Effect of Milling on Electrostatic Separation and Modeling Protein and Starch Content of Flour Fractions

Abstract

The objective of this research is to establish the effects of different milling techniques on the solvent-free electrostatic separation process for navy bean flour as well as to develop a model based on near infrared and fluorescence data to determine protein and starch content of the protein- and starch-enriched fractions using multivariate methods (i.e. partial least squares regression). Data fusion was used to combine the NIR and fluorescence spectra to try to achieve a model that had better predictability for protein and starch content. Protein content was measured using Kjeldahl digestion and starch content was measured using a dinitrosalicylic (DNS) acid array. The samples used in the NIR model are navy bean flour fractions from the electrostatic separation and the raw navy bean flour. There are 102 samples that are split in calibration (82 samples) and validation (20 samples) sets. The protein-enriched samples are collected from the electrostatic plate while the starch-enriched fractions are collected from the bottom of the electrostatic separator. The acquisition of reproducible infrared and fluorescence data from powder samples was successfully achieved. The pin milled navy bean flour had an average particle size almost three times smaller than the regular milled navy bean flour which could have contributed to the a high protein content (40.7%) of the protein-enriched fraction. The regular milled flour had a much higher protein extraction under optimum conditions but could only achieved a lesser protein content (32.5%) for the protein-enriched fraction. The regular milled navy bean flour also seemed to have particles disaggregate in the triboelectric charging process. Multivariate methods and pre-treatment techniques were compared for the NIR spectra of the navy bean flour fractions from electrostatic separation to measure the protein and starch content. The best method used Multiplicative Scatter Correction (MSC) pre-treatment with PLS regressions and had R2 values of prediction of 0.965 and 0.912 for protein and starch content, respectively. The N-way partial least squares (NPLS) regression was still a good model seeing as the R2 values of prediction for starch and protein content were 0.946 and 0.885, respectively. Two fluorophores were observed in navy bean flour: tryptophan and an unknown peak. It was observed that the starch model using the fluorescence dataset was highly correlated to the model’s predicted protein content (R2 of 0.978). The protein content model was better calibrated using the training set as well as providing a better prediction using the validation set for both NIR and fluorescence spectra. Data fusion was achieved by combining the NIR and unfolded fluorescence spectra of the navy bean flour fractions. The individual techniques had undergone pre-treatment separately and yielded the best model for determining protein content. Starch content was best determined using only the NIR spectra.