铁路隧道二次衬砌敲击检查声音特征分析及智能识别  

Feature analysis and intelligent recognition of percussion inspection sound of secondary lining in railroad tunnel

作　　者：高磊 刘振奎[1] 魏晓悦 张昊宇 张奎 GAO Lei;LIU Zhenkui;WEI Xiaoyue;ZHANG Haoyu;ZHANG Kui(School of Civil Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学土木工程学院,甘肃兰州730070

出　　处：《铁道科学与工程学报》2022年第7期1997-2004,共8页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(11662007,51268031);兰州市科技计划资助项目(2018-4-33)。

摘　　要：为实现铁路隧道二次衬砌背后空洞智能诊断,基于声音识别技术,建立隧道空洞敲击检查声音智能识别模型。收集645段检查锤敲击衬砌的声音样本,运用信号特征分析的基本方法,分析有空洞和无空洞状态下声音信号的时域和频域特征,并提取24维梅尔频率倒谱系数作为机器学习数据集。用主成分分析法降维,经混合粒子群算法优化的支持向量机训练后,建立铁路隧道空洞敲击检查声音智能识别模型,将该模型应用于实际铁路隧道验证其有效性。建立的声音识别模型训练时长为31 s,准确率达95.56%,且能准确对实际工程中的声音样本做出分类。研究结果表明:对2种状态下的声音样本时域特征和频域特征进行对比和分析,不同状态下短时能量和声纹都出现明显的不同。运用PCA-混合PSO-SVM建立的声音识别模型,有着较高的准确率和较快的训练速度,能够根据敲击检查声音准确判断出隧道背后是否存在空洞,如何根据声音特征判断衬砌背后空洞的大小和深度等,将是下一步研究的重点。目前铁路隧道快速无损检测还无法大范围普及,人工检查仍是使用最广泛的检查方法,通过研究敲击检查声音智能识别,为隧道智能化诊断做出新的探索,对加快人工检查速度、提高信息化程度和实现无纸化作业有着重要的意义。Based on sound recognition technology,the sound intelligent recognition model of cavity percussion inspection was established to realize the intelligent diagnosis of cavity behind the secondary lining of railroad tunnel.645 sound samples of inspection hammers striking the lining were collected.The sound signal processing methods were applied to analyze the time-domain and frequency-domain characteristics of these sound signals in the presence and absence of cavities.The Mel frequency coefficients of these sound samples were extracted as a machine learning dataset,which was a 24-dimensional matrix.These datasets were dimensionalized by principal component analysis and trained by support vector machine optimized by hybrid particle swarm algorithm to build an intelligent recognition model of railroad tunnel cavity knocking inspection sound.Finally,the model was applied to the actual railway tunnel to verify its effectiveness.The established sound recognition model,with a training time of 31 s and an accuracy of 95.56%,can accurately make classification of sound samples in real engineering.Conclusions are drawn.1)The short-time energy and sound pattern in different states are significantly different,by comparing and analyzing the time-domain characteristics and frequency-domain characteristics of the sound samples in the two states.2)The sound recognition model established by using PCA hybrid PSO-SVM has a high accuracy rate and faster training speed,and can accurately determine whether there is a cavity behind the tunnel based on the sound of knocking inspection.How to determine the size and depth of the cavity behind the lining based on the sound characteristics will be the focus of the next research.3)Rapid nondestructive testing of railroad tunnels is not yet widespread.Manual inspection is still the most widely used inspection method.Through the study of percussion inspection sound intelligent recognition,to make a new exploration of intelligent tunnel diagnosis,to speed up the speed of manual inspection,to impr

关 键 词：铁路隧道工程 声音识别 梅尔频率倒谱系数 主成分分析 混合粒子群优化算法 支持向量机 

分 类 号：U25[交通运输工程—道路与铁道工程]