Percussion-Based Pipeline Ponding Detection Using a Convolutional Neural Network

Pipeline transportation is the main method for long-distance gas transportation; however, ponding in pipeline can affect efficiency and even cause corrosion to some cases. A non-destructive detect using percussion acoustic signals a convolution neural network (CNN) proposed this paper. During process of detection, constant energy spring impact hammer used apply an on pipeline, percussive are collected. Mel spectrogram extract feature signal with different volumes pipeline. The transferred input layer CNN convolutional kernel matrix realizes recognition volume. results show that identify amount signals, which use as feature. Compared support vector machine (SVM) model decision tree model, has better performance. Therefore, percussion-based detection paper high application potential.