摘要
在引调水工程中,钢丝断裂易导致预应力钢筒混凝土管(Prestressed Concrete Cylinder Pipe,PCCP)出现结构破坏和功能失效。本研究通过智能学习模型分析原型试验信号特征并判别断丝类型。采用内径3.4 m,长度5 m埋置式PCCP开展断丝原型试验,利用分布式光纤传感器分别对切割断丝、腐蚀断丝和敲击噪声信号进行实时监测,并基于短时傅里叶变换和深度学习模型重构断丝信号,最终利用支持向量机建立了断丝信号识别模型。以Inception-ResNet-v2重构信号的断丝识别模型最低和最高准确率分别为92.9%和100%,之后利用t-SNE证明了信号重构的有效性。本研究结合不同智能学习方法实现了PCCP断丝类型的有效判别,为其长期运行断丝监测及预警分析提供了新的手段。
In water diversion projects,the breakage of steel wires can easily lead to structural and functional failure of Prestressed Concrete Cylinder Pipes(PCCP).This study aims to analyze signal characteristics and identify the type of wire breakage using intelligent learning models.In the research,A prototype test for wire breakage was conducted on an embedded PCCP with an inner diameter of 3.4 meters and a length of 5 meters.Real-time monitoring was carried out using a distributed optical fiber sensor to detect cutting wire,corrosion wire,and impact noise signals.Based on Short-time Fourier Transform(STFT)and deep learning models,the wire breakage signals were reconstructed.Finally,a wire breakage signal recognition model was established using a support vector machine.Based on the reconstruction of signals using Inception-ResNet-v2,the lowest and highest accuracy of the wire breakage recognition model are 92.9%and 100%,respectively.The effectiveness of the signal reconstruction was also demonstrated using t-SNE.This study has achieved effective recognition of wire breakage types by combining different intelligent learning methods.It provides new methods for long-term wire breakage monitoring and early warning analysis in PCCP operation.
作者
张野
袁思敏
李炎隆
温立峰
司政
孙凯宇
ZHANG Ye;YUAN Simin;LI Yanlong;WEN Lifeng;SI Zheng;SUN Kaiyu(State Key Laboratory of Eco-hydraulics in Northwest Arid Region,Xi’an University of Technology,Xi’an 710048,China)
出处
《水利学报》
EI
CSCD
北大核心
2023年第5期587-598,共12页
Journal of Hydraulic Engineering
基金
国家自然科学基金青年项目(52009109)
博士启动基金项目(104-451120005)
陕西省自然科学基础研究计划—引汉济渭联合基金项目(2021JLM-46)。
关键词
预应力钢筒混凝土管
深度学习
知识迁移
短时傅里叶变换
断丝信号
智能识别
prestressed concrete cylinder pipes(PCCP)
deep learning
knowledge transfer
short-time Fourier transform
wire broken signal
intelligent identification
