Maddox AI automates 100% inline inspection of piston housings and reliably detects defects such as chips, dents, cracks, porosity, missing material, scratches, contamination, and draw marks across multiple views.
Product
Industry
Total savings / year
Our customer manufactures piston housings for the automotive industry. One product variant had to be inspected for a total of eight defect classes, including chips, dents or pressure marks, cracks, shrinkage cavities or porosity, missing material or raw spots, scratches, contamination, and draw marks.
The quality inspection system needed to achieve at least 97% accuracy while ensuring that no defective parts passed through undetected. In addition, false rejects had to be limited to a maximum of 3%, and the cycle time of 10 seconds per part had to be maintained reliably.
A particular challenge was that multiple surfaces of each part had to be evaluated: the outer surface, the inner surface, and the top view. For implementation, Maddox AI was integrated into the existing inspection cells as a retrofit, allowing the customer to continue using the installed camera hardware.
The rule-based system was not reliable enough at detecting defective parts. This created a risk of defect escape and increased manual inspection effort.
With Maddox AI, the evaluation logic was replaced by an AI-based inspection system. Since implementation, production has been running stably with zero defect escapes.
Increasing the system’s sensitivity reduced defect escape, but it also raised the false reject rate. As a result, the workload for manual reinspection increased.
Maddox AI reduces false rejects without compromising inspection reliability.
When surface conditions changed or new defect patterns appeared, the legacy system had to be manually re-parameterized on a regular basis. This created significant effort during series production.
New NOK images can be annotated and used for retraining in self-service. Updated models can then be deployed directly to the existing industrial PCs.
The rule-based system was not reliable enough at detecting defective parts. This created a risk of defect escape and increased manual inspection effort.
Increasing the system’s sensitivity reduced defect escape, but it also raised the false reject rate. As a result, the workload for manual reinspection increased.
When surface conditions changed or new defect patterns appeared, the legacy system had to be manually re-parameterized on a regular basis. This created significant effort during series production.
With Maddox AI, the evaluation logic was replaced by an AI-based inspection system. Since implementation, production has been running stably with zero defect escapes.
Maddox AI reduces false rejects without compromising inspection reliability.
New NOK images can be annotated and used for retraining in self-service. Updated models can then be deployed directly to the existing industrial PCs.
Defect escape in production
False reject rate
Cycle time per part
Would you like to learn more about our solution? – Test for free now!
We are ISO 27001 certified.
