Real-Time Slaughterhouse Monitoring
Detecting classification deviations in Norwegian pig production — a monitoring platform used by Animalia to identify anomalies in meat percentage measurements across Nortura and Fatland facilities.
At Norwegian slaughterhouses, pigs are classified based on estimated meat percentage. This classification directly affects the value of each animal and is used to determine payments throughout the value chain.
Meat percentage is estimated using specialized classification systems such as GP7 and AutoFOM, operated by trained classifiers. While these systems are highly accurate, deviations can occasionally occur due to instrument calibration issues, human error, or changes in production conditions.
Because thousands of pigs can be processed within a few days, even small systematic errors may have significant financial consequences for both farmers and slaughterhouses. Detecting such deviations quickly is therefore critical.
The Challenge
Not every deviation indicates a problem. Changes in meat percentage can also occur naturally due to factors such as:
- Seasonal feeding strategies
- Differences in pig genetics and breeding lines
- Variations in slaughter weight and age
- Changes in supplier composition
The challenge was therefore not only to detect unusual changes, but also to provide tools that help explain why they occur.
Solution
I developed a monitoring platform that automatically collects classification data from slaughterhouses across Norway and analyzes incoming measurements on a daily basis. The system compares recent observations against historical population data and applies statistical process monitoring techniques to identify abnormal changes in meat percentage.
The platform provides:
- Real-time monitoring of slaughterhouse performance
- Statistical deviation detection using z-scores
- Trend analysis over time
- Automatic alerts for unusually high or low meat percentages
- Interactive dashboards for detailed investigation
Explainable Analytics
To support root-cause analysis, I also developed a machine learning module that investigates which factors influence meat percentage. Using decision tree models and feature importance analysis, the system helps identify variables that may explain observed changes, such as supplier composition, weight distribution, classification patterns, or seasonal effects. While the model was not designed for highly accurate prediction, it provides valuable insights into which variables contribute most to observed deviations.
Impact
The system is actively used by Animalia's classification advisors and quality specialists to monitor slaughterhouses operated by Nortura and Fatland. By detecting anomalies early, the platform helps identify potential instrument errors, classification issues, and production changes before they affect large numbers of animals. Feedback from Animalia's classification experts has shown that the platform is particularly useful for identifying potential drivers behind changes in meat percentage and supporting further investigation.