Establishing a novel procedure to detect deviations from standard milk processing by using online Raman spectroscopy

Abstract

Controlling milk processing steps is a crucial task as it affects the quality and safety of the final product. Using Raman spectrometer in combination with various evaluation techniques such as principal component analysis and regression, Gaussian process regression, and the autoencoder were checked to define an accurate method for detection of deviations from standard procedures. For this purpose, milk with 5% fat measured at 10 °C was considered as the reference milk. A temperature-controlled flow cell was used in a by-pass for online measurements. While the principal component regression was not able to predict the deviations, results demonstrate the capability of Gaussian process regression and the autoencoder to detect 5% added water and cleaning solution, 0.1% difference in fat content and variation of 5 °C in measurement temperature. It can be concluded that both procedures display promising results, however, the autoencoder can be trained once and used immediately for online supervision. Therefore, changes can be detected promptly, enabling companies to react instantly. • Deviations from standard samples during the milk processing was detected. • Raman spectroscopy equipped with high-quality quartz flow cell was utilized. • To find a proper monitoring method, various evaluation techniques were checked. • A suitable method for an online application was developed.