Frequently recorded sensor data may correctly provide health status of cows if data are handled carefully and errors are filtered away

Abstract

Description of the subject. The implementation of sensor based decision support in commercial dairy herds is highly dependent on having reliable systems. Problems with sensors give missing and noisy data hampering their use. Also, the presentation of results needs to be in a form which is simple and useful. These issues were addressed using a mastitis sensor and decision support as example. Objectives. This study aims at providing and evaluating a modular system applicable to the pipeline from sensor to decision support. Method. The case of mastitis was chosen as it is of economic importance and also affects welfare of cows, and because we have worked with a commercial sensor. The problems with sensors causing missing data and noise are described and a range of filtering and monitoring modules are shown to be important to make systems functional for herd management purposes. On top of this a solid method needs to be used to interpret and present data to end users, in terms of easy to read categories. Results. Filtering and pre-adjustments of raw data are important in making algorithms robust and reliable for daily use. Re-definition of traits is needed going from traditional few groups to continuous definitions, and then to new action oriented health classes. Also, for this case focusing on mastitis, assignment to "permanently sick" groups can be helpful in keeping focus on new acute cases. Conclusions. The combined use of filtering, fix-up routines and time series models leading into action oriented categories is needed to provide simple and robust decision support. The systems may be vastly improved by opening for transmission of data between user groups and to common databases – also with a few to use data in genetic selection.

Keywords

• Mastitis
• dairy cows
• sensors
• mathematical models
• data management