Label and artwork
Presence, position, skew, wrong-label checks, print defects and packaging appearance.
Intelligent inspection, controlled physical outcome
AI vision inspection with automatic rejection.
Combine machine vision, deep-learning inspection and deterministic code checks with reliable product tracking, rejection and verification.
Direct answer
When should you use AI vision inspection?
Use AI when the difference between acceptable and defective product contains natural variation that is difficult to express with fixed measurement rules. Conventional machine vision remains effective for edges, dimensions, presence, position, colour and well-controlled print.
A robust system may combine both: deterministic tools for codes and geometry, AI classification or segmentation for variable appearance, and automatic rejection for confirmed failures.
Packaging applications
Date and batch code
OCR or OCV for presence, legibility and comparison with an expected production value.
Barcode verification
Read and compare 1D, 2D, QR and Data Matrix codes; treat no-read or mismatch as a fail.
Cap and closure
Missing, loose, proud, skewed or incorrect closures and tamper-evidence features.
Seal and pack quality
Visible seal defects, contamination, deformation, tears and inconsistent presentation.
Product and component
Presence, count, orientation, assembly completeness and variable surface defects.
From image to action
Inspection is useful only when the result is controlled.
Acquire
Stabilise product presentation, lighting, trigger position and camera exposure.
Decide
Apply the agreed rules or trained model and produce a clear pass/fail outcome.
Track
Maintain product identity between image capture and the reject station.
Remove
Reject the single failed product using a mechanism suited to its format and speed.
Verify
Confirm arrival in the reject area and manage missing confirmation as a fault.
What determines AI vision feasibility?
Useful trials require representative good products, known defects, normal production variation, line speed, presentation, surface finish, inspection area and the cost of false rejects versus escapes. A polished sample set that excludes difficult variation can create an unrealistic result.
AI does not replace engineering controls
Model performance depends on image quality and representative data. The line still needs stable handling, defined acceptance criteria, controlled recipe changes, fault monitoring and a verified reject route. Oxford Vision Systems treats the camera and reject station as one production sequence.
Common questions
AI vision inspection answers.
Feasibility and performance are confirmed against the real application.
Does AI inspect every product?+
An inline system can inspect every presented product within the proven speed and presentation envelope; this depends on cycle time, imaging and line control.
Can the system learn new products?+
New formats normally require controlled setup, representative examples, acceptance criteria and validation before release to production.
Can it read expiry dates and batch codes?+
Yes. OCR and OCV tools can read or verify variable print, often alongside AI or rule-based appearance inspection.
What happens to an uncertain result?+
The acceptance strategy can classify uncertainty as a fail, route it for review or stop the line, depending on the agreed risk and process.