A multi-modal contamination detection stack for an FMCG plant layers four checks so each covers the blind spots of the others: metal detection for metal, X-ray for dense non-metal objects, vision for visible faults and check-weighing for what is inside the pack. No single technology catches every contaminant. Foreign material recalls more than doubled between 2024 and 2025, so layering is now a risk decision, not an optional extra. Quality leaders can map their own detection layers against the contaminants they face in a short session.
FMCG Contamination Control · Layered Detection
Four Detection Layers, One Line, and No Single Blind Spot
iFactory AI, an on-premise industrial AI platform, brings vision, X-ray, metal detection and check-weigh results into one reject and trend record.
Vision: sees the surface
X-ray: sees density
Metal: sees conductive objects
Check-weigh: sees mass
What Is a Multi-Modal Detection Stack?
It is a set of different inspection technologies, placed on the same line, that each find what the others cannot. Fortress Technology notes that no single technology can catch every possible contaminant. The recall numbers show why plants are layering up.
42 to 89
foreign material recalls, 2024 to 2025 (Meatingplace, 2025 recall review)
13
USDA-FSIS foreign material recalls in 2025, the leading cause (Food Safety Magazine, 2026)
26.64M
units hit by foreign material recalls in H1 2026 (Food Processing, 2026)
Recall numbers count events, not near misses. For every recall there are many units a layer quietly caught, and those reject records are the evidence a plant needs to justify the next layer.
Which Technology Sees Which Contaminant?
Metal detectors are generally ineffective on glass, plastics, rubber, stone or bone, while X-ray is density-based and covers many of them (Detectamet). The matrix shows typical coverage. Plants can ask for a coverage check against their own contaminant history.
| Contaminant | Metal | X-ray | Vision | Check-weigh |
|---|
| Ferrous metal | Yes | Yes | No | No |
| Stainless steel | Yes, hardest metal | Yes | No | No |
| Glass | No | Yes | Partly | No |
| Stone | No | Yes | Partly | No |
| Dense plastic | No | Yes | Partly | No |
| Soft plastic or rubber | No | Partly | Yes, if visible | No |
| Wood or cardboard | No | Partly | Yes, if visible | No |
| Bone | No | Yes | Partly | No |
| Wrong fill weight | No | Partly | No | Yes |
Coverage is typical, not guaranteed. Product, pack and settings change what each layer can see, so verify every line with its own test pieces.
Find the Contaminant Your Line Cannot See
Share your line layout and contaminant history. iFactory AI's team will show which hazards each layer covers and where a gap remains.
Where Should Each Layer Sit on the Line?
Detection works best as a path, from raw material to the final pack. Magnetic separators and sieves also belong early in the stack (Detectamet).
Intake
Magnets and sieves catch what arrives with ingredients.
Process
Vision watches open product for visible faults.
Packed pack
X-ray and metal detection check the sealed unit.
Weigh
Check-weighing confirms net content.
Reject
One locked, confirmed reject path.
What Is Each Layer's Blind Spot?
Metal detection
Misses non-metallic objects, and foil can limit it.
X-ray
Finds density differences, so low-density items can slip past.
Vision
Surface only; X-ray evaluates the inside (Mettler-Toledo, via Dairy Processing).
Check-weighing
Cannot detect foreign objects; weight changes can have other causes (Racklify).
A blind spot is not a fault in the equipment. It is the edge of what one technology can sense, which is exactly why the next layer exists.
Is a Second Layer Redundancy or Duplication?
Two detectors of the same type fail the same way. A different physics gives real redundancy. Teams can review which layers add true coverage for their range.
Duplication
Two metal detectors in series
Same blind spots, twice
More cost, little new coverage
Redundancy
Metal plus X-ray plus vision
Different physics, different misses
New coverage for each added layer
How Do You Choose Layers by Product Risk?
Dry snack in film
Metal detection first, vision for visible plastic, check-weigh for fill.
Sauce in a glass jar
X-ray for glass and stone, since metal detection is limited by the container.
Foil-wrapped pack
X-ray as the main layer, with vision on labels and seals.
These are worked examples. Your hazard analysis and customer codes decide the final stack.
How Do You Prove Each Layer Still Works?
Every layer needs its own challenge, because a layer that is not tested is only assumed. Vision systems are challenged with a known defect sample, and reject devices are checked at the same time (MangoApps verification log).
| Layer | Challenge Used | Also Confirm |
|---|
| Metal detection | Certified test pieces in product | Reject and bin-full alarms |
| X-ray | Certified density test pieces | Image quality and reject path |
| Vision | Known defect sample | Lighting, trigger and reject timing |
| Check-weigh | Certified test weights | Under and over reject |
What Should Every Layer Record?
