Impact of mid-FTIR homogenizer performance on repeatability and predicted values for major milk components

Abstract Text: Our objective was to determine the impact of  Fourier transform mid-infrared (mid-FTIR) in-line homogenizer efficiency on repeatability and accuracy of mid-FTIR predicted fat, true protein, and anhydrous lactose determination given by traditional filter and partial least squares (PLS) prediction models. Prior to the experiment, the slopes and intercepts for fat, true protein and anhydrous lactose were adjusted by running modified milk calibration samples through a mid-FTIR milk analyzer with an in-line homogenizer that produced homogenized milk with about 1.4 μm for a d(0.9). Five two-stage in-line homogenizers with different homogenization efficiency (i.e., produced d(0.9) from 1.35 to 3 μm on unhomogenized milk) were used to homogenize unpreserved, pasteurized, externally homogenized whole milk and unhomogenized whole milk. Homogenizers selection was based on performance determined in advance using laser light scattering particle size analysis. Milk that was externally homogenized and unhomogenized milk were each tested 18 times in sequence producing predicted values for fat B, fat A, fat PLS, true protein, true protein PLS, anhydrous lactose, and anhydrous lactose PLS. Repeatability and accuracy of fat, true protein, and anhydrous lactose determination using traditional filter models and PLS models were determined. Component predictions on externally homogenized milks (d(0.9)  = about 1.0 μm) had excellent repeatability and accuracy on all components when pumped through both efficient and inefficient homogenizers. Variation in homogenizer performance on unhomogenized milks had a much larger impact on accuracy of component testing than on repeatability. The largest absolute impact was on fat measurements, producing a lower fat B test by about 0.16% at a d(0.9) of 3 μm. The fat A and fat PLS predictions were low by 0.08 and 0.07% fat, respectively, at a d(0.9) of 3 μm. Variation in homogenization had very little impact on true protein measured using the traditional filter approach but did have a larger impact on true protein predicted with a PLS model (low by 0.1% at a d(0.9) of 3 μm). Effects of variation in homogenization on anhydrous lactose predictions were small. The USDA Federal Milk Market laboratories use a d(0.9) value of 1.7 μm as a criteria to make the decision to replace the homogenizer in an IR milk analyzer. In our study, a d(0.9) of 1.8 μm  produced a change in reading of < 0.04% fat and < 0.02% true protein. Both traditional filter and PLS mid-FTIR component prediction models were influenced by homogenizer performance.

Keywords: component accuracy, homogenization, mid-FTIR