This study presents an optical sensor system utilizing multi-reflectance spectroscopy (MRS), specifically designed for in-line applications to enable the real-time determination of fat and protein content in milk products, simultaneously. This method employs multiple light wavelengths and various illumination-detection geometries. A field study was conducted in a milk mixing plant, where measurements were obtained from milk products with varying fat and protein concentrations, with a particular focus on recombined milk samples and a brief comparison to conventional milk products. In a first step the experimental data are compared with simulation data obtained from an analytical MRS formula. The fundamental spectroscopic characteristics, particularly the dependence of reflectance values on fat concentrations, as well as the relationship between wavelength and reflectance, remained consistent. However, some experimental bias was observed in the absolute values when comparing the analytical and experimental results. Secondly, to get reflectance models multi-linear regressions (MLR) were carried out based on the experimental and analytical data as well fat and protein content obtained from traditional wet chemical methods. To estimate the model accuracy the root mean square error (RMSE) has been used yielding around 0.1 wt% for fat and protein. A validation procedure using recombined milk results in approximately 0.1 wt% for fat and around 0.2 wt% for protein. Finally, it is shown that the process sample temperature has only a small influence on the reflectance. In contrast the homogenization pressure significantly influences the reflectance and should be considered to ensure accurate monitoring.

Applications for machine learning (ML) and deep learning (DL) are constantly growing and have already been adopted in the field of particle measurement technology. Even though analytical (ultra-)centrifugation (AC/AUC) is a widely used technique for characterizing dispersed particle systems, ML and DL have not yet been applied in this area. Data evaluation and interpretation in AC/AUC can be challenging and often requires expert knowledge. DL models can help, but their development is limited by a lack of annotated training data. One solution is to generate and use synthetic data instead. In the first part of this study, a model was trained to synthesize data from experiments using a combination of Variational Autoencoder (VAE) and Generative Adversarial Networks (GANs). The results appear highly realistic. Novice users could distinguish real from synthetic samples with only 63% accuracy. Then, a classifier was trained on experimental AC data to categorize real-world examples based on their underlying separation kinetics, testing different DL architectures. After initial training, the models were further fine-tuned with synthetic AC data. ResNet34 models achieved the best performance with 94% accuracy, comparable to an AC expert (91%), while inexperienced users reached only 53%. In the second part of our study, a regression model was trained for the analysis of sedimentation coefficients. Therefore, various generative models were developed and evaluated for synthesizing AUC data based on numerically simulated sedimentation boundaries. The best results were achieved by combining VAE and GAN architectures with embedded physical constraints. However, the generative networks have so far led to additional smearing of the profiles, resulting in a broadening of the sedimentation coefficient distribution and indicating that further refinement is necessary.

In milk-based formulations, fat globule size is a fundamental quality parameter. Accurate measurement of globule size is critical for ensuring consistent product performance, preferably through in-line monitoring. However, most conventional techniques are off-line and unsuitable for integration into manufacturing processes. The objective of this study was to demonstrate the feasibility of a recently developed optical sensor based on Multi-Reflectance Spectroscopy (MRS) for in-line estimation of the fat globule Sauter diameter. The MRS sensor acquires multidimensional reflectance data across multiple wavelengths and defined illumination–detection geometries. Industrial validation was performed in a milk standardization plant using low-fat (1.5 wt%) and whole-fat (3.5 wt%) milk. Fat globule sizes were systematically varied by stepwise adjustment of homogenization pressure. Reference globule size distributions were obtained via analytical centrifugation and expressed as the Sauter diameter. Analytical reflectances derived from the reference size distributions were compared with experimental spectra to assess globule size dependency of the optical responses. Principal component analysis revealed a clear separation between fat concentration levels, motivating the development of concentration-specific partial least squares regression models. Two PLS models were trained and validated using independent datasets. Across the combined concentration range and full Sauter diameter span (0.7–3.5 μ m), resultet in a test RMSE of 271 nm. For the separate 1.5 and 3.5 wt% fat lines, RMSE values of approximately 80 and 40 nm were obtained, respectively. These results demonstrate that the MRS-based optical sensor enables accurate, real-time in-line monitoring of fat globule size in complex dairy emulsions and represents a promising process-analytical tool for industrial process control.