Application Studies On Rapid Analysis Of Composite Solid Propellant By NIR Spectroscopy

The determination methods of the components and impurities in three principalmaterials of composite solid propellants were investigated by means of near infrared(NIR) technique. The effect of chemometrics methods which were applied in spectraldata pretreatment, wavelength selecting, calibration model establishing, model transferand outlier detecting was also discussed.In partⅠ, A NIR method was investigated to determine the content of water in MAPO by standard addition method which includes various calibration models such as single wavelength regression, multiple linear regression, principal component regression, partial least square regression and artificial neural network. The influences of these different models on predictive results were compared. The experiment results indicate that a more accurate result can be obtained with the single wavelength calibration model. The recovery of the standard additions method was 95%～108% and the relative standard deviation (RSD) was 1.6%. There was no system error in the determination results between standard additions method and national military standard method.In partⅡ, The method to determine the contents of NG, BTTN and PEG in NEPE binder was studied by NIR, which used many methods including the correlation coefficient, salience variety analysis, interval PLS and genetic algorithm to select wavelength. The model transfer methods were studied as well. The experiment results show that spectral data shift transform can improve the correlation between absorbance and concentration of samples. The GA was better than both of the correlation coefficient method and the iPLS method for selecting wavelength, and the salience variety analysis was not suitable for this purpose. The model could be transfered between different apparatus with NEPE standard samples, but not with general standard samples when D/S algorithm was applied. The results of component content in NEPE binder obtained by NIR method was in accord with that by standard method. The difference between these two methods is less than 1%. The water contents and general heating process have no obvious effect on determination results.In partⅢ, The method to determine the contents of NG,TEGDN and PET in NEPEB binder was studies by NIR, which included outlier detection and establishment of robust calibration model. The outlier was detected by various methods such as the Mahalanobis distance, leverage, PC correlation plot, PC correlation three dimension plot and clusters dendrogram. The experiment results suggest that all of the outliers obtained by different detecting methods were not identical and the effect of these different methods on model was different. The Mahalanobis distance and the clusters dendrogram were more applicable for this system. The effect of detecting methods was different for each predicting object, which was best to improve predictive results of PET. It was less than 2% that the predictive residual error using NIR method to determine NEPEB binder component content after outliers eliminated.