Vision AI · MI-VAI-01
Visual Defect Inspection at Station
A trained model on one part family flagging pass/fail at line speed with a saved image per reject.
Vision AI · MI-VAI-01
An industrial camera, controlled lighting and an edge AI board at one inspection station, trained on your own part family to flag surface defects, missing components and misprints at line speed.
- ₹39,000pilot BOM · hardware only
- 6 – 10 daysto a live pilot
- Advancedintegration level
- 3validated providers
Why manual inspection drifts by shift
The problem
Manual inspection of surface defects, missing components and misprints varies by inspector, by shift and by the hour of the shift. Escapes reach the customer, rejects are counted but not photographed, and there is no data on which defect type is growing.
- Inspection consistency drops with fatigue and shift change
- No image record of what was rejected and why
- Defect types and trends are not measured
- Customer escapes cost far more than the inspection
The solution
A global-shutter industrial camera and an LED ring light are mounted at the station; an edge AI board runs a model trained on a few hundred images of your good and bad parts. Each part is classified pass/fail in milliseconds, rejects are saved with an image and defect class, and a dashboard trends defect types per shift.
- Pass/fail at line speed with a saved image per reject
- Model trained on your own parts in the pilot
- Defect-class trend per shift for root-cause work
- Reject signal available to a PLC or diverter
Validated service providers
Validated by TwoElectrons and ranked by validation score: verified credentials, completed jobs on the platform and client ratings.
- Sample Provider J — Bengaluru · Inspection & defect models · ★ 4.9 · 15 jobs
- Sample Provider K — Noida · Video analytics & PPE · ★ 4.7 · 9 jobs
- Sample Provider L — Pune · OCR & counting systems · ★ 4.6 · 8 jobs
Provider listings are samples pending live registrations.
Pilot BOM and cost
Reference hardware to run the pilot. Unit costs are indicative Indian market prices for the component class.
| Qty | Item | Unit cost | Line |
|---|---|---|---|
| 1× | Edge AI compute board GPU/NPU class, e.g. Jetson-class or SBC + accelerator | ₹22,000 | ₹22,000 |
| 1× | Industrial camera module global shutter, C/CS lens | ₹9,000 | ₹9,000 |
| 1× | LED ring / bar light with diffuser | ₹4,500 | ₹4,500 |
| 1× | Camera mount, enclosure, 12 V PSU | ₹3,500 | ₹3,500 |
| 1× | Labelling & training workflow Edge Impulse / Roboflow free tier | Free tier / open source | — |
| Pilot BOM total | ₹39,000 |
POC BOM cost only — hardware for a pilot. Installation, integration and provider services are quoted separately. Request for Quote
How camera-based defect inspection works
- Mount. Fix camera, lens and ring light at the station; set the field of view.
- Capture. Collect a few hundred good and bad part images.
- Train. Label and train the model; validate on held-out parts.
- Run. Deploy on the edge board; review escapes and false rejects weekly.
Pilot architecture
Industrial camera + ring light (global shutter · C-mount) → Edge AI board (GPU/NPU · inference) → Model & training workflow (Edge Impulse / Roboflow) → Inspection dashboard (pass/fail · reject images)
Inference runs on the edge board at the station; the cloud is used for labelling and training only.
Business case: escapes, consistency, defect trends
Ranges are typical figures reported for this class of solution; your pilot establishes the numbers for your plant.
- > 95% defect detection in the pilot class (typical for a well-lit, trained part family)
- 100% of rejects photographed (with defect class and timestamp)
- ms per part inference (at line speed)
- 6 – 10 days to a live pilot (one station, one part family)
Who this is for: Quality heads and QA engineers, Production managers, Auto components and castings, Electronics assembly and PCB, Packaging and printing, Any line with a manual visual inspection station.
Illustrative case study
Illustrative scenario · not a client reference
A die-casting plant inspecting machined housings in Coimbatore (illustrative)
Set-up. One camera station after machining, ring light in a shrouded fixture, edge AI board in the panel; 600 images labelled for porosity, burrs and missing tapped holes.
What happened. The model matched inspectors on porosity and beat them on missing holes, which the night shift had been missing. Escapes on that part fell to zero over the following quarter and the reject-image library became the basis for a casting-process fix.
- Pilot budget: ₹39,000 hardware
- Detection (missing holes): > 99%
- Escapes next quarter: 0
FAQ
How many images do we need to train?
A few hundred per class is usually enough for a pilot; the provider collects them at the station in the first days.
What about new defect types?
The model is retrained as new classes appear; the reject-image library makes this a routine step rather than a project.
Can it stop the line or divert the part?
Yes — the edge board can drive a PLC input, a reject solenoid or a light tower in the scale-up phase.
What does the ₹39,000 cover?
Camera, lens, light, edge board and mounting hardware on a free-tier training workflow. Fixture design, model training effort and provider services are quoted after a Request for Quote.
