Outcome summary
We built a vision station that recognises the variant of each part with AI and checks its dimensions with precision measurement, at production-line speed. The verdict goes straight to the line's sorting equipment.
Starting situation
The manufacturer makes many variants of the same product family, differing in head shape, marking and size. Telling them apart and checking dimensions by eye is slow, tiring and inconsistent at line speed. A mixed or out-of-tolerance batch that reaches a customer is expensive.
The system
- Industrial cameras with dedicated lighting look at each part from several viewpoints.
- An AI vision model recognises the variant of each part.
- A separate measurement layer checks the part's geometry. It uses classical precision metrology, not AI, so dimensions are reproducible and explainable.
- The verdict for each part goes to the line controller, which handles sorting and rejection.
- Everything runs on embedded GPU computers at the machine. No cloud is needed in production.
Implementation
Each station is driven by a configuration profile that connects cameras, AI models and measurement tools. Adding a product variant means retraining the model and updating the profile; the software itself does not change. The same profile can be rolled out to several identical stations.
Results and evidence
Variant recognition and dimensional checks are automatic and repeatable at line speed, replacing manual visual checks, and are integrated directly with the line's sorting equipment.
After launch
The system runs in production on several stations, with more being added. Models are refined as new variants are introduced.