At a glance
- Client
- An automotive components manufacturerName withheld for confidentiality
- Industry
- Manufacturing
- Service
- Artificial Intelligence
- Scope
- Camera setup, defect detection models and line integration
The challenge
Quality checks relied on inspectors examining a sample of parts by eye. At full line speed, small surface defects and dimensional faults were easy to miss, and inspector fatigue made results inconsistent between shifts.
Defects that escaped the plant were found by the customer, leading to returns, line stoppages at the customer's site and the risk of a costly recall.
Our approach
Define the defects
With the quality team, we built a catalogue of defect types with example images and acceptance limits.
Install cameras and lighting
Industrial cameras and controlled lighting were mounted at key stations so every part is imaged the same way.
Train detection models
Vision models were trained to spot each defect type and tuned to run at line speed on edge hardware.
Connect to the line
Failed parts are rejected automatically and every result is stored for traceability.
The solution
Every part is now photographed and inspected as it moves down the line. Parts with a defect are diverted automatically, and inspectors focus on borderline cases the system flags for review.
Each inspection result is linked to the batch and shift, so the quality team can trace problems back to a machine or a material lot.
Results
- Full inspection instead of sampling
- Consistent results across shifts
- Defects traced to a machine, batch or shift
- Inspectors freed to work on root causes rather than repetitive checks
Technology used
- Python
- OpenCV
- PyTorch
- Edge GPUs
- PLC integration
Facing a similar challenge?
Tell us what you are working on. We will walk you through how we would approach it, with no obligation.