Prediction of fat globule particle size in homogenized milk using Fourier transform mid-infrared spectra.

Our objective was to develop partial least square models using data from Fourier transform mid-infrared (MIR) spectra to predict the particle size distributions d(0.5) and d(0.9), surface volume mean diameter D[3,2], and volume moment mean diameter D[4,3] of milk fat globules and validate the models. The goal of the study was to produce a method built into the MIR milk analyzer that could be used to warn the instrument operator that the homogenizer is near failure and needs to be replaced to ensure quality of results. Five homogenizers with different homogenization efficiency were used to homogenize pasteurized modified unhomogenized milks and farm raw bulk milks. Homogenized milks were collected from the homogenizer outlet and then run through an MIR milk analyzer without an in-line homogenizer to collect a MIR spectrum. A separate portion of each homogenized milk was analyzed with a laser light-scattering particle size analyzer to obtain reference values. The study was replicated 3 times with 3 independent sets of modified milks and bulk tank farm milks. Validation of the models was done with a set of 34 milks that were not used in the model development. Partial least square regression models were developed and validated for predicting the following milk fat globule particle size distribution parameters from MIR spectra: d(0.5) and d(0.9), surface volume mean diameter D[3,2], and volume moment mean diameter D[4,3]. The basis for the ability to model particle size distribution of milk fat emulsions was hypothesized to be the result of the partial least square modeling detecting absorbance shifts in MIR spectra of milk fat due to the Christiansen effect. The independent sample validation of particle size prediction methods found more variation in d(0.9) and D[4,3] predictions than the d(0.5) and D[3,2] predictions relative to laser light-scattering reference values, and this may be due to variation in particle size among different pump strokes. The accuracy of the d(0.9) prediction for routine quality assurance, to determine if a homogenizer within an MIR milk analyzer was near the failure level [i.e., d(0.9) >1.7µm] and needed to be replaced, is fit-for-purpose. The daily average particle size performance [i.e., d(0.9)] of a homogenizer based on the mean for the day could be used for monitoring homogenizer performance.