【作者】 张顺平；

【导师】 谢长生；

【作者基本信息】 华中科技大学 ， 材料学， 2009， 博士

【摘要】 电子鼻是一种模拟生物嗅觉的气体／气味分析仪器，相对于传统的气体分析仪器，其具有分析快速、操作简单、可便携、成本低等优点，可应用于食品质量检测与控制、环境监测、公共安全、医疗卫生、航空航天等一系列国家重大需求领域。电子鼻中最核心的组成部分是气体传感器阵列，其中金属氧化物（MOX）气体传感器是应用最广泛的一类传感器。电子鼻技术应用推广中存在许多科学技术瓶颈。本文分析和探讨了电子鼻技术应用研究中的一些共性问题，其中主要包括MOX气体传感器响应信息提取的研究、MOX气体传感器敏感机理与响应模型的研究，以及气体传感器阵列的选择性优化研究。MOX气体传感器响应信息提取的研究中，分别在时域空间和相空间中分析了传感器响应曲线的特性，并建立了一种基于特征信息含量和相关性分析的零散特征快速提取方法，和一种基于相空间中传感器响应模式的全特征参数提取方法。其中所建立的零散特征快速提取方法，基于特征参数信息含量和相关性的分析，以在最短时间内提取对气体类别区分能力大、相关性小的参数作特征参数为原则，可快速提取信息量充足的特征参数。在易燃液体快速检测的一个应用中，基于该方法可在10s内提取积分、差分、微分和二次微分信号为特征参数，与传统的响应幅值特征参数提取方法相比有更优的性能，两者对各类易燃液体和不可燃饮料的正确识别率分别为85.7％和57.1％。所建立的全特征参数提取方法，基于对相空间中传感器响应动力学特性分析所得到的响应模式规律，提取了六个特征参数，同时基于这些参数可以还原出响应曲线，还原出的响应曲线与原始响应曲线的平均还原误差仅为5.4％。为了提高电子鼻的在线检测速度，缩短传感器阵列响应样本后恢复到初始态的时间，同时还分析了传感器恢复曲线的特性，并建立了一套快速提取恢复特征参数的方法与装置，其对样本的检测恢复时间仅为稳态测试条件下的42.7％，可以较大程度上降低检测恢复时间，提高在线检测的速度。最后基于海鲜中甲醛的检测为应用，对特征参数的稳定性进行分析与对比，分析得到恢复曲线特征参数min（dS_t／dt／）／max（S_t／）较传感器空气状态下的电阻R₀以及响应过程中的敏感幅值S更为稳定，三特征参数最小误差分别为2.1％、9.9％和7.2％。MOX气体传感器敏感机理与响应模型的研究中，基于氧离子化模型与氧空位模型推导出了一个MOX传感器与气体反应的动力学模型，并利用该模型定量描述了相空间中气体的响应模式，该模型描述数据中响应模式的平均模拟误差仅为3.98％。同时基于该动力学模型建立了一个传感器性能参数数据库，以及一个模式匹配的模式识别方法，利用该方法可以有效的对样本进行正确识别。同时还分析了该动力学模型可能的其它应用，包括研究工作温度、环境湿度、环境氧分压、敏感膜的厚度、敏感层晶粒的大小、敏感层表面形态、掺杂等对响应模式的影响。最后，基于模型假设与实验数据验证，证明传感器对不同类别气体存在不同响应模式的原因是，不同类别气体与气敏材料反应时，不同类别的气体是在与不同的氧空位／离子化的氧反应，或反应的几率不同。气体传感器阵列的选择性优化研究中，介绍了四种常见的特征选择阵列优化方法，并利用这些方法优化了MOX传感器阵列的工作温度，将10个初始工作温度为300℃的传感器组成的阵列，优化为一个工作温度为220℃的四传感器阵列。同时建立了一种基于亚阵列的阵列优化方法，利用该方法可以获知优化阵列中各传感器的独特功用，在利用六传感器（TGS813、TGS2600、TGS2602、TGS2610、TGS2611和TGS2620）对11种样本（苯、甲苯、二甲苯、乙醇、甲醇、丙酮、丁酮、甲醛、乙醛、正戊烷以及环己烷）的识别中，采用基于亚阵列的阵列优化方法，可将阵列优化为三传感器阵列（TGS2600、TGS2602、TGS813），且在该优化阵列中，传感器TGS2600独特的功用是识别丁酮和乙醛；传感器TGS2602独特的功用是识别苯和环己烷、甲醇和乙醇；传感器TGS813独特的功用是识别环己烷和正戊烷；传感器组合TGS2600和TGS2602独特的功用是识别丙酮和丁酮、丙酮和乙醛。更多还原

