To address the 6 clusters of problems outlined above, methodological developments must be made. These have been defined in 4 overall themes:
- Automatic learning and pattern recognition using sensor data. This theme involves development of methods for on-line learning of specific properties regarding important production and behavioral traits in growing pigs at animal, pen and batch level. Sensor data will include live weight, feed intake and water intake as well as activity level, measured in different ways (vision, infrared sensors, visits to feeding trough etc). The sensor data will be used for development of online classification and prediction models for monitoring and detection of welfare and production problems. The theme will be based on Cases 1-5 above.
- Methods for prioritization of alarms and intervention. Since time series from different data sources are monitored simultaneously there is a high risk of false alarms. Only some of the alarms will, therefore, require intervention. The decision to intervene will be based on the expected consequences for welfare and productivity combined with the strength and validity of the individual alarms. This calls for a solution based on a Bayesian approach as for instance graphical models. The theme will rely on alarms produced by methods developed in Cases 1-5 above.
- Optimization methods in a stochastic production environment. This theme will try to solve the important methodological challenges related to simultaneously take welfare and productivity into account (multi-criteria optimization) while still addressing risk and uncertainty. Furthermore, integration of decisions at animal, pen, section and herd level will be studied. The theme will be based on Cases 2-6 above.
- Evolutionary operations as a tool for improving welfare and productivity. Evolutionary operations (EVOP) and other techniques for process design, control and development have not been used systematically within livestock production. This is in contrast to the successful use for industrial production processes. The availability of online sensor recordings reduces the problems with the implementation of these principles, e.g. the data-recording is much easier. This theme will study how to implement these methods in order to reduce the occurrence of welfare problems such as tail-biting. The theme will be based on Cases 2-5 above.
Figure 1 illustrates the integration of the methodological themes and the problem cases. The PhD projects of the alliance will also be embedded into the four themes.