Page 62 - 《应用声学）》2023年第5期

第 42 卷 第 5 期 Vol. 42, No. 5
2023 年 9 月 Journal of Applied Acoustics September, 2023

⋄ 研究报告 ⋄

基于声发射和GAN-CNN的铝合金管道法兰

连接松动泄漏检测 ∗

王 新 1† 夏广远 2

(1 内蒙古农业大学职业技术学院 包头 014109)
(2 内蒙古科技大学机械工程学院 包头 014010)
摘要：面向管道法兰连接松动引起的泄漏检测需求，为解决数据样本不足和减少特征指标手动选取的繁琐环
节，提出了一种基于声发射和生成对抗网络 -卷积神经网络 (GAN-CNN) 的铝合金管道法兰连接松动泄漏检
测方法。首先，搭建管道泄漏标定和数据采集实验台，利用声发射技术获取不同等级的原始泄漏信号。其次，
采用生成对抗网络生成样本数据扩充原始数据，并利用统计特征评估样本生成质量。最后，将生成的样本数据
与原始数据设置为不同训练集，基于卷积神经网络构建智能分类检测模型，应用于管道泄漏检测。同时，设置
了新的泄漏工况，以及比较了小样本智能分类支持向量机模型的分类结果，以验证所提出方法的适用性和有
效性。结果表明，基于生成对抗网络和卷积神经网络的智能分类模型能够有效识别管道法兰连接松动泄漏，验
证了所提出方法的有效性。
关键词：法兰；泄漏；生成对抗网络；卷积神经网络
中图法分类号: TE973.6 文献标识码: A 文章编号: 1000-310X(2023)05-0954-09
DOI: 10.11684/j.issn.1000-310X.2023.05.008

Leak detection of aluminum alloy pipe due to loosening of flange connection

based on acoustic emission and GAN-CNN

WANG Xin 1 XIA Guangyuan 2

(1 Vocational and Technical College of Inner Mongolia Agricultural University, Baotou 014109, China)
(2 School of Mechanical Engineering, Inner Mongolia University of Science & Technology, Baotou 014010, China)

Abstract: For the detection requirements of pipe leak caused by loosening of flange connection, in order to
solve the insufficient data samples and reduce tedious process of manual selection of characteristic indexes, in
this paper, a method for leak detection of aluminum alloy pipe due to loosening of flange connection based on
acoustic emission and generative adversarial networks-convolutional neural network (GAN-CNN) is proposed.
Firstly, the experimental platform for pipeline leak calibration and acquisition are set up, and using the
acoustic emission technology, raw leak signals in different levels are obtained. Secondly, GAN is used to
generate sample data to expand the raw data. And, statistical characteristics are used to evaluate the quality
of sample generation. Finally, the generated sample data and raw data are set as different training sets, and
an intelligent classification detection model is constructed based on CNN, which is applied to pipeline leak
detection. Meanwhile, a new leak condition is set and the classification results of the support vector machine
(SVM) models are compared to verify the applicability and effectiveness of the proposed method. The results

2022-05-19 收稿; 2022-06-29 定稿
内蒙古自治区高等学校科学技术研究项目 (NJZY22530)
∗
作者简介: 王新 (1979– ), 女, 内蒙古呼和浩特人, 硕士, 副教授, 研究方向: 机械设计制造及其自动化, 自动化无损检测。
† 通信作者 E-mail: 2933717175@qq.com