Our customer produces filters in various designs in a 3-shift operation. Quality control was previously carried out using a rule-based camera system. However, due to the diversity of production, the existing Keyence camera system had to be recalibrated daily in order to limit the pseudo reject rate. Despite the regular recalibration of the camera system, the old camera control led to a pseudo reject rate of up to 80%, which in turn led to high scrap / re-sorting costs. With the introduction of Maddox AI, the inspection system only needs to be recalibrated / retrained in a few cases and the average pseudo reject rate has been reduced by over 90%. Defective filters are now reliably rejected, while OK filters are reliably classified as OK even with different production batches.