Study And Application Of Raman Spectral Analytical Algorithms For Different Mixture Systems

On-line detection and analysis technology for mixture composition,which is regarded as an important part of the process industrial information has been more and more widely applied in the industrial product process.Among different techniques,Raman spectroscopy detection generally becomes popular in the field of non-contact detection technology because it is fast and free of calibration.At the same time,the study of the related chemometric methods is also fast developed Aiming at solving the detection problems of the key components in the actual industrial product process,this thesis proposes a series of Raman spectral analysis algorithms corresponding to different types of mixture systems,based on which we have developed an online Raman spectral analysis function module and succeeded in applying it to product process.The thesis specifically includes(1)We propose the method of interference peak subtraction partial least square(IPS-PLS)for detection of the low-content benzene concentration in gasoline product.During the analysis,the algorithm first fits the peaks in the characteristic spectral band with nonlinear profile functions,and then subtracts the interference peak to obtain effective signal for partial least square(PLS)regression.In the experimental part,IPS-PLS and PLS methods were applied for benzene quantitative analysis based on samples of gasoline sampled in an oil refinery.The experimental results demonstrate that,IPS-PLS method is superior to the direct PLS in that it describes the spectral nonlinearity with nonlinear functions and helps decrease principal components needed by PLS regression model,meanwhile,IPS-PLS shows higher accuracy.As a conclusion,the IPS-PLS method based on Lorentz fitting function shows higher efficiency and accuracy,so it is more suitable for benzene detection in gasoline product process(2)For mixtures with parts of components known,we propose the method of alternating least square with peak addition and Lorentz constraint(PAL-ALS)to solve the spectra of unknown components and the coefficients for all the components.Based on the linear superposition model of the mixture spectra,the algorithm calibrates the mixture spectra directly on the spectra of known components and estimates whether there exist background spectra in the residual spectra and then appends the estimation of the new spectrum into the known pure spectral matrix.The algorithm finally optimizes the profile of the unknown spectra by alternating least square with Lorentz constraint The above steps are repeated until there is no more background spectrum emerging and the fitting error of the mixture spectra converges.In the simulation and experimental part,the algorithm was applied to resolve the mixture spectra of the p-xylene and its isomers with unknown components,and the result demonstrates that the algorithm can identify the unknown component spectra and resolve the coefficients for all components accurately.(3)For the case where the spectra of all the compositions are unknown or the deformation of mixture spectra is severe,a Raman spectral decomposition algorithm based on multivariate curves resolution-alternating least square with Lorentz function constraint is applied.This method is based on the linear superposition model forrmixture spectra,and only the mixture spectral matrix is needed to perform the analysis The algorithm first decomposes the mixture spectra with principal component resolution;and then the evolving factor analysis is performed to find how the distribution of the mixture spectral eigenvalues evolves along the pixel direction,so that it can determine the number of components in the mixture,the range of their peaks and the initial estimation for their spectra;in the last step,an alternative least square method with Lorentz function constraint is proposed to optimize the pure spectra.In the experimental part,our proposed algorithm was adopted to decompose the Raman spectra of the circulating fluid in an adsorption tower of a p-xylene producing unit,with which we can obtain the pure spectra in the mixture fluid Finally,with the repeated quantitative analysis experiments for its main component of p-xylene,we proved the effectiveness for the proposed algorithm(4)A function module is developed to monitor benzene concentration of blended gasoline based on an online Raman detection system developed by our laboratory and the IPS-PLS quantitative analysis method proposed in the thesis.With comparison to the offline chromatographic analysis results,the root mean square error and coefficient of determination of the online detection is 0.01%and 0.99 respectively,proving high accuracy of the function module.The function module has been successfully applied to monitor benzene concentration of gasoline products and effectively meet the requirements of gasoline quality monitoring in refinery.