A shared record format turns four machines into one evidence trail. Teams can ask for a shared record template before connecting equipment.
Time stamp and line
Product and batch
Layer that rejected the unit
Reason or contaminant type
Image or reading kept as evidence
Reject confirmed or not
Test result for the layer that shift
Operator who reviewed the reject
In What Order Should a Plant Build the Stack?
Most plants add layers in phases, driven by evidence from their own rejects and complaints.
Phase 1
Prove the layer you already own: tests, rejects and records.
Phase 2
Add the layer that covers your top documented gap.
Phase 3
Join the data and trend rejects across all layers.
Layers that do not talk
Four machines, four logs, and no single contamination story.
Tests on one layer only
Metal is verified daily while vision is never challenged.
Unlocked reject paths
Rejected units re-enter the line downstream of the detector.
Where Does iFactory AI Fit in a Detection Stack?
iFactory, an on-premise AI platform for manufacturing, does not replace the inspection hardware. It joins the results so a contamination story is read once, not four times.
Metal detectorX-ray unitVision camerasCheck-weigher
iFactory AI on-premise server
Unified reject logTrend and drift alertsAutomatic work orders
Quality director (illustrative): which layer caught the most rejects on line 4 this month, and what were they?
iFactory AI: the X-ray unit, mostly dense plastic from one infeed, with images and batches listed for review.
Expert Review: Judging a Detection Stack
Expert reviewLayered contamination detection for FMCG
What holds up
Using different physics for each layer, and matching layers to the hazards in your own contaminant history.
What to verify
That each layer is challenged with test pieces on real product, and that every reject path is confirmed and locked.
What to avoid
Buying a third layer before auditing the first two, and treating any layer as a guarantee.
Bottom line: add layers where a documented hazard has no cover, and join the data.
Written from established food safety engineering practice. Equipment limits come from your inspection suppliers.
Frequently Asked Questions
Why is metal detection alone not enough for FMCG lines?
Metal detectors are generally ineffective on glass, plastics, rubber, stone and bone (Detectamet). Those materials appear regularly in recall reports, so a second layer is needed. See what your current layers miss with our team.
What does X-ray detect that metal detectors cannot?
X-ray is density-based, so it finds glass, stones, dense plastic, rubber and bone as well as many metals (Detectamet). It does less well on low-density items. Ask support which products need X-ray in your range.
Where does vision inspection fit in the stack?
Vision checks the surface and visible faults, while X-ray evaluates the inside of the pack (Mettler-Toledo, via Dairy Processing). Together they cover more than either alone. See both views joined on one line.
Can a check-weigher find foreign objects?
No. A check-weigher confirms net content, and a weight anomaly can come from damage or wrong packing, not contamination (Racklify). It is paired with metal detection or X-ray. Plan the pairing for your line in a short session.
Does iFactory AI replace inspection equipment?
No. Your detectors, X-ray and cameras stay in place. iFactory AI joins their results into one reject log with trends and work orders. Walk through the joined view with our team.
{"@context":"https://schema.org","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"Why is metal detection alone not enough for FMCG lines?","acceptedAnswer":{"@type":"Answer","text":"Metal detectors are generally ineffective on glass, plastics, rubber, stone and bone (Detectamet). Those materials appear regularly in recall reports, so a second layer is needed."}},{"@type":"Question","name":"What does X-ray detect that metal detectors cannot?","acceptedAnswer":{"@type":"Answer","text":"X-ray is density-based, so it finds glass, stones, dense plastic, rubber and bone as well as many metals (Detectamet). It does less well on low-density items."}},{"@type":"Question","name":"Where does vision inspection fit in the stack?","acceptedAnswer":{"@type":"Answer","text":"Vision checks the surface and visible faults, while X-ray evaluates the inside of the pack (Mettler-Toledo, via Dairy Processing). Together they cover more than either alone."}},{"@type":"Question","name":"Can a check-weigher find foreign objects?","acceptedAnswer":{"@type":"Answer","text":"No. A check-weigher confirms net content, and a weight anomaly can come from damage or wrong packing, not contamination (Racklify). It is paired with metal detection or X-ray."}},{"@type":"Question","name":"Does iFactory AI replace inspection equipment?","acceptedAnswer":{"@type":"Answer","text":"No. Your detectors, X-ray and cameras stay in place. iFactory AI joins their results into one reject log with trends and work orders."}}]}
Cover Every Contaminant Your Line Has Already Met
See iFactory AI join metal, X-ray, vision and check-weigh results into one view. Book a stack review using your own recall and reject history.
