposed,which can accurately recognize the defect from MFL inner detection data. This method includes three parts： Aiming at the low accuracy problem of common base calibration algorithm in multi-channel data alignment,a base re-calibration algorithm is proposed to provide accurate data source for subsequent analysis; different from the dominant feature,such as peak-valley value in conventional method,six features that can better reflect the essential characteristics of the defect signal are proposed and corresponding feature extraction algorithm is proposed to provide accurate classification basis for the defect recognition; the defect recognition algorithm based on random forest is designed,which can accurately recognize various kinds of defects in MFL inner detection data. Finally,performance comparison analysis and experiment verification for the proposed method was carried out. The experiment results show that with the designed defect recognition algorithm,the recognition accuracy can reach 99. 59%,and the recognition sensitivity for small defects reaches 98. 66%,which proves that the proposed defect recognition method can achieve the object defect recognition effectively.