Illustrative scenario
Component packing: camera visual inspection
Inspect every part and keep an image of every defect, independent of tired eyes at the end of a shift.
This is an illustrative scenario with a hypothetical plant, not an Uptrol customer. The figures are estimates from the assumptions listed below, not achieved results.
- Industry
- Electronic components, packing
- Assumed size
- 1 packing line, 2 shifts, 2 inspectors per shift
- Scope
- Edge AI, PLC integration, quality traceability
- Estimated rollout time
- 8–14 weeks
Estimated benefits
2–4 peopleInspection staff moved to other work
Headline estimate216–432 million VNDLabour value per year
100%Defective parts with a stored image
See how we calculated
Assumptions
| Inspectors per shift | 2 people |
| Shifts per day | 2 |
| Cost of one person per month (salary and insurance, assumed) | 9 million VND |
| Share of inspection time that can move to other work | 50–100% |
Calculation
- 2 people × 2 shifts × 50–100% = 2–4 people
- 2 people × 2 shifts × 9 million VND × 12 months × 50–100% = 216–432 million VND
- Every NG part stored with time and shift = 100%
The level of automation is only confirmed after the test with real samples. Uptrol does not promise accuracy before testing.
Context
The hypothetical line has 2 visual inspectors per shift checking for missing parts and wrong or skewed labels. Results vary between shifts and there are no images as evidence when a customer complains.
The plant does not want images of its line to leave the local network.
What Uptrol deploys
- A test phase with real defect samples and acceptance criteria agreed up front
- Industrial camera, lighting and a light shield at the inspection point
- An AI computer at the line running the model; images stay in the plant
- OK/NG to the PLC to reject bad parts; defect images and shift statistics on the dashboard
Architecture
- 01Photo sensor triggers camera and light
- 02AI computer at the machine: image processing, OK/NG decision
- 03Line PLC: rejects bad parts
- 04Only results and defect images go to the Uptrol platform
Steps
- 01FeasibilityCollect good and defective samples, try lighting set-ups, agree criteria.
- 02Train and bench-testBuild the model, measure miss and false-reject rates on a test set.
- 03Install at the lineVibration-proof mount, synchronised with sensors and PLC.
- 04Run in parallel, hand overRun alongside manual checks, compare, hand over when criteria are met.
Related solution: Edge AI at the machine