A New Wavelengths Selection Methods For Modeling Of Molecular Vibrational Spectra And Fundamental Applications

The quality of product is the life of enterprises. It is imperative that modern industries should monitor the quality real time, and then control strategies can be made by the acquisitions of feedback information. Molecular vibrational spectra analysis technology has many prominent advantages, including capacity of nondestructive, rapid speed, easy operation, real time analysis and so on, which can provide abundant physical structure information combine with quantum mechanics. Molecular vibrational spectra analysis technology provides a new method to control the quality of product and optimizes the closed-loop operations for industries. In terms of different vibration, molecular vibrational spectra have several types, such as: infrared, Raman,Near infrared (NIR) spectra and so on. The phenomenon of Raman effect is due to molecular polarizability change, and NIR is due to the vibration of the molecular bonds which absorb spectral, Raman and NIR spectra are widely used in industries. As a kind of indirect measurement method, molecular vibrational spectra analysis technology must be combined with chemometrics methods to establish multivariate calibration model which can predict the unknow sample properties quickly. It is necessary that specific selection method should be used to remove redundant and abnormal wavelengths, and the purpose of wavelengths selection is to simplify model and improve prediction accuracy. This thesis proposes a new wavelengths selection method to pick out specific wavelengths. The validity of the new wavelengths selection method is verified with online standard NIR data set. The thesis researches the application of Raman spectroscopy analysis technology in new energy and new materials analysis. Then new method is adopted to select key variables of Raman Spectra, which also establishes a simplified and good accuracy modelThe main contributions of this thesis are as follows:1. In view of some parts of molecular vibrational spectra in common, propose a new wavelengths selection method which combine variable importance in the projection index and Partial Least Square Regression coefficient to pick out specific wavelengths. The new method eliminates the influence of randomness and subjective factors based on statistical ideas. The;validity of the new method is verified with standard NIR data set. For corn-oil standard NIR data set, the new method adopts only 11% wavelengths to build model, while RMSECV (root mean square error of cross validation) decreases by 33% and RMSEP (root mean square error of prediction) decreases by 29.2% compared with full spectra model.2.This thesis acquires quick and accurate results of mixing ratio by Raman spectroscopy analysis technology. In this research, build a platform to detect mixing ratio of biodiesel blend fuel real time, which is collected by Raman spectra. Then use data pretreatment methods to optimize the Raman data sets, and obtain a reliable data set for subsequent processing. The new wavelengths selection method is adopted to select key variables. Finially 16.7% wavelengths are selected to build model, RMSECV decreases by 39.6%, RMSEP decreases by 26%, only 6 factors are choosed.3.In view of polyacrylamide's type confusion and adulteration, this thesis explores the feasibility of Raman spectroscopy analysis technology coupled with pattern recognition methods in the application of polyacrylamide type distinction and adulteration identification, providing a method for industries to analysis polyacrylamide quickly. The new wavelengths selection method also works well in polyacrylamide Raman spectra wavelengths selection.