【作者】 贾雅琼；

【导师】 王殊；

【作者基本信息】 华中科技大学 ， 生物信息技术， 2013， 博士

【摘要】 气体探测的应用领域非常广泛,覆盖了工业、农业、环保、国防、航天航空及日常生活等各方面。相比其他传统气体探测技术,声学气体探测具有诸多优点：无需标定、重复性强、实时性强、可同时检测多种气体、无需预处理气体、不损耗气体等,已成为气体信息传感与检测领域中的前沿技术和重要方法。气体超声波谱由气体声吸收谱和声速谱组成,分别对应气体声吸收系数和声速随声波频率的变化曲线。构建气体超声波谱的方法是声学气体探测的重要基础,故深入分析气体的声弛豫过程以构建气体超声波谱是声学气体探测要解决的关键问题之一。而利用信号处理技术完成基于超声波谱的气体探测理论及其研究方法,是声学、量子物理、信号处理等学科交叉的前沿研究。本论文根据声在气体中的传播特征,结合气体在声波扰动下的声波方程,在研究气体声吸收理论、气体声速理论和超声信号处理理论的基础上,得出了气体声弛豫过程的分解对应模型和复合弛豫时间模型,并完成了气体超声波谱的构建,最终利用小波多分辨率分析和支持向量机对气体声弛豫吸收谱进行分析,实现了混合气体成分识别。本论文进行的研究工作主要有以下几个方面：1.通过研究气体声弛豫过程中振动自由度与平动自由度(V-T)以及振动自由度之间(V-V)的分子能量转移模型,给出了有效比热容与弛豫时间的分解对应关系及其通用获得方法。该分解模型与现有的声弛豫模型相比,反映了分解后的V-T和V-V弛豫过程中振动比热容与弛豫时间的对应关系,并发现了较高能级是引起对应声弛豫过程的决定因素。将基于该分解模型获得的气体声弛豫吸收谱经碰撞直径微调改进后,其结果比现有理论更接近实验数据,证明了该分解对应关系的正确性和合理性。2.提出了复合弛豫时间的倒数和模型,并利用其与有效定容比热的关系构建气体的弛豫声吸收谱。混合气体的有效弛豫频率可以直观地从弛豫声吸收谱上得到,有效弛豫时间可以通过和有效弛豫频率的关系得到。利用这种方法得到的混合气体的弛豫声吸收谱与已有实验数据相比较,误差小,且与谱的变化趋势一致,论证了该方法的正确性。同时将该方法得到的弛豫声吸收谱的数据与多个已有理论对应的声弛豫吸收谱比较,数据的变化规律一致,证明了该方法的有效性。利用该方法构建了空气及其含碳气体的声弛豫吸收谱,为后续的气体识别提供基础。3.通过统计已发表文献中关于气体超声波谱的理论和实验数据,并利用本文提出的有效比热容与弛豫时间的分解对应关系的理论模型扩展这些数据,建立了不同成分和浓度下N2、O2、H2O2、CH4、H2、CO2等多种混合气体和空气的超声波谱数据库。重点研究和提取了气体超声波谱线上的核心信息点——主弛豫点的声吸收系数及其对应声频率,通过统计气体超声波谱数据库的核心信息点数据,建立了常见气体超声波谱作用区统计图；发现了利用气体超声波谱作用区定性探测气体成分的理论方法,并初步形成了利用主弛豫点的吸收系数变化幅度和声频率变化幅度定量检测气体成分的理论方法。该研究利用统计学方法从物理上证明了基于超声波谱的气体探测的正确性和可行性。4.将小波多分辨率分析和支持向量机分类等经典信号处理方法引入超声波谱气体探测。声弛豫吸收谱线的数值化分析是基于超声波谱的气体探测的关键,利用小波多分辨率分析提取声弛豫吸收谱的特征,并通过计算特征参数各自的识别率选出同时具有高识别率和低计算代价的特征参数,用于多分类支持向量机的训练和检验。使用训练后的支持向量机,完成了对空气、空气和CO、空气和CO2、以及空气和CH4等四种混合气体的多分类识别,仿真结果表明识别率达到100%。从而实现了从多元混合气体和空气等复杂背景中探测一氧化碳、二氧化碳和甲烷等一种和多种气体信息的传感方法,完成了基于超声波谱的气体探测的理论研究工作。本论文工作得到了国家自然科学基金项目“基于超声波谱的气体探测”(编号：60971009)和“基于有效弛豫时间的气体探测方法研究”(编号：61001011)的资助。本论文的研究成果不仅从微观上深入分析了气体声弛豫过程,得出了有效比热容与弛豫时间的分解对应关系,结合平动弛豫时间和振动弛豫时间得出了复合弛豫时间倒数和的理论模型,并将其应用于构建宏观气体超声波谱；通过大量统计文献中的气体超声波谱的理论和实验数据,构建了气体超声波谱的数据库,从物理基础上证明了基于超声波谱的气体探测的可行性；在此基础上,利用小波理论和支持向量机理论进行声学气体探测,提供了一种新的超声气体探测的思路和方法。更多还原

