(206e) Information Content Analysis of near Infrared Spectral Data for in-Line Monitoring of Batch Crystallisation

Despite being widely used in agriculture, food production, environmental monitoring and regarded as on-line chromatograph in petrochemical and biochemical industries, little work can be found on NIR application to process for particle formation through crystallization or precipitation. Unlike mid IR, NIR spectra contain information regarding the solid phase in addition to concentration in the liquid phase. The purpose of this proposed research is to develop NIR as an on-line technique for pharmaceutical crystallization process for simultaneous measurement of the properties of both the liquid and solid phase including concentration, polymorphs, size and shape. This work will demonstrate through a number of carefully designed experiments that the NIR spectra contain information of both solid and liquid phase, and therefore can be used for simultaneous measurement of these properties.

Experiments carried out on the two polymorphic type of L-Glutamic acid namely alpha and beta, at temperature between 20 and 80 Degree Celsius with six different particle size ranges. Reference techniques such as X-ray diffractometery and microscopic images used to identify the polymorphs, Malvern Mastersizer 2000 and cascade sieves used to measure particle size and Mettler AB54 weighing instrument used to measure sample weights.

Chemometrics techniques; principal component analysis (PCA) and independent component analysis (ICA) used to extract information from the NIR spectral data. Results on both chemometrics techniques provided information of polymorphs, particle size, solution and solid concentration. Having analysed these information, a calibration model have been built to measure mean particle size, solution and solid concentration by partial least square regression. The effect of a range of different NIR data pre-treatments on calibration and prediction precision were investigated. Over all, raw absorbance data were found to produce regression model with the best predictive ability.