Automated reading of codes on raw forged parts for maximum accuracy and reliability.

Traceability of forged industrial parts is a persistent challenge in the metal-mechanical industry. At Castertech, quality assurance required absolute precision in identifying batch codes. However, freshly forged parts present rugged characteristics—such as surface variation, inconsistent reflectivity, and natural distortions in embossed characters.
Relying heavily on manual inspection created bottlenecks in cycle time and increased the risk of traceability errors.

DB’s technical assessment concluded that standard sensors or barcode reading systems would not perform reliably in this harsh environment. A custom computer vision model was required, specifically trained to handle the visual “imperfections” of forged metal components.
The system needed to go beyond isolated character recognition, interpreting the full surface context of each part. In addition, the software architecture had to be lightweight and optimized for real-time interaction with existing industrial robots on the shop floor, supporting layout variability and production flexibility.

The team developed and deployed a robust AI-based image processing application. The model was trained on thousands of samples to achieve high resilience against the imperfections of embossed markings on forged parts.
The solution was successfully integrated into two robotic cells on Castertech’s factory floor. To ensure production flow was not impacted, DB performed deep performance optimizations, reducing image processing latency to fractions of a second and enabling reliable operation across six different traceability layouts.