基于MSCNN－LSTM的注意力机制U型管道缺陷识别模型

PDF(2264 KB)

振动与冲击 ›› 2023, Vol. 42 ›› Issue (22) : 293-302.

论文

朱雪峰1，冯早2，马军2，范玉刚2

作者信息 +

Siphon defect recognition model based on the MSCNN-LSTM and attention mechanism

ZHU Xuefeng1， FENG Zao2， MA Jun2， FAN Yugang2

Author information +

文章历史 +

摘要

对于承担缓震功能的特异U型管道，其结构复杂使得管内和管壁缺陷具有时延性和多源多征兆等特点。针对U型管道缺陷难以有效识别的问题，提出一种基于MSCNN－LSTM的注意力机制U型管道缺陷识别方法。采用主动声学检测方法获取管道声学响应信号，将原始声学信号作为模型输入，训练多尺度卷积神经网络（Multi-Scale Convolution Neural Network，MSCNN）提取重要细粒度局部特征。然后，多尺度局部特征融合为一个特征向量输入至长短期记忆网络（Long Short-term Memory Neural Networks，LSTM)中抽取潜藏在时序规律的粗粒度上下文特征。下一步引入注意力机制，对提取的特征赋予不同的权重，使模型更关注于最具类别区分度的特征，滤除冗余特征，提高模型缺陷识别能力。最后，在输出端通过Softmax分类器实现U型管道缺陷识别。实验结果表明：与其它常用的分类方法相比，该方法拥有更快的收敛速度，实现98.44%的缺陷识别准确率。此外，采用Grad-CAM类激活可视化方法对所提模型的特征学习和缺陷分类机理实现了过程分析和展示。

Abstract

In order to avoid physical damage to buried pipelines caused by resonance, most large buildings above ground are designed with sinking siphon. To solve the multi-source and multi-symptom defect recognition problem caused by siphon’s complex structure, the attention mechanism defect recognition method based on Multi-Scale Convolution Neural Network (MSCNN) and Long Short-term Memory Neural Networks (LSTM) was proposed. The pipeline response acoustic signal was acquired through active acoustic detection method. With the original signal as model input, MSCNN was trained to extract important fine-grained local features. Then, multi-scale local features were fused into a feature vector and input into LSTM to extract coarse-grained contextual features underlying time sequence regularity. By introducing the attention mechanism and assigning different weights to the extracted features, the model paid more attention to the features with the highest category discrimination, and the redundant features were filtered out, thereby improving the defect recognition ability. Finally, siphon’s defect recognition was achieved through Softmax classifier at the output. According to the experimental results, the proposed method has faster convergence speed and high defect recognition accuracy rate 98.44%. Moreover, Grad-CAM class activation visualization method was adopted for the process analysis and demonstration of feature learning and defect classification mechanism of the proposed model.

导出引用

朱雪峰1，冯早2，马军2，范玉刚2. 基于MSCNN－LSTM的注意力机制U型管道缺陷识别模型[J]. 振动与冲击, 2023, 42(22): 293-302

ZHU Xuefeng1， FENG Zao2， MA Jun2， FAN Yugang2. Siphon defect recognition model based on the MSCNN-LSTM and attention mechanism[J]. Journal of Vibration and Shock, 2023, 42(22): 293-302

