A 100% inspection system sees every defect. It does not tell you what to do with the roll. Here is where each One Box product sits, and what changes when the roll record leaves the inspection PC.
The vision model answers from the image alone: class, size, confidence. It stays a replaceable component, which keeps retraining cheap and avoids the classifier learning shortcuts from context it should not see.
Roll Grader answers from the stream of classified events plus machine-direction position, cross-direction position, time and the plant's event feed. Every diagnosis is explainable to a process engineer: a repeat every 412 mm names the roll with that circumference.
| Camera-only inspection system | Detects and classifies defects on one line. Data lives on that line's PC. The operator marks or tags, and someone downstream looks for the tag. |
|---|---|
| ImageFlow | The same job, done well, on any converting process. Pipeline tools, recipes without a programmer, operator and engineering interfaces, Excel reports per run. |
| Roll Tagger | Adds what the camera cannot know: roll identity, encoder truth, operator events and plant I/O. The defect gets a true position on a named roll, and the slitter or unwind can stop on it. |
| Roll Grader | Adds what no single line can do: review and grade every roll from every line against a policy, plan the slit, archive with genealogy, and learn from every reviewer decision. |
Borderline material is graded and located before it ships, with a record to show the customer.
The slitter stops on the defects that matter and runs through the rest.
Quality by product, line and shift, not by inspection PC.
Control stays on the line. If the hub is down, nothing stops; grading waits.