- A food company experienced manufacturing issues with one particular product, where the product was unsellable by the time it reached the end of the production line. They needed to find the root cause of this issue in the manufacturing process to avoid scrap and waste.
- The plant which manufactures this product has several production lines with an array of machines performing various tasks. These machines generate “historian data” – measures such as temperature, pressure, and belt speed. The machines produced data at 1300 data collection points. We also incorporated external data, such as ambient temperature and humidity.
- Using machine learning for both discovery and impact estimation, we isolated the cause of the problem. For one SKU (recipe) in particular, a combination of the temperature set-points for one of the ovens, as well as high ambient humidity, were the primary drivers of the problem.
- A re-examination of temperature set-points for each recipe was recommended. Also, the tolerance for deviations from set-points was better understood.
- The reduction of scrap product for this single item alone was worth $3.1 million per year.