Tools
for Monitoring Growing Pigs
PhD Thesis by Thomas Nejsum Madsen
Summary
There is a growing need for computer based methods for herd management within pig production. The average herd size is increasing as well as the number of animals managed per person. Moreover, the possibilities of real-time monitoring at farm-level have improved during the last years concurrently with the development of modern computer technology. There is a variety of studies dealing with methods for herd management, but very few are concerned with methods for monitoring the production. Monitoring is basically a matter of recording relevant parameters at a time scale and evaluating data as soon as they are available for the purpose of giving the caretaker a warning immediately when there is a problem in the production.
This thesis deals with two different methods for monitoring weaners and finishers. In Chapter 4 a new principle for estimation of daily gain is introduced. In contrast to the traditional way of calculating average daily gain as 3-month averages (or sometimes one-month), the new method provides estimates every time there has been a delivery to the slaughterhouse (often one delivery each week). The estimation method results in a time series of growth rate estimates, and it is illustrated how the the Kalman filter can be used for data filtering of the series.
The estimation principle described above is an improvement compared to traditional methods, but it still suffers from the fact that the estimates only cover the animals that have left the production unit. Chapters 5 and 6 present a method that gives a more current picture of the pigs' well-being, by online-registration of their water consumption. This study has proved that growing pigs show a very stable diurnal drinking pattern and that changes in the well-being often affect this pattern. Chapter 5 describes how the pigs' drinking pattern, recorded as one-hour sums of water consumption, can be modeled by means of a state space model including cyclic components. Among other things it is discussed how to estimate variance components in the model.
In Chapter 6 a model for detection of changes in the pigs' drinking pattern is developed. The model uses one-hour ahead prediction errors from the model described in Chapter 5, in conjunction with a Cusum control chart, to provide warnings in case of changes in the drinking pattern. Examples illustrate how the warning system reacts in case of outbreak of diarrhoea. Experience from the farms that joined this study shows that in the case of outbreak of diarrhoea warnings are given 24-36 hours before there are any visual signs on the pigs.
The monitoring system based on water consumption has been implemented as a commercial software package (FarmWatch), which is available for the pig producers. The relations between the monitoring method, the user interface and communication of alarms are described in Chapter 7.
Full thesis
The thesis is available as a PDF file.
