文章摘要
冯雪松,文玉梅,甄锦鹏,张雪园,李平,文静.管道泄漏声振动信号的特征分析[J].声学技术,2015,34(5):413~418
管道泄漏声振动信号的特征分析
Feature analysis of pipeline leakage acoustic signals for leak identification
投稿时间:2014-10-15  修订日期:2015-01-21
DOI:10.16300/j.cnki.1000-3630.2015.05.007
中文关键词: 管道泄漏识别  特征提取  模式识别  频域分布  统计特征
英文关键词: pipeline leak identification  feature extraction  pattern recognition  frequency distribution  statistical property
基金项目:国家自然科学基金资助项目(61174017)
作者单位E-mail
冯雪松 重庆大学光电工程学院传感器与仪器研究中心, 重庆 400044  
文玉梅 重庆大学光电工程学院传感器与仪器研究中心, 重庆 400044 ymwen@cqu.edu.cn 
甄锦鹏 重庆大学光电工程学院传感器与仪器研究中心, 重庆 400044  
张雪园 重庆大学光电工程学院传感器与仪器研究中心, 重庆 400044  
李平 重庆大学光电工程学院传感器与仪器研究中心, 重庆 400044  
文静 重庆大学光电工程学院传感器与仪器研究中心, 重庆 400044  
摘要点击次数: 1173
全文下载次数: 3692
中文摘要:
      管道泄漏声振动是泄漏过程中多种事件共同作用产生的,所以使用多种特征才可能比较准确地描述管道泄漏声信号。合理的选取和使用这些特征对于泄漏信号识别至关重要。通过分析泄漏过程,确定选取信号的随机性和频率分布特性作为泄漏特征。由于随机性和频域特性可由多种参数描述,于是比较了各种参数作为泄漏特征值的辨识效果。使用支持向量机作为分类器,对比了使用单种特征以及组合使用多种相同或不同类特征时,实际供水管道声振动及管道泄漏的识别效果。使用两种特征的识别准确率普遍高于使用单种特征的情况,然而使用更多的特征却并没有进一步提高准确率。其中样本熵和功率谱分布特征的组合准确率最高,达到了93%,而且使用此特征组合能够正确区别管道周围常见噪声。
英文摘要:
      Leakage acoustic signal of pipelines is originated from the concurrent events during leaking. This physical fact suggests that only by combining multiple features of the signal can a leak be uniquely identified. Reasonable selection and appropriate application of features is the key to develop a valid leak recognizing pattern. According to the mechanism of leaking, the characteristics of randomness and frequency distribution are chosen as leak features. Since the randomness and frequency distribution can be described with various characteristics, a single characteristic and the combination of multiple characteristics from the same or different classes are compared for identifying leaks based on abundant acoustic signal samples collected from practical water-supplied pipelines .The Support Vector Machine is used for recognition. The recognition effect with two characteristics is better than that with a single characteristic, particularly the combination of sample entropy and power spectral distribution obtains the highest correct rate of 93%. However, more characteristics fail to produce further improvement in the correct rate of recognition. With the selected features, common noise and mimicked leakage sound can also be identified correctly.
查看全文   查看/发表评论  下载PDF阅读器
关闭