Inspecting Components Of Pharmaceuticus By Near-infrared Spectroscopy Analysis

The paper focuses on the study of Near Infrared(NIR) Spectra on inspecting the main components of the Metronidazole and develops several methods for NIR inspecting compositions in the Pharmaceuticus.In this thesis, 48 samples selected from total 72 Metronidazole Samples were used for instrument calibration, while surplus 24 samples were used for validating the calibrations. In the process of treatment, the mathematical method of linear forward-step regression was adopted at NIR original absorption spectroscopy, the first derivative spectra and the second derivative spectra, and 3 calibration equations were developed from 48 calibration samples. the correlative coefficients, the average relative errors and the standard errors were regarded as 3 evaluating indexes for calibration equations. Partial Least Squares(PLS) regression was utilized on two types of spectra (original absorption spectra and second derivative spectra) to develop calibration models. Also BP artificial neural network(ANN) was used at original absorption spectra .From predicting outcome of surplus samples(24),we found the highest precision got from BP-ANN with 3.521% of the average relative error, 0.0301 of the standard error and 0.9825 of the correlative coefficiect. The study showed the calibrations developed from several mathematical methods were good, PLS was better and BP-ANN was best.The paper also was concerned with the error of NIRSA, the sources of the errors and the influences of the errors. Also the mathematical methods of the calibrations were discussed. From the study we can develop fast NIR detector for Pharmaceutical-analysis and give thinking for practical application.