Autenticação lipídica de manteiga usando espectroscopia por infravermelho próximo (NIRS)

Abstract

The authenticity of dairy products, especially butter, represents a challenge for the industry and regulatory agencies due to the occurrence of economically motivated fraud, such as the partial replacement of milk fat with lower-cost vegetable lipids, such as palm stearin. In addition, changes resulting from processing and storage may compromise product quality, reinforcing the need for rapid, reliable, and non-destructive analytical methods. In this context, this thesis aimed to develop a methodology based on near-infrared spectroscopy (NIRS) associated with chemometrics for butter authentication and the physicochemical changes during storage, based on the assumption that adulteration may interfere with the spectral signature of butter. The experimental design comprised the production of butters prepared from two milk creams, subjected to the addition of palm stearin at different concentrations (0, 5, 10, and 15%) and evaluated during 120 days of refrigerated storage. The samples were characterized by reference physicochemical analyses, including moisture, titratable acidity, peroxide value, fat content, and solids-not-fat, and by NIR spectroscopy in the range of 908 to 1676 nm. The spectral data were subjected to different preprocessing techniques and analyzed by exploratory chemometric methods, supervised classification using PCA-LDA, LDA, PLS-DA, SPA-LDA, GA-LDA, and one-class authentication using the models DD-SIMCA and OCPLS. The results demonstrated that the addition of palm stearin compromised butter stability throughout storage, favoring changes related to syneresis, increased acidity, and the progression of oxidative processes. Exploratory analysis showed that storage time constitutes a source of spectral variability, while the supervised models exhibited a high ability to discriminate authentic and adulterated samples. Among them, GA-LDA stood out by combining high performance with a significant reduction in spectral dimensionality, concentrating the discriminant information in regions associated with C–H vibrations of lipids. In the authentication assessment, the Linear-OCPLS model demonstrated superior performance, achieving 100% efficiency in rejecting samples with process and compositional deviations, in addition to presenting a sensitivity of 48.57% in identifying authentic butters, representing low recognition of the authentic class, which demonstrates a confirmatory application. On the other hand, although the DD-SIMCA method achieved 100% specificity for adulterated samples and samples with process deviations, it showed low sensitivity in recognizing genuine samples, resulting in a rejection rate of 97.5%. The results obtained demonstrate that NIR spectroscopy associated with chemometrics constitutes a rapid, non-destructive tool with the potential to reduce reagent consumption and waste generation for butter authentication and quality monitoring. In addition to expanding knowledge about the application of spectroscopic methods in food authentication, this thesis demonstrates the potential of variable selection for the development of simplified NIR sensors and highlights the applicability of one-class classification models in quality control and inspection programs, contributing to strengthening the integrity of the dairy product supply chain and combating food fraud.