AI quality control
Detecting surface, assembly, completeness and labelling defects. An OK/NOK decision with the defect location marked and the image archived.
Machine vision and production
AI visual inspection and production automation
We build AI-based machine vision systems that check every part on the line, detect defects, measure and count, and pass the result on: to the PLC, a robot or the shift report. We have experience in automation and robotics, and we design our own electronics and prototypes.
65 min technical consultationQuote in writingNo commitment
What is an AI vision system
A machine vision system is a camera, lighting and software that judge a product from an image. Classic systems follow fixed rules and handle repeatable measurements well. AI adds what rules struggle with: defects whose appearance varies, such as scratches, inclusions, discolouration or incorrect assembly, and products that differ between batches. The model learns from images of good and defective parts from your own line.
Quality inspection station
Choose a part and run the inspection. You'll see how the system pinpoints the defect, makes a decision and logs it in the shift report. Values in the demo are examples.
Choose a part and press “Inspect part”.
Shift report0 pcs
- OK0
- Scratch0
- Missing hole0
- Dimension0
Values are examples. In a real deployment the model learns from images of your parts and your defects, and we set the reject thresholds together with your quality team.
How we deploy a vision system
Station analysis
We look at the line, the parts and the defects and set the requirements: cycle time, resolution, lighting conditions, how parts get rejected.
Feasibility test
Using images from your production, we check whether the defects can be detected with the required accuracy before you buy any hardware.
Build and integration
Choice of cameras and lighting, the AI model, integration with the PLC, robot or reject unit and with your reporting system.
Commissioning and retraining
Go-live on the line, tuning thresholds together with your quality team, retraining the model on new cases.
What we do
Measuring, counting and identification
Dimension measurement, counting parts and packages, reading codes and markings, recognising the product variant.
Electronics and prototypes
We design electronics, controls and station prototypes. When an off-the-shelf device doesn't fit, we build our own.
Production automation
Integration with robots, PLCs and production systems. Shift reports, alerts and quality data go where they're needed.
Where we deploy vision systems
They work best where inspection is manual today, defects vary and an escaped defect is expensive.
Metalworking and parts
Scratches, pitting, missing holes, threads, out-of-tolerance dimensions. Also part inspection before shipping.
Assembly and electronics
Assembly completeness, component placement, connectors, cables, labels and markings.
Packaging and logistics
Package completeness, label legibility, piece counting, pallet checks before loading.
AI visual inspection vs. classic machine vision
You don't always need AI. If the task is measuring one dimension under repeatable conditions, a classic vision system is often simpler and cheaper. AI has the edge with irregular defects, variable products and difficult lighting. During the analysis we tell you straight which approach makes sense for your line.
Experience in automation and robotics
The vision system is only part of the station. What matters is what happens after the decision: the signal to the PLC, the reject unit, the sorting robot, the entry in the report. We have experience in automation and robotics, so we design the whole station rather than just the camera, and we build the electronics and prototypes ourselves.
How much does an AI vision system cost
Cost depends on the number of cameras and stations, the line's cycle time, the type of defects, the required accuracy and the integration with automation. We start with a feasibility test on your images so you know what to expect before buying hardware. You get the quote in writing.
Frequently asked questions
How many images do you need to train the model?
It depends on the defect and how much the product varies. A few hundred images are often enough to start, and we keep retraining the model while the line runs. We set the number in the feasibility test.
Will the vision system keep up with our line's cycle time?
We size the hardware to the cycle time. The model runs locally at the line (edge AI), so the decision is made in real time without sending images to the cloud.
Can you integrate the system with our robot or PLC?
Yes, integration with automation is part of the deployment. We check the specific interfaces during the station analysis.
Where do you work?
We're based in Wrocław, Poland, and deliver projects across Poland and Europe. We usually do the station analysis on site; later work is partly remote.
What about defects the model has never seen?
The system flags uncertain cases for a person to review, and those images go into retraining the model. You set the decision thresholds together with your quality team.
Data security: nine principles we follow on every project.How we protect data
Show us your line
Send a few images of defective parts or tell us about the station. In 65 minutes we'll assess whether AI visual inspection makes sense here and what a feasibility test would look like.
Book a consultation