参考文献

[1] YU Yicheng, SAFARI Ali, NIU Xudong, et al. Acoustic and ultrasonic techniques for defect detection and condition monitoring in water and sewerage pipes: A review[J]. Applied Acoustics, 2021, 183:108282.
[2] HAWARI A, ALKADOURR F, ELMASRY M, et al. A state of the art review on condition assessment models developed for sewer pipelines[J]. Engineering Applications of Artificial Intelligence, 2020, 93: 103721.
[3] CHE, Tongchuan, Duan, Huanfeng, Lee, Pedro J, et al. Transient wave-based methods for anomaly detection in fluid pipes: A review[J]. Mechanical Systems and Signal Processing, 2021, 160: 107874.
[4] KHAN M, PATIL R. Acoustic characterization of pvc sewer pipes for crack detection using frequency domain analysis[C]. 2018 IEEE International Smart Cities Conference, 2018: 1-5.
[5] 李帅永，程振华，毛维培，等. 基于改进经验小波变换及互谱相位差谱的供水管道泄漏声振动定位方法[J]. 仪器仪表学报, 2019, 40(12): 80-91.
LI Shuaiyong, CHENG Zhenhua, Mao Weipei, et al. Leak location based on the improved EWT and cross-power phase difference spectrum of vibro-acoustic signal in water-supply pipelines[J]. Chinese Journal of Scientific Instrument, 2019, 40(12): 80-91.
[6] SAADE M, Mustapha S. Assessment of the structural conditions in steel pipeline under various operational conditions - A machine learning approach[J]. Measurement, 2020, 166:108262.
[7] 伍林峰，冯早，朱雪峰. 基于优化VMD和连续隐马尔科夫模型的管道堵塞状态评估[J]. 振动与冲击, 2020,39(22):214-222+233.
WU Linfeng, FENG Zao, ZHU Xuefeng. Pipeline blocking state assessment based on optimized VMD and continuous hidden Markov model[J]. Journal of Vibration and Shock, 2020,39(22):214-222+233.
[8] Cheng J, Wang M. Automated detection of sewer pipe defects in closed-circuit television images using deep learning techniques[J]. Automation in Construction, 2018, 95: 155-171.
[9] Gong C, Li S, Song Y. Experimental validation of gas leak detection in screw thread connections of galvanized pipe based on acoustic emission and neural network[J]. Structural Control Health Monitoring, 2020, 27(1): 2460.
[10] Xie Y, Xiao Y, Liu X, et al. Time-Frequency Distribution Map-Based Convolutional Neural Network (CNN) Model for Underwater Pipeline Leakage Detection Using Acoustic Signals[J]. Sensors, 2020, 20(18):5040.
[11] 王冉，石如玉，胡升涵，等.基于声成像与卷积神经网络的轴承故障诊断方法及其可解释性研究[J].振动与冲击,2022,41(16):224-231.
WANG Ｒan，SHI Ｒuyu，HU Shenghan, et al.An acoustic fault diagnosis method of rolling bearings based on acoustic imaging and convolutional neural network[J]. Journal of Vibration and Shock, 2022,41(16):224-231.
[12] 张俊红，孙诗跃，朱小龙，等.基于改进卷积神经网络的柴油机故障诊断方法研究[J].振动与冲击,2022,41(06):139-146.
ZHANG Junhong, SUN Shiyue, ZHU Xiaolong, et al. Diesel engine fault diagnosis based on an improved convolutional neural network[J].Journal of Vibration and Shock, 2022, 41(06) :139-146.
[13] 孙洁娣，王利轩，温江涛，等. 改进稠密块轻量化神经网络的管道泄漏孔径识别[J]. 仪器仪表学报, 2022, 43(03): 1-11.
SUN Jiedi, WANG Lixuan, WEN Jiangtao, et al. Pipeline leakage aperture recognition based on lightweight neural network with the improved dense block [J].Chinese Journal of Scientific Instrument, 2022, 43(03): 1-11.
[14] 温江涛，付磊，孙洁娣，等. 压缩感知结合卷积网络的天然气管道泄漏孔径识别[J]. 振动与冲击, 2020, 39(21): 17-23.
WEN Jiangtao, FU Lei, SUN Jiedi, et al. Recognition of leakage aperture of natural gas pipeline based on compression sensing and convolution network[J]. Journal of Vibration and Shock, 2020, 39(21):17-23.
[15] 宁方立，韩鹏程，段爽，等. 基于改进CNN的阀门泄漏超声信号识别方法[J]. 北京邮电大学学报, 2020,43(03):38-44.
NING Fangli, HAN Pengcheng, DUAN Shuang, et al. Identification Method of Valve Leakage Ultrasonic Signal Based on Improved CNN[J]. Journal of Beijing University of Posts and Telecommunications, 2020,43(03):38-44.
[16] 周倩倩，司徒祖祥，腾帅，等. 基于卷积神经网络的排水管道缺陷智能检测与分类[J]. 中国给水排水, 2021, 37(21):114-118.
ZHOU Qianqian, SITU Zuxiang, TENG Shuai, et al. Intelligent Detection and Classification of Drainage Pipe Defects Based on Convolutional Neural Networks[J]. China Water & Wastewater, 2021, 37(21):114-118.
[17] 李哲, 封皓, 刘欣, 等. 基于CNN的水管道小泄漏声信号识别[J]. 噪声与振动控制, 2021, 41(4):66-72.
LI Zhe, FENG Hao, LIU Xin, et al. Acoustic Signal Identification of Small Leakage in Water Pipelines Based on CNN[J]. Noise and Vibration Control, 2021, 41(4):66-72.
[18] 聂维, 江竹, 刘伯相,等. 一维卷积长短期记忆神经网络的管道泄漏检测方法[J]. 中国农村水利水电, 2022, (1):147-152+157.
NIE Wei, JIANG Zhu, LIU Bo-xiang, et al.One-dimensional Convolutional and Long Short-term Memory Neural Network Method for Pipeline Leak Detection[J]. China Rural Water and Hydropower, 2022, (1):147-152+157.
[19] 许子非,金江涛,李春.基于多尺度卷积神经网络的滚动轴承故障诊断方法[J].振动与冲击,2021,40(18):212-220.
XU Zifei ，JIN Jiangtao，LI Chun.New method for the fault diagnosis of rolling bearings based on a multiscale convolutional neural network[J]. Journal of Vibration and Shock,2021,40(18):212-220.
[20] Greff K, Srivastava R, Koutník J, et al. LSTM: a search space odyssey[J]. IEEE Transactions on Neural Networks Learning Systems, 2016, 28(10): 2222-2232.
[21] Niu Z, Zhong G, YU H. A review on the attention mechanism of deep learning[J]. Neurocomputing, 2021, 452: 48-62.
[22] Selvaraju R, Cogswell M, Das A, et al. Grad-cam: Visual explanations from deep networks via gradient-based localization[C]. Proceedings of the IEEE international conference on computer vision, 2017: 618-626.
[23]FENG Z. Condition classification in underground pipes based on acoustical characteristics[D]. University of Bradford, Bradford: University of Bradford, 2013: 1-259.