【Abstract】 Gas detection has been used in many areas:industry, agriculture, environmental industry, national defense, aerospace industry, and daily life. Compared with traditional technologies of gas sensing, acoustic-based gas sensing has many advantages:no calibration, strong repeatability, real-time response, simultaneously detecting several kinds of gas, no need to preprocess gas, and no loss of gas. Acoustic gas detection (AGD) has become cutting-edge technology in the field of gas-information sensing and detecting techniques.AGD is primarily based on establishment of gas ultrasonic spectrums (GUS) which consist of gas acoustic absorption spectrums-gas acoustic absorption coefficient dependent on the acoustic frequency, and sound speed spectrums-sound speed dependent on the acoustic frequency. So the first thing needed for AGD is analyzing the gas acoustic relaxation process to establish GUS. After constructing the GUS, signal processing technology is used for the further research in ultrasonic-spectrum-based gas detection which is cutting-edge research of interdisciplinary including acoustics, quantum physics, and signal processing.Based on acoustic propagation-in-gas characteristics and acoustic wave equations under the condition of gas being disturbed by acoustic waves, via researching gas acoustic absorption theory, gas sound speed theory, and ultrasonic signal processing theory, a decomposition model and a multi-relaxation time model are obtained in the process of gas acoustic relaxation; GUS is also built up. Based on gas acoustic relaxation absorption spectrum (GARAS), the gas detection is realized by utilizing wavelet multi-resolution analysis (WMRA) and multi-class support vector machine (MSVM).The main research results are as follows:1A decomposition model---the decomposition relationship between effective specific heat capacity (ESHC) and relaxation time---and the general method of getting the model is obtained via researching the model of molecule energy transfer among vibration-translation (V-T) and vibration-vibration (V-V) in gas acoustic relaxation process. Compared with current acoustic relaxation models, this model has two characteristics:(1) the relationship between vibrational specific heat capacity and relaxation time in the process of V-T and V-V relaxation is obtained;(2) it is discovered that higher energy level is the determining factor of causing relaxation. Then, the model is modified by fine tuning collision diameter. The modified model suits the GARAS closer to experimental data by comparing with spectral lines from existing theories.2A reciprocal-sum model of multi-relaxation time is proposed, and using the relationship between the model and ESHC, GARAS is built up. GARAS functions to offer two physical effective relaxation frequency and effective relaxation time. Such method of establishing GARAS is correct:via comparison between collected experimental data and GARAS obtained by this method, only small error exits and the trend of spectra lines are the same. And the method is effective:through comparing data between spectrums from this method and spectrums from several existing theories, it is discovered that the changing rule of data are the same. This method is used to establish the GARAS of air and the gas mixtures including air to supply fundament in the further research.3Based on GUS theories and data collected systematically in published papers, and the data extant by the decomposition model, a database of gas acoustic spectra, which includes acoustic absorption spectrums and sound speed spectrums, has been built to record common gas mixtures from gases as diverse as N2、O2、H2O、CH4、H2、CO2and air. By statistics of GARAS key information---the primary peaks---from the database, a graph concerning GARAS key information areas is obtained. By researching the graph, two methods are obtained:(1) detecting qualitatively gas-components;(2) sensing quantitatively gas-components based on the changing range of the maximum acoustic relaxation absorption coefficient and acoustic frequency. Physically, the result proves that gas compositions detection based on gas acoustic spectra is feasible.4WMRA and MSVM, classic signal processing methods, are introduced for the first time in AGD. WMRA and MSVM are used for analyzing GARAS numerically---the key of AGD technology. WMRA is used to get features of GARAS. And then, the feature coefficients with high recognition rate and low computation cost are selected from these features and put into MSVM to train and to test it. The trained MSVM will help recognize four types of gas mixtures (air, air and CO, air and CO2, air and CH4) successfully. The simulation results demonstrate that the recognize accuracy of the approach is100%for four types of gas mixtures. In place of a traditional way is a new method of detecting one or several gases (CO, CO2, and CH4) from multi-component mixtures of gas such as air---a theoretical research of gas detection based on ultrasonic spectrums.This dissertation is funded by two projects listed in the National Natural Science Foundation of China:ultrasonic-based gas detection (Grant Nos.60971009) and the research of effective-relaxation-time-based gas sensing (Grant Nos.61001011). This dissertation introduces several research findings:(1) the relationship between ESHC and relaxation time is obtained via analyzing gas acoustic relaxation process;(2) the theory model of reciprocal-sum of multi-relaxation time is obtained by utilizing translational relaxation time and vibrational relaxation time and used to build up GUS;(3) the GUS database is established, which physically proves that ultrasonic-spectrum-based gas detection is practical;(4) WMRA and MSVM are used to acoustic gas detection, which suggests a new train of thought in the area of acoustic gas detection.更多还原