Explanatable deep learning framework for pipeline leakage monitoring

The invention discloses an interpretable deep learning framework for pipeline leakage monitoring, which comprises the following steps: firstly, analyzing a leakage acoustic emission signal by adopting three different wavelet basis functions, and processing to obtain time-frequency characteristics wh...

Full description

Saved in:
Bibliographic Details
Main Authors DU SHUHONG, ZHANG ZHIFEN, XING JI, CHENG WEI, CHEN XUEFENG, MIAO ZHUANG, HUANG JING, WEN GUANGRUI
Format Patent
LanguageChinese
English
Published 02.07.2024
Subjects
BLASTING
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
ELECTRIC DIGITAL DATA PROCESSING
HEATING
LIGHTING
MECHANICAL ENGINEERING
PHYSICS
PIPE-LINE SYSTEMS
PIPE-LINES
STORING OF DISTRIBUTING GASES OR LIQUIDS
WEAPONS
Online AccessGet full text

Cover

More Information
Summary:The invention discloses an interpretable deep learning framework for pipeline leakage monitoring, which comprises the following steps: firstly, analyzing a leakage acoustic emission signal by adopting three different wavelet basis functions, and processing to obtain time-frequency characteristics which show correlation and complementarity; on the basis, an interpretable neural network with different wavelet convolution kernels is designed, and abstract feature details of the broadband acoustic emission signals are extracted through a multi-stage dynamic receptive field. Then, a feature fusion module based on channel importance weighting is designed to highlight contribution degrees of different channels to optimize a learning process of the network. Results show that the method provided by the invention can effectively extract the distinguishing features of the leakage acoustic emission signals, and the recognition accuracy of different leakage states can reach 98.32%, which is significantly higher than that
Bibliography:Application Number: CN202410422892