| In the drugs, the active components are not only the hardcore of them, but also the main constituents to decide drug action. The quality and content of active components affect directly the therapeutic of a drug. If the content of active components is low enough, it can't have the due drug action; and if the content is high enough, it may bring the minus effect to hurt body. So in the course of producing drugs, the quality and content of active components are the important index in the course of producing and selling drugs, and must be monitored strictly. Because the technique of Near Infrared Spectroscopy Analysis (NIRA) is convenient, fast, without pretreatment to samples, and applies to inline analysis, it has the recognition and application continually in the field of pharmaceutical analysis.The wavelength region of near infrared spectroscopy is 780～2526nm(12820～3959cm-1). In this region, there are absorption about frequency multiplication and sum of fundamental frequencies of vibration spectrum with hydrogen radical groups which are O-H,N-H,C-H,C-H,S-H primarily, so the spectrum band is wide, the overlap is quite severely, and the absorption signal is weak, the information is complicated to analyze. Although this spectrum region is discovered earlier, the value of analysis can't be regarded. In recent years, with the development of supercomputer and the software of chemometrics, especially, in depth research and extensive use of chemometrics, NIR becomes the spectral technique with the fastest development and most remarkably, so its applications are more and more abroad in every field. Only the range of application in the field of drug analysis is quite abroad, it applies to not only drug's multi different states, such as those preparations of raw material, integrate tablet, capsule and liquid, but also the analysis of different kinds of drugs, such as protein, Chinese herbal medicine, antibiotics. NIR applies to analyze and detect the purity of bulk drug, packing material, etc., and monitor production process. It will carry out monitoring with online and continuous analysis to production process through different fiber optic probes.This thesis will analyze two sets of drugs (One is dealt with vitamin C; another is dealt with acetaminopher and artificial cow-bezoar.) with NIR quantitative analysis. We collect the drugs' near infrared diffuse reflection spectrums with Fourier infrared spectrometer, adopt to partial least square (PLS) and artificial neural net (ANN) to process spectral information with mathematical technique, and create each corrected model.PLS is a new multiple statistical data analysis method. It can solve the problem of linearity correction and gain satisfied results to measure many mass parameters to actual samples. PLS can select voluntarily for associated spectrum data. The data can be entire band, and can also be part data. When PLS analyzes two sets of samples' NIR, we select each spectral range according to the different researched component. We will regress for researched component between concentration and absorption spectrum, get corrected model to modeling analysis to gain regressive mean square deviation (RMSEC) and correlation coefficient (R). Forecast for every samples with the model, gain the comparison diagram with actual value. The correlation coefficients of regression models about two experiments are 0.998 at least, the deviation about calculated value with actual value are all low that between ±1%. To artificial cow-bezoar, the deviation is between ±0.05%.Because PLS is a multiple linear data analysis method, the changes about actual course are almost non-linear. If we almost linearize them, dealt them as linear model, the precision of non-linear fitting will be low. So we should find a nonlinear modeling method to correct spectrum data with high precision. BP net in ANN is a nonlinear modeling method with strong ability of approximation of function, so we use it to conduct gamma correction to one set of sample's spectrum data in the experiment.
|
PDF Full Text Request