In-line quality control of food powder blends using near infrared spectroscopy

Near-infrared (NIR) spectroscopy is employed as an in-line control system to check the quality product of food powder blends in real time while they are still inside the mixer and in the packaging area. Quality control was so far performed off-line: few samples were taken after the end of production and quality of the whole batch was assessed just analysing these small quantities. Results were obtained after products were already packaged, and in case of unconformities blends could not be reworked and they were wasted. This work aims to measure composition of powder mixtures in real time and to check every single bag produced, so extending product quality to the whole batch. Productivity will be also enhanced increasing the mixer throughput and reducing the waste.
Different food powder blends are analysed: bread and roll concentrates, confectionery mixes, baking and dough powders, and cake emulsifiers. Two non-contact NIR fiber-optic probes are installed in a conical screw mixer and in a double screw conveyor. When blends reach the fully homogeneity, the mixture in the vessel is scanned by a light in the near-infrared frequency. Spectra are soon analysed and powder blends with concentrations within the acceptancy range are discharged passing through a double screw conveyor where a second NIR probe is installed. Powders are scanned inside the conveyor just before to be packaged, so checking the quality for every single bag produced.
The Bruker MATRIX-F FT-NIR spectrometer installed in the vessel was calibrated following a two-step approach. In the first step, samples of food powder blends in different concentrations were made and then scanned by an off-line Bruker MPA FT-NIR spectrometer. After cutting spectra to remove noisy regions, a calibration model was built using the PLS algorithm and then cross validated. Outliers were removed using PCA, and minimum value for RMSECV was obtained by choosing the best combination of number of factors, pre-treatment and frequency ranges. In the second step, spectra of powder inside the mixer were collected using the in-line NIR spectrometer and composition of corresponding samples was measured with the MPA FT-NIR spectrometer previously calibrated. Using these spectra and the corresponding concentrations, Bruker MATRIX-F FT-NIR spectrometer was calibrated using the same approach described for MPA. Finally MATRIX-F model was used to measure concentration of powders inside the vessel and the screw conveyor.
High correlation was obtained for the organic components in the mixture, and minimal difference between predictions and true value was measured. As expected, RMSECV for MPA was smaller than for MATRIX-F: in the first case samples were made up and concentration values were already known, while in the second case concentration was measured, so involving a bigger uncertainty.
This work demonstrates that near infrared spectroscopy can massively improve the quality control in food industries. Concentration of powder mixtures can be measured in real time in a non-invasive, quicker and safer way as no staff is involved and no chemical reagent is needed. The quality of the entire production batch is assessed, so providing a greater confidence to customers.