Conceptual equipment illustration.
OCR inspection converts visible characters into data. A production check must then decide whether that data is the value expected on the pack. A code can be clearly readable and still contain the wrong batch, date or product identifier.
For a machine vision inspection system, specify both the reading task and the business check that follows it.
Recognition and Verification Have Different Targets
Optical character recognition reads an unknown string within its supported character set. Optical character verification compares characters with an expected pattern or value. Cognex’s OCV overview describes checking expected characters in expected positions for tasks such as date and lot codes.
Define which values are fixed, recipe-driven or supplied by another system. A broad pattern that permits letters and numbers might confirm format while still allowing an incorrect batch.
The code-reading and traceability scope covers how the read result joins the product record. OCR of human-readable printing is also distinct from decoding or grading a barcode.
Build a Challenge Set for the Print
Include acceptable print variation and known failures. Trial low contrast, missing strokes, displaced printing and characters that can be confused, such as O/0 or I/1, where those characters occur in the actual code.
Keep the image-acquisition settings with the trial. A software result trained on well-lit flat samples may not establish the same reading performance on the moving curved package.
The label-inspection guide covers presentation, placement and other label defects. OCR is one task within that wider pack check.
Validate the Value as well as the Read
For a date code, check the permitted structure and a valid calendar value. For a lot code, compare with the active batch or approved source. Define how the inspection obtains that source and what it does when the source is unavailable.
A high-confidence read can still be a correctly read wrong value. Keep confidence/reading status separate from the expected-value result.
For pharmaceutical serialization, associate human-readable data with the relevant pack record and other required code checks. OCR alone does not complete the serialization workflow.
Follow the Pack to Rejection
Test a missing code, an unreadable code, a readable wrong code and the loss of the expected-value input. Confirm that each result remains attached to the correct moving pack and that the physical reject is verified.
Save the image and decision evidence according to the specified retention requirement. If only selected images are kept, name that selection so that later investigation knows what is available.
Contact Motionwell about the printed-code check to discuss reading, expected values and the reject path together.