Research On Key Technologies Of FAIMS Electronic Nose

Electronic nose technology is a gas detection technology based on bionic olfaction.It has been widely used in the fields of food,environment,medical diagnosis,and public safety owing to its advantages of small size and fast response.At present,improving the discrimination accuracy of electronic nose technology for gas detection needs in different fields is still a hot research topic.The intelligent recognition of machine learning algorithms and multi-dimensional information acquisition of sensor combinations are the main means to improve the performance of electronic noses.High field asymmetric waveform ion mobility spectrometry(FAIMS)is a gasphase ion separation technology working at atmospheric pressure.Benefiting from the introduction of microelectromechanical system(MEMS)technology,FAIMS has the advantages of fast,high sensitivity,and system miniaturization in gas detection,which provides a new method for the development of the electronic nose field.At this stage,the FAIMS electronic nose needs to solve the applicability of machine learning algorithms and the contradiction between multi-dimensional information acquisition and miniaturization in the field of gas detection.Aiming at the above problems,this dissertation carried out research on the applicability of machine learning algorithms and the multi-dimensional information acquisition of integrated dual ionization sources for FAIMS electronic nose.The main work is as follows:(1)A new FAIMS electronic nose system was developed and its qualitative and quantitative capabilities were studied.Key hardware modules have been developed through MEMS technology and circuit technology,measurement and control software has been developed using LabVIEW programming,and spectral processing software has been developed using Python programming.Finally,a FAIMS electronic nose system was built by combining machine learning-based spectral analysis methods.Benzene,toluene,and xylene were measured on this system platform,and the FAIMS spectra of the three compounds were obtained.The experimental results showed that the compensation voltage values corresponding to the ion peaks in the FAIMS spectra of the three compounds were different,and the three compounds were distinguished.In addition,the signal intensities of the three compounds were linearly related to their concentrations,and the correlation coefficients were all higher than 98.5%.Thus,this study validated the trace detection,high qualitative,and quantitative capabilities of the FAIMS electronic nose system.(2)The research on applicability of various machine learning algorithms in the application of FAIMS electronic nose was conducted.Taking soy sauce,an important condiment with complex ingredients,as the research object,the applicability of various machine learning algorithms on the FAIMS electronic nose was systematically screened.The features of FAIMS data were extracted using wavelet packet decomposition(WPD)and principal component analysis(PCA),witch were used as input to the classification model.Then,classification models were established using linear discriminant analysis(LDA),k-nearest neighbor(k-NN),support vector machine(SVM),and gradient enhanced decision tree(GBDT),and different types of soy sauce samples were successfully classified.Finally,the classification performances of the models were evaluated using cross validation.The training score,test score,and average cross validation score of the optimal model were all 100%.The results showed that the variety of machine learning algorithms realized the intelligent recognition of FAIMS electronic nose data,PCA showed a better feature extraction effect than WPD,and GBDT showed better classification performance than the other three models.(3)Aiming at the contradiction between multi-dimensional information acquisition and miniaturization of FAIMS electronic nose,the FAIMS electronic nose technology integrating multiple ionization sources was first proposed.The two ionization sources were integrated in the FAIMS drift tube,and more information dimensions could be obtained by utilizing the differences between the ultraviolet photoionization source and the corona discharge source in the range of ionization objects,ionization efficiency,and product types.The complementary advantages formed by two ionization sources in selectivity,quantification,and sensitivity were verified by experiments.The machine learning algorithm was used to process the FAIMS spectral data in different ionization modes,which verified the unique advantage of integrating multiple ionization sources in increasing the recognition dimension while maintaining miniaturization of FAIMS.