基于声学特征分析地下供水/排水管道安全状态的模式识别系统研究

The issue of water supply/water waste pipeline condition assessment in our country is that we mainly focus on “passive detection” which applied after the events occurred, and target on leakage conditions only. It is a demand of the society’s development that the pipeline detection methods to transform gradually into “active techniques”. Acoustic based techniques are increasingly adopted in condition assessment of utilities since they work on relatively low-budget equipments while present satisfying capability of collecting signals underground and it is easy to perform. Recent researches in this field are concentrating on improving the sensitivity of acoustic sensors and developing comprehensive signal analysis techniques. We introduce pattern recognition theory into this project looking to classify and identify water pipe defects (leakage, sediment, deformation etc.) based on extracted acoustic features, furthermore, to be able to provide a systematic safety evaluation of a period length of water pipeline. The laboratory simulated data have proven the feasibility and encouraging accuracy of this research. The main objectives of this project are: to develop the methods of processing field collected data containing multiple defects with more complicated background noise; to improve the efficiency and the stability of the system, and to lower the costs as much as possible without sacrificing the system’s performance. The main achievements of this project shall include: a suit of signal processing, feature extraction and pattern classification methods for acoustical signals; a database containing groups of acoustic features/parameters with each group represents a common pipe defect or pipe accessory. The original data shall be collected from pipes of different diameters and materials so that the database could provide overall information for research fellows in this field to compare with and use for reference.

目前我国对地下水管道的安全检测方法偏向以被动检测为主且主要针对单一泄漏故障检测，管道检测技术逐渐由事后补救的被动检测向主动预防检测转变是社会发展的需要。基于声学的检测方法具有使用成本较低，操作较简易，信号易于收集等优势而被广泛采用，其相关技术当前存在的问题主要集中在信号采集设备的敏感度和对声学信号的深入分析方法等方向。本课题引入模式识别方法通过对声学特征的提取和分析达到对多发性管道缺陷（泄漏，阻塞，变形等）的识别进而实现对管道的整体预防性风险评估。实验室模拟数据已初步证明了此研究的可行性和较高的准确度。如何有效处理背景环境更加复杂的现场数据，提高识别系统的性能和效率，降低使用成本是本课题的主要研究目标。本课题研究结果包含建立常用管材管径的地下水管道的多发性故障和主要管道配件的声学参数数据库，及一套信号分析，特征提取和分类识别的理论分析方法，将为同类研究提供数据参考和方法借鉴。

针对由于材质老化，设计缺陷，地理条件变化和人为原因造成的埋地排水/供水管网的运行故障例如管道泄漏，阻塞和变形等，本项目采用声学主动检测手段，利用低频声波在封闭空间内传播不易受管道运行工况影响、传播距离长且衰减缓慢等相关特性，对管道内以复杂多重堵塞为主的多类故障展开了分析识别研究。通过项目的实施，完成了对声波信号在通用类型管道中的传播特性研究，利用声波的衰减、散射和反射等特性，采用数字信号处理方法对声信号进行分析和提取，获得了一系列能够代表和反应管道状态和不同状态类别间区别的声学特征参数，可为同领域的相关研究提供参考和借鉴。在充分和深入讨论了声波传播特性及其特征参数的基础上，设计完成了能够对声学特征实现有效分类进而对其反映的管道状态实现准确识别的模式识别系统，同时考虑到管道状态类别在应用场景下的样本分布不均衡性和故障小样本特性，对识别算法进行了调整和优化，使本系统更具有应用前景。采用实验室模拟运行的实验数据，本系统优化后对单一堵塞等故障的识别准确率达到95%以上，对复杂多重故障的识别准确率达到90%左右。目前我国对地下管道的检测方法基本以事后检测为主，本项目采用主动检测机制对地下水管网进行检测和预防控制，可实现对管道日常运行状态的定期监测，在故障发生之前对其进行预防控制，本研究所采用的低频声波传感器及信号处理软件在造价成本和操作难易度上都大大低于传统使用的闭路监控和管内机器人等，更在对检测对象的定量和定位上有直观的估测，更易实现大范围的工程应用普及。通过本项目的实施，为地下管道安全检测从事后检测向主动监测的逐步过渡提供了新的思路和一套理论分析方法，建立了包含多种常见管道故障的声学参数数据库并填补此类研究的空白，为声学检测技术、计算机技术和信息技术的融合提供了数据和理论支撑，进一步促进了管道故障的科学控制管理。