Application of an Experimental Design to the Detection of Foreign Substances in Milk

Abstract

NIRA was used in order to study interactions between temperature (T) and soluble substances and to detect solutes added to milk. A two-level factorial design was planned to estimate the effects of the main factors and of the most interesting interactions on spectral data. Water ( W), NaCl ( N), reconstituted skim milk powder ( P) and W with N were alternately added to milk. The plan was repeated on eight different milks. The interactions studied were W × N, W × T, N × T, P × T, W × N × T. The values assigned to the experimental design model were 1 and —1 for active and non-active factors, respectively. This model and the corresponding analysis of variance were assessed independently at each wavelength, making it possible to plot results versus wavelengths as a spectrum. Thus, absorbance (predicted variable) was considered as the response to experimental design variations (predictors). Principal Component Analysis (PCA) was applied to the overall data collection to identify the main variation factors. The ability of principal components to predict the presence of foreign substances was assessed by Principal Component Regression. Several PCRs were calculated on two sub-collections (HIGH and LOW temperature), obtained from the overall collection (TOTAL), in order to assess whether recording spectra at two temperatures was useful to predict the presence of foreign substances. PCA on the overall data collection identified T as the most important factor affecting the total spectra variation; N and P groups were also separated according to the third and the fourth principal components. The parameters of the analysis of variance confirmed that the main factors N, P and T and the T interactions were more significant. T and N induced a shift of water absorption bands around 1450 and 1940 nm. There was no prediction improvement in recording milk spectra at two temperatures. PCA on the LOW sub-collection identified W and P according to the first factor and N according to the second one.