Authors: D Krause C Holtz M Gastl M A Hussein T Becker
Publish Date: 2014/12/07
Volume: 240, Issue: 4, Pages: 831-846
Abstract
This work is focused on a new strategy for quality analysis of brewing malt using near infrared NIR spectra taken from malt kernels in reflection as fingerprint to classify directly to processability of malt One part of the study deals with calibrating a partial least squares discriminant analysis PLSDA model with NIR spectra classifying malt into the three different classes resulting in a fivecomponent model Therefore suitable preprocessing algorithms for spectra were tested The target for calibration is given by an expert opinion on lautering runs filtration step in brewing The accuracy achieved using pilot plant data in relation to the expert classification “good” “normal” and “bad” was 906 and 927 in validation and calibration respectively The second part of the study is presenting the transfer of these analytical tools to industrial scale This was established via adjustment to corresponding system conditions The accuracy achieved using similar algorithms as mentioned before was 936 and 766 in calibration and validation respectively Independent from this two numerical possibilities were established for automatic process evaluation classifying the different processes in three categories good normal bad the first is calculating the residual standard deviation of a process based on multivariate statistical process control and the second is discretizing each process individually based on its single online trends Both methods were compared to the expert opinion coinciding with 84 and 85 respectively
Keywords: