期刊
PROCESS SAFETY AND ENVIRONMENTAL PROTECTION
卷 124, 期 -, 页码 279-289出版社
ELSEVIER
DOI: 10.1016/j.psep.2018.11.009
关键词
Small leakage detection; Sound pressure spectrum; Characteristic frequency band; MFCCs; SVM classifier
资金
- National Natural Science Foundation of China [61473205, 51604192, 61773283]
Direct acoustic inspection with an inner detector has recently been suggested as an efficient leakage detection method. Because of its high sensitivity, this method is very promising for pipeline health monitoring. However, few studies have used automatic recognition based on the acoustic signature of leakage signals for direct acoustic inspection. In this study, a hybrid computational fluid dynamics and computational aeroacoustics mathematical model of small leakage based on Lighthill's analogy was constructed to simulate the acoustic signature in the vicinity of a leakage source. The obtained results show that the leakage sound is a broadband signal with dispersion characteristics. Furthermore, because of the appearance of a stationary wave, amplitude peaks emerge at specific frequencies in the sound pressure spectrum. According to the simulation parameters, a specific experimental environment was constructed that matched the simulation results. Inspired by the formant recognition method used in automatic speech recognition, a novel acoustic signature extraction approach was proposed based on Mel-frequency cepstral coefficients. Moreover, instead of a full band, a redesigned frequency band was applied in response to the influence of noise interference and the sound pressure spectrum. An efficient support vector machine classifier was employed to solve the binary classification problem (regardless of leakage), and the obtained results are very encouraging: an almost 10% performance improvement was achieved with a redesigned frequency band, and both the accuracy and specificity of the recognition system reached up to 97%. (C) 2018 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据