Effect of homogenizer performance on accuracy and repeatability of mid-infrared predicted values for major milk components.

Our objective was to determine the effect of mid-infrared (MIR) homogenizer efficiency on accuracy and repeatability of Fourier transform MIR predicted fat, true protein, and anhydrous lactose determination given by traditional filter and partial least squares (PLS) prediction models. Five homogenizers with different homogenization performance based on laser light-scattering particle size analysis were used. Repeatability and accuracy were determined by conducting 17 sequential readings on milk homogenized externally to the instrument (i.e., control) and unhomogenized milk. Milk component predictions on externally homogenized milks were affected by variation in homogenizer performance, but the magnitude of effect were small (i.e., <0.025%) when milks were pumped through both efficient and inefficient homogenizers within a MIR milk analyzer. Variation in the in-line MIR homogenizer performance on unhomogenized milks had a much larger effect on accuracy of component testing than on repeatability. The increase of particle size distribution [d(0.9)] from 1.35 to 3.03μm (i.e., fat globule diameter above which 10% of the volume of fat is contained) due to poor homogenization affected fat tests the most; traditional filter based fat B (carbon hydrogen stretch; -0.165%), traditional filter-based fat A (carbonyl stretch; -0.074%), and fat PLS (-0.078%) at a d(0.9) of 3.03μm. Variation in homogenization efficiency also affected traditional filter-based true protein test (+0.012%), true protein PLS prediction (-0.107%), and traditional filter-based anhydrous lactose test (+0.027%) at a d(0.9) of 3.03μm. Effects of variation in homogenization on anhydrous lactose PLS predictions were small. The accuracy of both traditional filter models and PLS models were influenced by poor homogenization. The value of 1.7µm for a d(0.9) used by the USDA Federal Milk Market laboratories as a criterion to make the decision to replace the homogenizer in a MIR milk analyzer appears to be a reasonable limit, given the magnitude of effect on the accuracy of fat tests. In the future, as new PLS models are developed to measure other components in milk, the sensitivity of the accuracy of the predictions of these models to factors such as variation of homogenizer performance should be determined as part of the ruggedness testing during PLS model development.