In automatic milking systems, sensors are used to find sick cows and prevent bad milk to end up in the bulk tank. The objective this project was to predict somatic cell counts and clots in milk, two important and well established indicators of udder health and mastitis. This was done using regularly generated sensor data form the milking robot as input to machine learning models. It is not the first time data from sensors in automatic milking systems are combined to predict sick cows or milk of bad quality. However, few earlier studies attempt to predict somatic cells and no previous studies aimed to predict clots in milk. Additionally, there is a lack of knowledge regarding occurrence of clots at quarter level among cows milked in automatic milking systems.
The models could predicted somatic cell count with a low prediction error and the prediction improved further if the cows earlier level of somatic cells was known. Using system data from 3 days was sufficient for the prediction, meaning that the cow does not have to be milked in the system during a longer period in order to predict the somatic cell count. By using models based on data generated on daily basis, applications could be created where the somatic cell count could be displayed by the system without sampling the cows and as often as the user prefers. Predicted values of somatic cell count could also be included in mastitis prediction models to improve the prediction performance.
The occurrence of clots at quarter level among cows milked in automatic milking systems was investigated for the first time. Among many interesting relationships, we discovered that longer milking intervals increased the risk of having clots in the milk as well as occurrence of clots in a quarter at a previous milking. Cow milkings free from clots were correctly predicted with a high certainty by our models, while the occurrence of clots was harder to predict, especially the milder cases of clots. In the future we hope to combine the knowledge of this research to create applications were the user easily could obtain predicted somatic cell counts for individual cows, hopefully also combined with the occurrence of clots in the milk.
This PhD project is a collaboration between DeLaval International and the Swedish University of Agricultural Sciences. Parts of the funding was provided by the Swedish Foundation for Strategic Research.
Dorota Anglart obtained her Master of Science degree in Animal Sciences in 2010 at the Swedish University of Agricultural Sciences, Uppsala. After graduation she joined the team at DeLaval International AB in Tumba, Sweden, working as a test leader for automatic milking systems. In 2016, she took on a new role as the first industrial PhD at DeLaval working on a project for improved mastitis in automatic milking systems. Her recurrent role at DeLaval is mastitis management specialist. In her spare time she likes to train her dogs and take them for long walks.
Text and picture: Dorota Anglart
