No description available.

A mid-infrared chalcogenide fiber optic probe was employed to develop a Fourier transform infrared spectroscopy-based partial-least-squares (PLS) calibration model for the quantitative analysis of sucrose, glucose, fructose, maltose, total sugar and water content in chocolate syrup. Based on the comparison of the pure component and correlation spectra extracted from chocolate syrup and aqueous sugar solutions based models, it was determined that the tightness of the concentration ranges and the ratios of the sugars in the chocolate syrup samples did not allow to draw adequate information to build a robust PLS calibration model. PLS regression models developed using infrared spectra of chocolate syrup calibration standards prepared by addition of sugar solutions to increase the concentration range did not yield conclusive results. A different approach used for standard preparation consisted of diluting chocolate syrup samples to different degrees. This new method provided an increased concentration range for the sugars but maintained an almost constant sugar to sugar ratios. The PLS models based on these new calibration standards yielded high calibration correlation coefficients and low errors on the external validation. Accuracy, repeatability, long-term stability and ruggedness were tested and the results demonstrated that the calibration models were robust and had a better repeatability than the reference high-performance liquid chromatography method. The fact that the calibration model was developed using standards having very similar sugar profiles precluded its use for the analysis of chocolate syrup samples of different formulations. The resulting formulation-specific PLS regression model required a preclassification step to ensure that the model is applied to the appropriate sample type. A probabilistic neural network (PNN) model was developed to fulfill the preclassification requirement. PNN yielded excellent classification results. The modeling uncovered