Offer · Cameras and edge AI on your line: fewer defects reach your customer
KaizoVision Quality Control
Visual quality inspection on the production line or in the warehouse. Industrial cameras and a local GPU computer next to the line detect defects and damage, send OK/NOK to the PLC and log every result per serial number. Images never leave the plant.
For whom
Automotive suppliers (Tier 1/2) first, then series production (appliances, furniture, electronics), logistics and warehouses (damaged parcels and loads), and food and pharma (labels and packaging). Our team brings 15 years in automotive, ISO 26262 and ASPICE, and hands-on vision and edge AI.
Runs on
AI Edge Box Camera or Line at each line, plus one AI Office Box Pro or Pro+ that collects results and calculates KPIs. Compare AI Box packages →
System diagram
Inputs, outputs and interfaces
What goes into the AI Edge Box at the line, what comes out, and through which interface. Everything runs inside your plant: images never leave it.
Your network: your data stays here
Inputs
Industrial cameras and lighting
An image of every part, set up for the defect: scratches, missing features, dimensions, assembly.
GigE Vision or USB3 VisionPart trigger and serial number
A signal when the part is in position, and its serial number for traceability.
PLC: digital I/O or OPC UA
AI Edge Box Camera / Line at the line
- 1Image capture and pre-processing
- 2Defect detection and segmentation model
- 3Vision-language model that describes and classifies unusual cases
- 4OK/NOK decision, with the image kept as evidence
- 5Results from every line collected on one AI Office Box Pro / Pro+, which calculates the KPIs
Outputs
OK/NOK to the PLC
The PLC rejects the defective part within the line's takt. Not connected during the shadow-mode pilot.
Digital I/O, OPC UA or PROFINETResults in the MES
Result, defect type and image per serial number, for traceability.
MES interface, per your MESOperator screen
Doubtful parts with the image; the operator confirms OK or NOK.
Touch screen at the stationKPI dashboard
Live PPM, escape rate, false reject rate and defect trends per shift and machine.
Web dashboard in your LAN
Learning loop: every part the operator confirms becomes a new training sample, and the model is retrained as part of the subscription.
Interfaces
| Interface | Direction | Protocol / format | What passes |
|---|---|---|---|
| Cameras | Line → Edge Box | GigE Vision or USB3 Vision, with industrial lighting | An image of every part |
| PLC | PLC ↔ Edge Box | Digital I/O, OPC UA or PROFINET, agreed with your automation team | Part trigger and serial number in; OK/NOK reject signal out |
| MES | Edge Box → MES | REST API, database or file export, depending on your MES | Result, defect type and image link per serial number |
| Operator screen | Operator ↔ Edge Box | Touch screen at the station | Doubtful parts and the operator's decisions |
| KPI dashboard | Office Box → quality team | HTTPS in your LAN, any browser | PPM, escape rate, false reject rate, trends |
| Remote service | Our team ↔ AI Boxes | VPN, under the SLA | Model updates, retraining, support |
What we measure
KPIs we measure before the pilot and prove after it
We measure these with you before the pilot (the baseline), so the return can be proven in złoty afterwards.
| KPI | Definition | How it turns into money |
|---|---|---|
| PPM | Defective parts per million delivered to the OEM | Complaint and sorting cost × incidents; risk of penalties and losing supplier status |
| Escape rate | % of defects that passed inspection and went further | Every escape is a potential complaint; the KPI the OEM watches most |
| Scrap rate | % of production rejected as scrap | Material and labour in rejected parts |
| False reject rate | % of good parts wrongly rejected | Unnecessary scrap and re-inspection; decides whether production accepts the system |
| Inspection FTE | Visual inspectors across all shifts | Employer cost per FTE × FTEs; people move to higher-value work |
| Inspection time / takt | Time to inspect one part vs line takt | Whether inspection slows the line |
| OEE | Availability × performance × quality | The overall effect, for the plant director |
Architecture
From camera to PLC and dashboard
Image acquisition
Industrial cameras and lighting chosen for the defect: scratches, missing features, dimensions, assembly.
Edge AI
A local GPU computer next to the line: a defect detection and segmentation model, plus a vision-language model that describes and classifies unusual cases. Data does not leave the plant.
Integration
OK/NOK signal to the PLC (part reject), results written to the MES, traceability per serial number.
Dashboard
Live KPIs: PPM, escape rate, false reject rate, defect trends per shift and machine.
Learning loop
The operator confirms doubtful cases, and new samples are used to retrain the model.
Rollout
Five stages, each with a result you can check
- 11–2 weeks
Discovery
Interview with the quality team, choice of one line and 1–3 defect types, baseline KPIs. Result: agreed KPIs and success criteria.
- 22–3 weeks
Data
Camera installation, collecting and labelling OK/NOK images. Result: a training dataset.
- 33–4 weeks
Offline proof of concept
Model training and a test on historical data. Result: an accuracy report against manual inspection.
- 44–6 weeks
Pilot in shadow mode
The system runs alongside the inspectors, with no effect on production. Result: hard data on escape and false reject rates.
- 54–8 weeks
Rollout
Integration with PLC and MES, training, documentation and handover. Result: a production system and a subscription.
Return on investment
Example calculation
Illustrative assumptions; in the discovery call we replace them with your baseline numbers.
| Item | Assumption | Saving per year |
|---|---|---|
| Fewer OEM complaints | 6 incidents a year × PLN 50,000, halved | PLN 150,000 |
| Less scrap | Scrap of PLN 200,000 a year, reduced by 25% | PLN 50,000 |
| Lighter manual inspection | 2 FTE × PLN 90,000 moved to other work | PLN 180,000 |
| Total benefit | PLN 380,000 | |
| Solution cost (year 1) | Rollout + subscription | about PLN 200,000–300,000 |
ROI = (lower complaint cost + lower scrap + cost of freed FTEs − solution cost) / solution cost. Our target: payback in under 12 months.
All our offers run on the same on-premise platform: an AI Box in your local network with your own models. Your data does not leave the company, which matters for GDPR, trade secrets and medical data. We take care of the hardware, updates and secure remote service (VPN, SLA).
Start with KaizoVision Quality Control
Book a short discovery call. We measure your baseline KPIs first, so the result can be proven in numbers.
Request MVP