模型可再生的管道缺陷故障诊断理论与技术研究

项目名称： 模型可再生的管道缺陷故障诊断理论与技术研究

项目编号： No.61473069

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 刘金海

作者单位： 东北大学

项目金额： 81万元

中文摘要： 漏磁内检测方法是保障长输管道安全运行的主要技术,其中缺陷故障诊断是该技术的核心。但已有的故障诊断方法难以解决大量检测数据及未知缺陷（统称大数据）且故障样本有限条件下的管道缺陷故障诊断问题。本申请拟以解决大数据和故障样本有限条件下的管壁缺陷故障诊断问题为目标开展研究。内容包括：研究建立多通道数据动态校正及滤波方法，为后续研究提供高质量数据；建立基于变尺度数据窗的缺陷快速检测方法，完成大数据下快速的缺陷检测；首次提出模型可再生缺陷故障诊断方法，巧妙地通过模型再生机制实现现有缺陷故障诊断方法的平滑切换，克服现有方法的缺点同时保留其优点；以模型可再生故障诊断方法为核心设计焊缝缺陷检测与故障诊断方法，解决焊缝中缺陷的故障诊断这一国际难题。最后建立大数据和故障样本有限条件下模型可再生故障诊断理论框架。本研究成果能有效处理大数据和故障样本有限时的管壁缺陷故障诊断问题，具有重要的理论意义与实际应用价值。

中文关键词： 数据驱动；故障诊断；模型再生；故障分类；异常检测

英文摘要： MFL based inspection technology is one of the main way for the safty running of pipeline, in which flaws diagnosis is the key technology. The existed methods of flaws diagnosis can not deal with flaws well when the inpection data and flaws are huge(named as big data) but flaw samples are small. For solving this problem, we plan research several contents, as follows: Build methods of dynamic adjust and data filter for muti-channel data which can provide the valid data source for following research; Build fast method for large flaws detection based on alterable data windows, which can detect flaws quickly in big data; Build flaw diagnosis method based on autogeny models(FDRM), which can overcome the backdraws of existed two main flaw diagnosis and reserve their advantage by model autogenyion. A method of flaw diagnosis for weld is proposed based on FDRM, which is the fisrt method to deal with the flaw diagnosis for weld in the world. At last, base on the achievements, a new frame of pipeline flaws diagnosis is built, which can find and diagnosis flaws in the situation of big data and the limited flaw samples. The proposed frame have not only theoretical significance but also practical utility.

英文关键词： data-driven;falut diagnosis;model autogeny;falut classification;anomaly detection