Development of a PLS based method for determination of the quality of beers by use of NIR: spectral ranges and sample-introduction considerations

Abstract

Near infrared spectroscopy (NIR) has been used to determine important indicators of the quality of beers, for example original and real extract and alcohol content, using a partial least squares (PLS) calibration approach. A population of 43 samples, obtained commercially in Spain and including different types of beer, was used. Cluster hierarchical analysis was used to select calibration and validation data sets. Absorbance sample spectra, in transmission mode, were obtained in triplicate by using a 1-mm pathlength quartz flow cell and glass chromatography vials of 6.5 mm internal diameter. The two methods of sample introduction were compared critically, on the basis of spectral reproducibility for triplicate measurements and after careful selection of the best spectral pre-processing and the spectral range for building the PLS model, to obtain the best predictive capability. For each mode of sample introduction two calibration sets were assayed, one based on the use of 15 samples and a second extended based on use of 30 samples, thus leaving 28 and 13 samples, respectively, for validation. The best results were obtained for 1 mm flow cell measurements. For this method original zero-order spectra data in the ranges 2220–2221 and 2250–2350 nm were chosen. For the real extract, original extract, and alcohol d_x-y and s_x-y values of −0.04 and 0.07% w/w, −0.01 and 0.13% w/w, and −0.01 and 0.1% v/v, respectively, were obtained. The maximum errors in the prediction of any of these three indicators for a new sample were 2.2, 1.2, and 1.9%, respectively. This method compares favorably with the automatic reference method in terms of speed, reagent consumed, and waste generated.