Development of novel methodologies and models for milk quality monitoring (PhD)

Optical sensors are widely used in the agro-food sector to measure and monitor product quality. At the Livestock Technology research group of KU Leuven, we develop innovative and customized near-infrared spectral sensors to measure the composition of raw milk whilst being extracted from the cow’s udder. The obtained information can be used in real-time to separate milk based on its properties, safeguarding the quality of the delivered milk, or to create milk batches with unique characteristics. Moreover, as the production of milk is a dominant factor in the metabolism of dairy cows, involving a very intense interaction with the blood circulation, the extracted milk contains valuable information on the health status of the cow. As a result, the analysis of the milk components for each individual cow can provide valuable information on the cow’s udder health and metabolic and nutritional status. This can support early detection of altered health and welfare and reduce the use of antibiotics by taking preventive actions.

As all process analytical technologies, our sensors and their measured spectral data are subject to drift and structural noise. These discrepancies can originate from small differences in the hardware of sensor replicates, wear and maintenance, environmental fluctuations or variations in the cow management. In the past years, novel chemometrics methodologies and machine learning algorithms have been developed to account for these effects or obtain models that are more robust against these sources of drift.

The main goal of this applied project is to implement, optimize and validate different strategies for robust calibration, calibration transfer and calibration maintenance so that the models to predict the milk quality traits from the measured spectral data can cope with changing environments and conditions. Furthermore, the robustness of these prediction models will be evaluated and improved with the vast amount of on-farm data and experience that we have available. Finally, the variation of the sensor measurements and the outcome of the models will be studied in relation to the milk quality and cow health and combined with advanced data-processing techniques to obtain robust monitoring and early-warning systems.