Electronic Nose: Studies On Gas Sensor Arrays, Systems And Its Applications

Olfaction is a physiological reaction of creatures to odour or volatile substance molecules. Of all the human sensory systems, olfaction is the least understood in terms of the primary receptor mechanism and biological transduction. Electronic mimicking olfaction is still a developing technology. In this dissertation, we report the fabrication of gas sensors, development of the electronic nose systems and applications of the electronic nose in environment monitoring and food quality controlling. The structure and characteristics of the electronic nose, gas sensor materials, gas sensor arrays, array optimization, volatile organic compounds (VOCs) recognition and quantification, liquor identification, characterization of vinegar and influence of pattern recognition to the performance of the electronic nose are systemically studied. ZnO nanoparticles were prepared by thermal evaporation with metallic zinc as raw materials. The obtained ZnO nanoparticles display a mixture of rod shape with 20–50 nm in width and about 150 nm in length and needle shape with 5–10 nm in diameter and about 200 nm in length. Different sensitive materials were produced by mixing the ZnO nanoparticles and some commercial metals or oxides. TEM images proved that the dopants could restrain the grain growth when the ZnO thick films were sintered. XRD results showed that some of the dopants reacted with ZnO nanoparticles and the phase transformation, which related to their resistances and sensitivities, could be observed during the sintering process. Doping can considerably reduce the resistances and improve the sensitivities of the thick films. The sensor arrays were prepared by using integrated circuit method and laser micromachining method. There are 4 ZnO thick films on each array prepared by laser micromachining method. Each array is in size of 7mm×4mm×0.635mm, and low-power-consumed, well-reproduced and low-cost. The electronic nose software system was developed by using the concept of virtual instrument in LabVIEW development platform and contained 7 modules, such as parameters setting, data acquisition, data saving, data opening, data deriving, data calculating and data analyzing. Pattern recognitions were realized in Matlab and the results were transmitted to the electronic nose via an interface between Matlab and LabVIEW. DZB2005, the first portable electronic nose prototype in China, was developed for real-time, on-line and in-situ detection. The advantages of the DZB2005 include rapid to respond, easy to train, power to recognize, etc. Statistical analysises, such as characteristics analysis, correlation analysis, principal component analysis, were used to optimize the original array contained 27 differently doped nano-ZnO thick films. The optimized array was composed of 6 gas sensors by doping different substances, namely, 1wt%TiO2, 5wt%TiO2, 1wt%MnO2, 1wt%CeO2, 4wt%CeO2 and 0.92mol%Ag. The ability of the electronic nose was improved through eliminating the abnormal sensors, reducing the array scale and restraining signal redundancies by array optimization. Consequently, it is proved that the selectivity of ZnO thick films to VOCs were improved by doping. 5wt%TiO2 and 1wt%MnO2 doped ZnO thick films show the different sensitivity features to alcohol and acetone as those of the others sensors in the optimized array. But the selectivity of the thick films to benzene, toluene and xylene were poor. The sensitivities to xylene were always higher than those to toluene. The sensitivities to benzene were the lowest. The accuracy of BP-ANN in terms of predicting individual alcohol, acetone, benzene, toluene and xylene in air was 96%. The liquor flavour is a key element to the liquor quality. 5 commercial liquors, alcohol, diluted alcohol were measured by the electronic nose. The responses of the liquors differed from their flavour type. Liquors belongs to different flavour type were easy to identify, while liquors belongs to the same flavour type were difficult to classify. Learning vector quantization (LVQ), which recognized with the samples similarities and trained with the teacher samples, was better than principal component analysis incorporating with discriminant analysis (PCA-DA) and back-propagation artificial neural network (BP-ANN) in recognition. The accuracy of PCA-DA, BP-ANN and LVQ in terms of predicting liquors was 76.8, 71.4 and 89.3%, respectively, and total accuracy of PCA-DA, BP-ANN and LVQ was 80.4%, 85.7% and 94.6%, respectively. Characterization of vinegars is difficult but very useful to vinegars analysis and vinegars quality control. 