Study On Data Driven Spectral Analysis Of Complex Systems And Its Applications

Multi component analysis of complex system,which provides key information support for industrial informatization,represents one of the challenges in on-line monitoring and process control of industrial production.With the accelerated development of industry informatization in recent years,a high request to multi component analysis of complex system is set by industrial field,which requires the detection techniques to be high-throughput,real-time response,nondestructive and noncontact.Among all the detection techniques,spectrum detection technologies represented by vibrational spectroscopy have become the most preferred method in complex system analysis for its advantages of high-throughput and rapidness.However,due to the highly overlapping and complex spectral band,the processing of complex system spectra must rely on the chemometrics to effectively extract the quantitative information of analytes from the matrix interference.Therefore,the development of novel chemometrics methods is the key to realize the analysis of complex system spectrum.Using vibrational spectroscopy along with chemometrics methods,this dissertation investigate the production mechanism and analytical theory of vibrational spectra,and thus sets up novel spectral analysis methodologies based on data driven theory.Through analysis of the spectral characteristics and principles,these strategies are capable of encoding the analyte information contained in complex system.The main contents and contributions of this dissertation are listed as follows:1.In order to solve the problem of overlapping spectral of multiple components in complex system,this thesis established a novel strategy of data driven spectral analysis(DDSA).This strategy begins with the experimental design.With the well-designed data sets,DDSA guides the subsequent steps to accurately extract the data characteristics of target object.A higher-density discrete wavelet transform(HDWT)method is then introduced to process the spectral data,and this method significantly enhances the spectral resolution by using the double domain oversampling,which effectively reduce the interference between the overlapping spectra.Finally,a variable selection method is used to select the optimal combination of variables from the complex HDWT coefficients for the DDSA.2.This dissertation sets up an on-line gas analysis system for mud logging on oil and gas exploration field,and then develops an over sampling data driven(OSDD)algorithm to encode Raman spectra of the gas mixtures.With the OSDD,the resolution and reliability of Raman spectral analysis were significantly improved.As a result,the Raman system quantifies 12 alkane and non-hydrocarbon complex mixture gases,which testifies the capability of the OSDD when dealing with the overlapping spectrum in complex system.The OSDD also compared favorably with the gas chromatography(GC)method,illustrating the feasibility of OSDD in mud logging industry.3.In order to overcome the collinearity problem in the spectral analysis of complex system,a prior knowledge oriented data driven(PKODD)algorithm is developed.Meanwhile,a highly sensitive infrared detection device is established to detect the methanal contained in baby clothes.The spectral data were collected by a FTIR spectrometer and then processed by the PKODD algorithm.The research result shows that the PKODD can effectively suppress the variable interference that caused by the collinearity of methanal and methanol,thus effectively detects the presence of methanal in baby clothes.4.A digitally labeled data driven(DLDD)algorithm is developed to precisely extract the weak analytical information from the raw Raman spectra containing fluorescence interference.DLDD efficiently isolates the features of triclosan from matrix components,enabling the nondestructive quantitative analysis of triclosan in real antibacterial hand soaps.We demonstrate the feasibility of DLDD using a practical test in which we seek to quantify triclosan in several kinds of antibacterial hand soaps nondestructively.The results illustrate DLRS as a promising tool for nondestructive detection of triclosan in antibacterial hand soaps without any sample pretreatment,which may well extend to digital separation of some analytes of interest in the presence of complex chemical substances.The results illustrate that the DDSA strategy provides a promising tool for quantitative analysis of complex system.With the excellent adaptability,this strategy may greatly promote the applications of vibrational spectroscopy in industry.