Home / Solutions / Rogue/Foreign Can detection
Stop the One Wrong Can Before It Ships
Verify the identity of every can on your filling line, not just its label. MarshallAI catches mismatched products and mixed labels from the very first can of a new SKU, with no re-programming and no learning period.
Also detects fallen and damaged cans.
Can be combined with our industry leading Date label verification (OCR).
Minimize recalls and complaints
Stay compliant in every market
Avoid Hold-For-Inspect(HFI)
Zero Changeover Downtime: Recognizes each SKU automatically as it starts running. No operator “train” step, no re-teaching the system between batches. Perfect for even small batches.
Protected From Can One: Flags a mismatched can the moment it appears. No ramp-up period or sample run needed to learn what’s “normal” before protection kicks in.
99.99995% Detection Accuracy: Holds this accuracy at full line speed, on every SKU, without hand-tuning sensitivity per run.
2500+ Cans a Minute: Every single can inspected. Performance can be scaled with processing hardware.
Fallen, misaligned and damaged can detection: Every single can inspected.
Why Rogue Cans Slip Past Legacy Vision Systems
A rogue can happens when an empty can meant for one product ends up on the wrong filling line — a domestic-market can filled with an export formulation, or a leftover can from the previous batch mixed into a new run. Because many SKUs in a portfolio share near-identical can geometry and print design, the two cans can look almost identical to a camera scanning for print pattern alone.
- Traditional label verifiers are not built to tell apart two versions of the same label with only a subtle formulation difference.
- Traditional systems can be hard to tune, tightening thresholds to catch subtle differences means risking more false rejects actually OK cans.
Rogue Can Detection: MarshallAI vs. Legacy Label Verifiers
| Operational Aspect | Legacy Label / Deco Verifiers | MarshallAI |
|---|---|---|
| Detection basis | Compares printed pattern/color against a reference; flags deviation past a set threshold | Verifies each can’s full product identity against the SKU that should be running |
| Near-identical SKU variants (e.g. domestic vs. export) | Prone to miss — differences can fall inside the pattern-matching tolerance | Built to catch — trained on every SKU’s distinct signature, not just gross print differences |
| New batch / SKU changeover | Operator triggers a “learn” step for the new label; protection relies on the previous batch’s memorized pattern until it’s retrained | Recognizes the new SKU automatically — protected from the first can, no learning step |
| Detection reliability | Depends on sample size and a sensitivity setting, trading off missed defects against false rejects | 99.99995% accuracy, consistent across SKUs and runs |
| Scope | Single-purpose module for label/deco inspection | One station covers rogue can, fallen can, and damaged cans. |
Case studies
-
Eliminating Rogue Cans: High-Speed QA in Beverage ManufacturingA global beverage manufacturer spent 15 years trying to reliably intercept “rogue cans” on their high-speed filling lines. Discover how MarshallAI deployed a minimal latency edge vision system to inspect 2,500 cans per minute, preventing mismatched products and costly mechanical jams with near-perfect precision.
Beverage industry Quality assurance
Questions & Answers
Does MarshallAI need to “learn” each new SKU before it can protect the line?
No. Every SKU in your portfolio is modeled in advance, so the system recognizes what’s running the moment a changeover happens. There’s no operator-triggered training step and no window where the line is protected only by a memorized previous batch.
What happens when a rogue/foreign can is detected?
We can issue an alert, halt the line or integrate with a rejection mechanism. Timelined statistics are also provided for total products and faulty products for OEE optimisation and root cause analysis.
Can it tell apart two versions of the same can — like a domestic and export variant?
Yes. This is the core case MarshallAI was built for. Because the model verifies a can’s full identity rather than comparing print patterns against a similarity threshold, it catches differences that fall below what pattern-matching systems are tuned to flag.
Does inspection data leave our facility?
No. Processing runs entirely on-site via our Universal Edge Stack. Can images and production data stay on your network.
Will this slow down our line or require a new SKU program for every product?
No. MarshallAI runs at full line speed — 2,500+ cans a minute — and handles SKU changeovers automatically, with no reconfiguration required.
