Spectra selection methods:A novel optimization way for treating dynamic spectra and in-line near infrared modeling

Near infrared(NIR)spectroscopy is now widely used influidized bed granulation.However,there are still some demerits that should be overcome in practice.Valid spectra selection during modeling process is now a hard nut to crack.In this study,a novel NIR sensor and a cosine distance method were introduced to solve this problem in order to make thefluidized process into"visualization".A NIR sensor wasfixed on the side of the expansion chamber to acquire the NIR spectra.Then valid spectra were selected based on a cosine distance method to reduce the influence of dynamic disturbances.Finally,spectral pretreatment and wavelength selection methods were investigated to establish partial least squares(PLS)models to monitor the mois-ture content.The results showed that the root mean square error of prediction(RMSEP)was 0.124%for moisture content model,which was much lower than that without valid spectra selection treatment.All results demonstrated that with the help of valid spectra selection treatment,NIR sensor could be used for real-time determination of critical quality attributes(CQAs)more accurately.It makes the manufacturing easier to understand than the process parameter control.

Application of near infrared spectroscopy in monitoring the moisture content in freeze-drying process of human coagulation factor VIII

As an important process analysis tool,near infrared spectroscopy(NIRS)has been widely used in process monitoring.In the present work,the feasibility of NIRS for monitoring the moisture content of human coagulation factor VIII(FVIII)in freeze-drying process was investigated.A partial least squares regression(PLS-R)model for moisture content determination was built with 88 samples.Different pre-processing methods were explored,and the best method found was standard normal variate(SNV)transformation combined with 1st derivation with Savitzky–Golay(SG)15 point smoothing.Then,four different variable selection methods,including uninformative variable elimination(UVE),interval partial least squares regression(iPLS),competitive adaptive reweighted sampling(CARS)and manual method,were compared for eliminating irrelevant variables,and iPLS was chosen as the best variable selection method.The correlation coe±cient(R),correlation coe±cient of calibration set(Rcal),correlation coefficient of validation set(Rval),root mean square errors of cross-validation(RMSECV)and root mean square errors of prediction(RMSEP)of PLS model were 0.9284,0.9463,0.8890,0.4986% and 0.4514%,respectively.The results showed that the model for moisture content determination has a wide range,good linearity,accuracy and precision.The developed approach was demonstrated to be a potential for monitoring the moisture content of FVIII in freeze-drying process.