【Abstract】 Electronic nose is an instrument for gas/odor analysis which simulates the biologic olfaction. Comparing with the traditional gas analyzing instruments, it has the virtues of high speed for analyzing, facility in operation, easiness to carry, low cost etc. Electronic nose could be used in many application fields for important national requirements, such as food quality assessment and control, environment, security, sanitation, avigation etc. Gas sensor array is the most important part in electronic noses. Metal oxide （MOX） gas sensor is the widely used sort of gas sensors. There are several problems in the applications of electronic nose. These problems mainly included three subjects, which were the feature extraction from the response curves of MOX gas sensors, reaction model analysis of MOX gas sensors, and the sensor array optimization. These problems were analyzed in this paper.In the research of feature extraction analysis, the characters of response curves in the time domain and in phase space were analyzed. A piecemeal signal feature extraction method based on information and relativity analysis and a entire feature extraction method were established. In the piecemeal signal feature extraction method, features from the 10th second of the integrals, differences curves, and the 5.7th second of the primary derivatives curves, and the 6.2th second of the secondary derivatives curves were extracted. In the entire feature extraction method, six features were extracted. With these features, the response curves could be reconstructed. The mean error of the reconstructed response curves from the original signal response curves was 5.4%. In order to reduce the response-recovery time of electronic nose, the characters of recovery curves were analyzed. And an electronic nose with nine metal oxide gas sensors and a method of feature extraction on sensor recovery curves were established to reduce response-recovery time. With the electronic nose and the feature extraction method, the mean response-recovery time in the measurements was 33.5 s, which was about 42.7% of the response-recovery time in typical traditional gas sample measurements. Finally, the feature stabilities were compared in the measurements of formaldehyde-containing detections in octopus. The minimum relative errors of static features R₀ （resistance in the air） , S （sensor response） , and one dynamic feature DR （desorption rate） were 2.1%, 9.9% and 7.2%. In the reaction model analysis of MOX gas sensors, a reaction model of MOX gas sensors is established to simulate the sensor response patterns, where the mean simulation error was 3.98%. A performance database of sensors and a pattern matching method were built for gas sort classification without any usual pattern recognition methods. The other applications of the reaction model were also analyzed, including the researches of the influence of temperature, O₂ concentration, humidity, film thickness, grain size, doping, catalyzer etc. on the sensor response properties. Finally, the reason of different response patterns with different gas sensors was analyze. The results showed that on the surfaces of sensing materials, different gases sorts reacted with different active dots, or with different reacting probabilities.In the sensor array optimization analysis, four common feature selection method were used in working temperature selection of MOX sensor array, where a 10 sensor array with working temperature of 300℃was optimized to a 4 sensor array with working temperature of 220℃. In the sensor array optimization analysis, a sensor array optimization method based on sub-array was also established to analyze the unique functions of each sensor in the optimized array. A measurement with a 6 TGS sensors array （TGS2600, TGS2602, TGS2610, TGS2611, TGS2620 and TGS813） to classify 11 gas sorts （benzene, toluene, xylene, acetone, butanone, methanol, ethanol, formaldehyde, acetaldehyde, pentane and cyclohexane） was used in the data validation. The sensor array was optimized to 3 sensors with the method. Each sensor in the optimized array had unique functions to solve different difficult tasks. TGS2600 had the unique functions to discriminate Butanone and Acetaldehyde. TGS2602 had the unique functions to discriminate Benzene and Cyclohexane, Methanol and Ethanol. TGS813 had the unique functions to discriminate Cyclohexane and Pentane. The combination of TGS2600 and TGS2602 had the unique functions to discriminate Acetone and Butanone, Acetone and Acetaldehyde. The proposed method might be a new generation of sensor array optimization methods.更多还原