Developing simulation models of Automatic and Rotary Milking Systems to enhance production efficiency : Case study at DeLaval International AB

Sammanfattning

Currently, many farmers are transitioning to Automated Milking Systems (AMS), recognizing the ability to reduce labor demands and improve animal welfare. However, this shift is also driven by external economic pressures, such as labor shortages and the need for cost-efficient milk production. By developing a simulation model of the milking system, farmers and engineers can evaluate different barn layouts, cow traffic patterns, and resource allocations without disrupting live operations. Recognizing the potential of simulation, DeLaval has shown interest in using it to improve barn layouts and milking processes. However, there is a notable gap in existing models; few combine agent-based modeling with discrete event simulation, a hybrid approach well-suited to capturing both cow behavior and system dynamics. In collaboration with DeLaval, this master’s thesis investigated the modeling of two milking systems: AMS, using a batch milking cow traffic design, and Rotary Milking Parlors (RMP). The objective was to create simulation models that closely emulate real farm conditions to explore improvements in production efficiency. It focused on developing realistic agent-based simulation models grounded in actual farm layouts, cow behavior, and empirical input data using AnyLogic as a simulation tool. The study outlined a systematic approach for model development, demonstrating how the simulation tool, together with real-world data, can be used to replicate a farm’s milking systems. To assess the accuracy of the models, validation was performed using both statistical and face validation methods, comparing simulation outputs to data from the real farms. The results demonstrated the potential of a hybrid modeling approach for evaluating and optimizing farm designs. The models were successfully validated with an average relative error of less than 2% and shown to reflect the real-world behavior of the system. However, certain simplifications, such as the use of circular hitboxes, limited input and output data, and the omission of behavioral traits like dominance and stress, introduced some limitations. Nevertheless, the models identified key decision parameters and their influence on productivity, providing a foundation for further refinement of models with more complex logic and behavior for more realistic replication, allowing for optimization opportunities.