基于气味分析的设备异常检测方法研究

发布时间：2018-04-06 21:03

本文选题：异常检测　切入点：气味模拟　出处：《国防科学技术大学》2011年硕士论文

【摘要】：载人航天、深海探测、大型飞机等技术的发展,对宇宙飞船、潜艇、大型客机等具有非开放空间的大型系统内设备安全运行提出了越来越高的要求。对于这类系统而言,密闭座舱内大量设备长期运行,特别是状态异常所产生的各种污染是影响空气质量的主要原因之一。同时,空气成份中的这些污染也包含了反映设备运行状态的重要信息。基于仿生原理的人工嗅觉分析(电子鼻)技术是对气味检测的一种重要手段,可望为非开放空间设备密集系统的健康监控和早期异常检测提供新的技术途径。为此,本文在综述电子鼻技术研究现状的基础上,针对密闭空间设备异常容易产生的油液渗漏和导线过热问题,系统开展了设备异常气味识别与分离检测方法研究。本文的主要研究工作包括: (1)在分析电子鼻系统工作原理与结构组成的基础上,对嗅觉检测涉及的传感器阵列、信号预处理、模式识别等关键技术的国内外研究现状进行了总结和探讨。 (2)对密闭空间油液渗漏和导线过热进行了气味模拟实验与响应分析。在对实验环境、实验方法及实验装置进行介绍的基础上,分析了电子鼻系统传感阵列响应过程的影响因素,并对设备异常传感阵列信号进行了初步分析。 (3)针对设备异常状态气味定性识别问题,在对实验数据进行预处理的基础上,通过主成分分析提取了实验数据的主要特征,分别采用线性判别分析和前馈神经网络实现了异常气味的定性分类。 (4)针对异常状态气味的定量识别问题,研究了基于补气过程中气味浓度补偿的复频域分析方法,提出了嗅觉传感器信号与浓度的非线性关联模型和主成分回归模型,并利用实验数据对这两种模型进行了验证。 (5)对设备异常状态混合气味的分离识别问题进行了探索研究,利用独立分量分析,提出了气味源盲分离模型,并对混合气味的实验数据进行了分离识别。
[Abstract]:With the development of manned spaceflight, deep-sea exploration, large aircraft and other technologies, the safe operation of equipments in large systems with closed space, such as spaceships, submarines, airliners, etc., has been put forward more and more high requirements.For this kind of system, a large number of equipments in the airtight cabin run for a long time, especially the pollution caused by abnormal state is one of the main reasons that affect the air quality.At the same time, the air pollution also contains important information to reflect the operation state of the equipment.Artificial olfactory analysis (electronic nose) based on biomimetic principle is an important means of odour detection, which is expected to provide a new technical approach for health monitoring and early abnormal detection of closed space equipment intensive systems.In this paper, based on the review of the current research situation of electronic nose technology, aiming at the problems of oil leakage and overheating caused by abnormal equipment in confined space, the methods of equipment odour recognition and separation detection are studied systematically.The main research work of this paper includes:1) based on the analysis of the principle and structure of electronic nose system, the research status of sensor array, signal preprocessing, pattern recognition and other key technologies involved in olfactory detection are summarized and discussed.(2) the odour simulation experiment and response analysis of oil leakage and wire overheating in airtight space were carried out.Based on the introduction of the experimental environment, the experimental method and the experimental device, the factors influencing the response of the sensor array in the electronic nose system are analyzed, and the signal of the abnormal sensor array of the equipment is analyzed preliminarily.3) aiming at the problem of qualitative identification of odour in abnormal state of equipment, the main characteristics of experimental data are extracted by principal component analysis on the basis of pretreatment of experimental data.Linear discriminant analysis (LDA) and feedforward neural network (FNN) are used to classify odors qualitatively.(4) aiming at the problem of quantitative identification of odors in abnormal state, the complex frequency domain analysis method based on the compensation of odor concentration in the process of replenishing gas is studied, and the nonlinear correlation model and principal component regression model of the signal and concentration of olfactory sensors are proposed.The two models are verified by experimental data.In this paper, the problem of separation and recognition of mixed odors in abnormal state of equipment is studied. By using independent component analysis, a blind separation model of odour source is proposed, and the experimental data of mixed odors are separated and recognized.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3;TB17

【参考文献】