Virtual Sensing for Cluster Vacuum Control

Sammanfattning: The purpose of this project was to develop a virtual sensing model in order to estimate the vacuum level in the milking cluster with data collected further along the milk tube. Measuring the cluster vacuum level is desirable for many applications, such as control and monitoring, but due to the cow's tendency to kick, chew and stomp on anything within reach, placing a sensor in or close to the cluster is both difficult and impractical. The project was performed at Delaval and their requirements mirrored these issues with a focus on the solution's ability to be implemented into a milking setup as well as numerical limitations on the accuracy of the estimate. This project employed a methodology of experiments where the main focus was to compare a general model to a model specifically calibrated for the chosen setup. These models were also compared on four different setups that were selected after an initial test round. The solution revolved around calibrating a vacuum loss model with a temporary sensor placed at the cluster and used two pressure sensors placed further along the milk tube for permanent use. The results showed that a high degree of accuracy can be achieved as long as a reliable flow estimation is available. The attempt at designing a Venturi tube to measure the flow gave mostly underwhelming results and is an area fit for future development. In conclusion, it was found that by calibrating a simplified two-phase vacuum model using polynomial regression, in combination with a calibrated pressure-based flow estimate, the cluster vacuum could be estimated with high to very high accuracy depending on the milking setup.