Mathematical Representation Of Raman Spectra And Applications In Quantitative Analysis

Raman spectroscopy has been revealed as an important technique in nondestructive testing field. This fact is associated with its main characteristics, such as analytical sensitive, high analytical speed and minimum sample preparation. However, there are still some difficulties in practical application of quantitative analysis based on Raman spectra, such as overlapping band of different components in a mixture, nonlinear mixture effects of spectra, shorting of training samples and so on. To overcome these problems, a series of Raman spectra quantitative analysis methods have been presented in this thesis, which were proposed based on the theory of spectral analytic model. The thesis specifically including:1) The primary task of spectral hard modeling method is to represent the measured spectrum as a sum of Voigt peaks. The precision of the spectral model has immediate impact on the accuracy of the regression model. A spectrum often includes dozens or even hundreds of Voigt peaks, which mean that the spectral modeling process is an optimization problem with high dimensionality in fact. So, the computational process is time consuming and the results may not be the optimal solution due to ill-condition of the optimization problem. Herein, a rapid spectral modeling method was proposed, which reduces the dimensionality of optimization problem by only adjusting the parameters of overlapping peaks in stead of all peaks in the process of spectrum modeling. Experimental results showed that the spectral model built by the new method is more accuracy and need much shorter running time than the conventional method.2) A new characteristic peaks extracting and quantitative analysis method (Direct hard modeling, DHM) based on Voigt function was proposed to overcome the problem of strong spectral overlap in mixture. By DHM algorithm, the spectra of mixtures were modeled as a sum of Voigt functions at first; and then the characteristic peaks of every pure component in mixture can be extracted by analyzing the linear correlation between the intensity of Voigt peaks and the concentration of components; a linear calibration model will be built based on the intensity of characteristic peaks and the concentration of pure-components at last. Comparative studies were taken on to quantitatively analysis the incoming stock components of paraxylene absorption tower. Experimental results showed that the DHM algorithm has the advantages of generalization and requiring less training samples compared to partial least squares algorithm.3) The quantitative analysis based on Raman spectra will become very difficult if there are some unknown components in mixture and the spectral profile of tested component present nonlinear changes. To overcome these problems, a new quantitative analysis method based on spectral model was proposed. At first, the spectral profile of tested component is modeled mathematically, and then the nonlinear effect on spectrum of tested component can be handled by adjusting the parameters of its spectral model. If there are some unknown components in mixture, the spectrum produced by unknown components can be fitted by a sum of Voigt peaks during an iterative algorithm, and then the fitting spectral profile can be deducted to eliminate the undesired signal come from unknown components. This method was applied to determine the methanol concentration in methanol gasoline and the analysis results were satisfied no matter the type of base gasoline is known or unknown. Compared with partial least squares and least squares support vector machine algorithms, the new method has the advantages of accuracy of prediction and requiring less training samples.