The camera that catches what the checklist can't.
MAFLO's vision layer runs alongside your existing IQC, PQC and FQC — every coil, every sheet, every carton gets a first-pass camera check at line speed. Passes release into stock. Fails auto-raise a nonconformance with the image evidence attached. No form filling. No shift-end catch-up. No customer surprises.
Four leakages a form-based QC cannot.
Manual QC catches roughly one in three defects on a busy line — the rest reaches your customer as a complaint. A vision agent inspects every unit, every second, without fatigue. Here's where it earns its keep first.
Vendor-side dimensional drift on inbound coils
Copper coils that look OK but come in 0.02 mm under-gauge quietly steal 3-5% of your yield. A fixed camera at the goods-in bay measures every coil in under 2 seconds — pass rows enter Stock Available, fails auto-raise an IQC NC with the image attached.
Surface defects your operator stopped seeing at 2 pm
Scratches on laminate sheets, pinholes in enamel wire, cold solder on lighting boards. The vision agent sits over the line at 25 fps, tagging each defect with a barcode. The line supervisor sees a spike before the shift ends — not the customer.
Wrong-label, wrong-carton, wrong-count dispatch errors
The FQC camera reads the carton label, cross-checks against the sales order, counts the units inside and confirms the dispatch photo. A mismatch blocks Ready-to-Ship and creates a task on the QC lead — before the truck leaves the yard.
Machine states you were reconstructing from WhatsApp
A CCTV feed on the plant floor becomes a live machine-state ledger — Running / Idle / Setup / Down — without touching the PLC. Feeds the same Shift Output ledger your barcode scans populate. The OEE number stops being a guess.
Vision sits on your existing spine.
The vision layer is not a parallel system. Every camera event flows through the same IQC → PQC → FQC ladder your handheld QC already uses. Same NC register. Same barcode. Same MRP workbench. Same audit trail your ISO 9001 auditor already trusts.
| Stage | Role | What the vision agent does |
|---|---|---|
| IQC | Incoming raw material | Auto-inspect every GRN line. Pass to Stock Available. Fail creates IQC NC with image evidence and a supplier debit note draft. |
| PQC | In-process on the line | Continuous camera over the drawing / press / assembly line. Every defect is barcode-tagged and routed to Rework or Scrap. Feeds the Shift Output ledger. |
| FQC | Final QC + dispatch | Reads carton label, verifies count, checks against sales order. Mismatch blocks Ready-to-Ship. Passes generate the dispatch photo the customer receives. |
| MES | Shop-floor state | Non-invasive CCTV feed becomes a machine state ledger (Run/Idle/Setup/Down) and OEE. No PLC integration required. |
| MRP | Exception feedback | A fail-decision that scraps 200 kg of RM pegs back into the MRP workbench as a shortfall exception. The owner sees the leakage the moment it happens. |
Where we're training the first models.
Vision models earn their trust on your own samples. Each vertical starts with a supervised pilot on one line, in your plant, on your defect definitions. Once a model hits the pass/fail threshold you signed off on, it goes into production.
Wires & Cables
LiveCoil diameter drift · surface scratch · label OCR
See the vertical page →Lighting & Electrical
PilotSolder-joint AOI · driver-can dent detection · lot barcode read
See the vertical page →Laminates & Plywood
PilotSheet surface defect · press-mark detection · pack-count verify
See the vertical page →Resins & Chemicals
DesignFill-level check · label + COA verify · drum seal integrity
See the vertical page →Three tiers. Pick per checkpoint.
You don't rip out your line to add vision. You add it, one checkpoint at a time, on hardware that fits the task.
Android phone in a rugged case
Your PQC / FQC lead points the phone camera at the coil, sheet, or carton. Model runs on-device, decision in under a second.
- · No extra software for existing PQC / FQC checkpoints
- · Works offline · syncs when back online
- · One rugged Android phone per QC station
IP camera over the line
A 2-5 MP camera looks down at every unit passing the checkpoint. Frames stream to a shop-floor NUC for inference at 10-15 fps.
- · Scoped per line — quoted after a site visit
- · PoE, no shop-floor wiring changes
- · Best for surface / label / count checks
NVIDIA Jetson-class inference
A Jetson Orin at the line takes multi-camera input at 25+ fps for high-speed defect detection. Same MAFLO event schema out.
- · Pilot = one line, one camera, one model on your samples
- · Add lines one at a time as models prove out
- · Best for drawing / press / high-throughput
Bring us five defect samples. Get a model in three weeks.
We train the first model on your own defect photos, on your own line — not a generic dataset. You sign off on the pass/fail threshold. Only then does it go live.