Study On In-Process Determination Of Contents And Blend Uniformity Of Primary Components In Modified Double Base Propellant Intermediate

A fiber-optic probe was inserted into proprllant intermediate samples stirred constantly to record spectra. Massive water and continued agitation are the physical condition of proprllant intermediate, their impactions on NIR spectra were studied. A RDX calibration model was constructed based on partial least squares (PLS) regression algorithm. The model was optimized by selecting right spectrum regions and pretreantment methods. A rubust model was builted based on the value of Rcv2, RMSECV and RPD were 0.9970,0.146 and 18.2574. The developed model was assessed by extermal validation with samples of prediction set. The ratios of root mean square error of predicted divided by the mean content value of prediction set (RMSEP/mean) was 0.47%; the ratios of RMSEP divided root mean square error of calibrated (RMSEP/RMSEC) was 1.156. The result showed that the performance of quantitative model for RDX was accuracy.precise and rubust. The model can afford on-line determination of RDX content for quality control purpose.A fiber-optic probe was inserted into proprllant intermediate samples stirred constantly to record spectra. Two quantitative calibration models were constructed based on partial least squares (PLS) regression algorithm for RDXand NC. The two models were optimized by selecting right spectrum regions and pretreantment methods. The developed models were assessed by external validation with samples of prediction set. For RDX and NG, the ratios of root mean square error of predicted divided by the mean content value of prediction set (RMSEP/mean) are 1.79% and 2.63% respectively; the ratios of RMSEP divided root mean square error of calibrated (RMSEP/RMSEC) were 1.0375·and 1.0117 respectively; the residual predictive deviation (RPD) statistic that is defined as the standard deviation (S.D.) of the reference values divided by RMSECV (S.D./RMSCV) were 8.3624,7.9057 and 7.9957 respectively. The result showed that the performance of quantitative model for RDX and NC were accuracy.precise and rubust. The model can afford on-line determination of RDX and NC content for quality control purpose.Near infrared spectrum technolygy combined with Moving block of standard deviation (MBSD) algorithm was applied to monitor the homogeneity of propellant intermediate and the mixed end was judged. When the blednd fininshed, a fiber-optic probe was inserted into proprllant intermediate samples stirred constantly to record spectra. Three quantitative calibration models were constructed based on partial least squares (PLS) regression algorithm for RDX, NC and NG. The three models were optimized by selecting right spectrum regions and pretreantment methods. The developed models were assessed by external validation with samples of prediction set. For RDX, NC and NG, the ratios of root mean square error of predicted divided by the mean content value of prediction set (RMSEP/mean) were 2.73%, 2.53% and 2.56% respectively; the ratios of RMSEP divided root mean square error of calibrated (RMSEP/RMSEC) were 1.2207,1.0476 and 1.07 respectively; the residual predictive deviation (RPD) statistic that is defined as the standard deviation (S.D.) of the reference values divided by RMSECV (S.D./RMSCV) were 9.449and 6.377 respectively. The result showed that the performance of quantitative model for RDX and NC were accuracy.precise and rubust. The model can afford on-line determination of RDX and NC content for quality control purpose.