17 commercial vinegars, acetic acid and diluted acetic acid weremeasured by the gas sensor array contained 9 nano ZnO thick films doped by some commercial metals or oxides. The type, raw materials, total acidity, fermentation method and production area of the vinegars were chosen to describe and character the 17 vinegars. PCA and cluster analysis (CA) results showed that all these characteristics were not of independent, and the type and fermentation method are more effective than raw materials, total acidity and production when the vinegars were analyzed by the electronic nose. LVQ performed the role of recognition and classification of the vinegars. The accuracy in term of predicting tested vinegar samples was 72.1%, 76.5%, 77.9%, 94.1% and 82.4% according to type, raw materials, total acidity, fermentation method and production area, respectively. The fingerprints of the vinegars were established according to their type, raw materials, total acidity, fermentation method, and production area. There were a good deal of similarities between the outputs of pattern recognition and the fingerprints. The accuracy in term of predicting the 17 vinegars was 85.3% according to the fingerprints. Identification of liquors and characterization of vinegars proved that broad spectrum responses were obtained when liquors and vinegars were measured. The trace components, such as ethanol, ester, acid, acetal and phenol in liquors or vinegars, are the key elements to the odour of liquors or vinegars, and constitute the fingerprints of different liquors or vinegars, though their main ingredients are alcohol or acetic acid, respectively. This is the reason the great differences can be found between the commercial liquors and alcohol and diluted alcohol, and the commercial vinegars and the acetic acid and the diluted acetic acid. The findings of this study would suggest that the electronic nose is powerful to distinguish the forged liquors or forged vinegars from the commercial liquors or vinegars. Linear regress methods, such as multiple linear regression (MLR) and principal component regression (PCR), were not suitable to analyze individual VOCs concentrations because of the non-linear relationships between the responses and the concentrations. Lack of ability to take account all 5 VOCs into a whole was another disadvantage of linear regression methods. Artificial neural network (ANN) can predict individual alcohol, acetone, benzene, toluene and xylene concentrations simultaneously, but their errors were too high to accept. A hybrid two layers ANNs was developed to decrease the errors. The hybrid ANNs includes 2 parts, which contains one sub-ANN and 5 sub-ANNs, respectively. Thepredicting precision was considerably improved, and the average relative error of predicting the alcohol concentrations was only 2.9%. The average relative errors of predicting the other VOCs concentrations were about 10%. Multicomponent analysis (MCA) of binary mixture showed that the gas sensor sensitivities varied from the dopants and the type of VOCs. When alcohol and acetone binary mixture were measured the average absolute error was 7.43ppm while the measured concentration was less than or equal to 10ppm, and the average relative error was 14.26% while the measured concentration was more than 10ppm. The additive property can be observed when alcohol and benzene binary mixture were measured, and the corresponding average absolute error and average relative error improved to 5.13ppm and 9.46%, respectively. Pattern recognition algorithm has become a critical component in the successful implementation of gas sensor arrays and played an important role in odour measurement. Comparisons were made based on four qualitative criteria (calculating speed, training speed, memory requirements and robustness) and one quantitative criterion (classification accuracy). K-nearest neighbor (k-NN), linear discriminant analysis (LDA), BP-ANN, probabilistic neural network (PNN), LVQ and self-organizing map (SOM) pattern recognition algorithms were compared for their ability to classify 4 different datasets obtained by metal oxide semiconductor (MOS) gas sensor array. It is shown that none of the pattern recognition algorithms can meet all the aforementioned criteria. When high selective signals are applied, the statistical theory based algorithms, k-NN, LDA, etc, should be the fittest algorithms. LVQ and PNN are recommended for applications where the selectivity of sensor array is low because of good accuracy and rapid training speed, respectively. Electronic nose system is a huge project and involved with biology, materials, electronics, mechanism, computer, applied mathematics, etc. Many developed country have placed it on the preferential technology to develop. Our works have also suggested the potential applications of the electronic nose for monitoring environment and controlling food quality. Unfortunately, misclassifications and errors occurred frequently in recognition of VOCs, identification of liquors, characterization of vinegars and quantification of VOCs. Further research will be devoted to removing the misclassifications and the errors.Development of new sensing materials, improvement of gas sensor preparation and betterment of pattern recognition algorithms are expected to overcome these limitations and to promote the electronic nose technology. In order to ultimately eliminate the misclassifications and the errors, studies on principle of human olfaction and sensing mechanism of gas sensors are needed.