PLATFORM
From the first image to the decision on the line.
AIVQC is three specialised components that together prepare, publish and run an approved inspection — instead of a single, closed "AI box".
→TRAIN
AIVQC Trainer
You mark the defects and set the goal and time — automatic training does the rest. Trainer turns images of your product into a tested inspection model without machine-vision expertise and publishes it to AIVQC Production stations.
- +Automatic training
- +Leak-free data
- +Evaluation like the line
View component →
→CONTROL
AIVQC Server
Server is the control point between engineering work in Trainer and real inspection on the line — it publishes only approved versions, and only to the stations you select.
- +Local deployment
- +Access control
- +Release history
View component →
→INSPECT
AIVQC Production
Production is the component running right at the line — it analyses images locally, makes an OK/NOK/Error decision and passes it on to the process, with no dependency on a constant Internet connection.
- +Local image analysis
- +Recipes and rollback
- +Modbus TCP and history
View component →
WHERE EACH PART RUNS
Every component in its place — and on your network.
The split into three components is deliberate. Experimental work does not mix with production, publishing goes through one controlled point, and the decision is made where the product is — at the station on the line.
Each component runs on the plant network and uses hardware you already have, if it meets the station requirements. Details are on the Trainer, Server and Production pages.
| Component | Where it runs | Who uses it | What it is responsible for |
|---|---|---|---|
| AIVQC Trainer | One engineering workstation per site, optionally with a GPU worker on the local network | Process or quality engineer | Data, annotation, automatic training, evaluation, packages |
| AIVQC Server | A Docker container on a Linux server, Synology NAS or Windows — or a service on the Trainer computer (offline mode, up to 3 stations) | IT administrator | Identities, package distribution, history, licensing |
| AIVQC Production | A computer at each inspection station | Operator, technician, expert | OK/NOK decision, PLC signals, inspection history |
CLOSED LOOP
A model that gets better with every change on the line.
The three components form one cycle: difficult cases from the line go back to training, and a new model version goes back to the line — in a controlled way and with a full history.
- 01
Images
Photos and recordings of the product, marked defects and confirmed defect-free images — in Trainer.
- 02
Training
Automatic training within a time budget chooses the model and parameters — no machine-vision expert needed.
- 03
Publishing
A verified package goes through Server to the selected station only.
- 04
Inspection
Production decides locally and sends signals to the PLC — even without a network.
- 05
Feedback
NOK and low-confidence images return through Server to Trainer as material for the next version.
DEPLOYMENT
We deploy it together — or your team does.
You can deploy the platform with us — from hardware and data qualification to line integration — or on your own. If you have your own engineers and IT staff, you have everything you need: Server installs like any Docker container, and automatic training needs no machine-vision expert.
With an in-house deployment you pay for a software licence, not for integrator hours — usually the biggest saving. We quote both ways individually.
PRINCIPLES
No cloud, no closed ecosystem, no guessing.
The platform is designed for plants where production images should not leave the network, and where a station must not stop just because the Internet went down.
- No cloud: images, datasets and models stay on site, and inspection and publishing work without the Internet. You activate the licence online or offline, with a file.
- No closed ecosystem: we first check the camera and computer you already have instead of selling a new set.
- No guessing: when a station is not sure of the result it reports it instead of letting the product through, and an error is never treated as OK.
- Traceability: every package has checksums, every publication and activation is recorded, and every result has its recipe and person attached.
- Roles instead of a shared password: separate permissions for operator, technician, expert and administrator — including for decision-threshold changes.