Visual quality control in wheel production

Requirement

Visual inspection of surface defects could not keep pace with line speed, and undetected defective products reached the customer.

Our approach

A YOLO-based model integrated with Basler industrial cameras and trained on images collected in the field was optimised with TensorRT and deployed to an edge device. Results are sent to the PLC, and defective products are rejected automatically on the line.

Outcome

Defects are detected at line speed, and model accuracy improves over time through operator feedback.

Technologies used

• Basler
• YOLO
• TensorRT
• PLC entegrasyonu