Vision systems on cartoners now do far more than reject a bad pack at the end of the line. The 2026 generation tracks subtle defect drift, correlates upstream variables, and updates its own reference set during artwork changes. None of that is universal. Knowing what cameras actually catch reliably, and what they still miss, is the difference between a vision system that pays back and one that adds rejects without solving the underlying problem.

What vision systems on cartoners catch reliably

Six inspection types are now solved problems on a 2026-grade vision station. Print quality on the carton outer, including registration, smudging and missing text. Date and lot code legibility at the standard inkjet or laser-marked positions. Carton presence and basic orientation at the infeed and at reject. Glue bead presence and rough position on closure flaps. Closure quality, including flap up or down and corner pop-off detection. Stack alignment at the discharge.

Reject rates on those six inspection types typically run below 0.3 percent on calibrated systems. False reject rates run below 0.1 percent. Those are operational numbers, not vendor claims.

What cameras still miss on cartoner inspection

Four defect types remain difficult for vision systems on cartoners in 2026. Subtle colour drift on offset and digital print is hard to catch reliably because the camera baseline drifts with ambient lighting. Partial occlusion, where part of the carton sits behind another pack at the inspection station, defeats most rule-based systems and pushes AI systems toward false rejects.

Internal defects are largely invisible to optical inspection at production speed. Split flute, ink bleed-through and partial product fill inside a closed carton all sit in this category. X-ray or weight-check inspection covers these, but at extra capex and integration cost.

Multi-pack pattern errors, where the individual packs are correct but the multipack arrangement is wrong, show up only on systems specifically trained for the pattern. Most generic cartoner vision setups do not check pattern by default.

AI-based versus rules-based vision

Rules-based vision uses fixed reference templates and tolerance bands. The system passes or rejects based on whether the inspected pack falls inside the band. The advantage is determinism. Auditors and validation engineers know exactly why a reject happened. The disadvantage is the manual retraining work needed at every artwork change, which historically consumed an engineer for half a shift.

AI-based vision uses neural network inference. The system trains on a reference set of good and bad packs. The advantage is robustness to small artwork changes and edge cases that would defeat a rules-based system. The disadvantage is auditability. Why did the AI reject this pack is a harder question to answer than why the rule rejected it.

Most UK FMCG cartoners now run hybrid systems. Rules-based inspection covers the deterministic checks like date code and carton presence. AI inspection covers the harder checks like print quality and subtle defect drift. The hybrid approach lets the rules-based section carry the audit weight while the AI section catches what rules cannot.

Lighting, calibration and retraining

Lighting setup is half the battle on any cartoner vision system. Diffuse or directional lighting choice depends on the surface type. Flat matt cartons usually want diffuse light. Glossy or coated stock typically wants a darkfield arrangement to suppress specular reflection. Getting lighting wrong locks the rest of the system into firefighting from day one.

Calibration drifts. Cameras settle, lighting ages, line vibration moves alignment. A daily calibration check, integrated into the start-of-shift checklist, catches drift before it shows up as false rejects in production. Most calibration checks take an operator under five minutes.

AI retraining frequency depends on artwork churn. Lines running stable artwork for months between changes might retrain twice a year. Lines with weekly promotional artwork changes need retraining tooling that runs in under thirty minutes per change.

Spec a vision system on a 2026 cartoner RFQ

Five questions belong on every cartoner RFQ that includes vision inspection. The first is the defect type list. Which defects matter operationally and commercially. The second is the reject station capability. Single-lane reject, two-lane reject, automatic infeed re-injection or full reject diversion to a discrete bin.

The third is data logging. Most modern vision systems can log every inspection, every reject and every false reject to a time-stamped database. Auditors increasingly expect this data to be queryable rather than locked inside the vision system.

The fourth is AI versus rules-based versus hybrid. Validate the supplier’s recommendation against your audit position. The fifth is integration. The vision system needs to talk to the cartoner controller, the line OEE system and the broader factory data layer. OPC UA support is the practical answer in 2026.

Common pitfalls when specifying vision systems on cartoners

Three pitfalls show up consistently on cartoner vision projects. The first is over-specifying the resolution. Higher-megapixel cameras add cost, increase processing time and rarely improve real defect detection past a certain threshold. Most cartoner inspections need 1.3 to 5 megapixel sensors. Anything past that is buying margin that production never uses.

The second is under-specifying the lighting budget. Lighting is the single most important variable on a vision system, and it is usually the smallest line item on the quote. A vision system with brilliant cameras and bad lighting performs worse than a moderate system with the right lighting. Spec lighting first.

The third is integration with reject station logistics. A vision system that flags a reject is only useful if the reject station can actually divert the bad pack reliably. On high-speed cartoners, the reject station has throughput limits of its own. Build the reject capability into the original cartoner specification rather than retrofitting it after the vision system identifies the problem.

Key takeaways for vision systems on cartoners

Vision systems on cartoners catch six inspection types reliably in 2026. They miss four. AI-based and rules-based both have a role and most lines run hybrid systems. Lighting, calibration and retraining decide whether the system pays back. RFQ specifications need to cover defect lists, reject capability, data logging, AI position and integration. Vendor demonstrations using the customer’s actual cartons under the customer’s actual lighting are worth more than any specification sheet.

Talk to our engineers if you want to specify or troubleshoot a vision system on a cartoner line.

Specify your cartoner vision system with our engineers.

Talk to our engineers