Visible defects
Scratches, stains, tears, shape or colour outside the normal range, according to what you consider a defect.
Quality · Today and tailored
Quality control with machine vision: defects, labels and real count
A fixed camera looks at the product as it passes: it detects visible defects, checks that the label is correct, inspects the packaging and counts what actually comes out. It is a second pair of eyes. The decision on each part is still made by a person.
Batch labelCorrect
Part on conveyorPossible defect
Box countMatches
Quality control alert: a part with a possible defect. A person decides whether to mark it as bad.
Illustrative example
What it detects
What the camera checks
You show it examples of good and bad parts from your production; without real examples there is no good detection.
Labels
It checks that there is a label, that it is legible and that it is the one corresponding to the batch being manufactured.
Packaging
Lid, closure and presentation of the packaging, where it can be seen from the outside.
Actual count
It counts what happens and compares it with what the machine or the delivery note says.
Defective part with reason, batch and photo
The operator marks the defective part on the tablet with its reason, and can take a photo. The part is linked to the batch and OEE uses that data for quality.
View defective parts and batchesPhoto that ADRI reviews as an adviser
If you enable it, ADRI gives an opinion on whether the photo confirms the reason. It never changes quantities or reasons. It comes switched off.
How the alert arrives
From the data to the person who can act
An alert that does not reach the person who can fix it is useless. That is why we design it back to front.
- The camera sees the part
A fixed camera with good light looks at each part or each box as it passes. Light is half the job: it is taken care of from the start.
- The system alerts
If something does not add up, it alerts the person in quality control with the image of the part. It does not mark anything on its own.
- A person decides
The person who knows the product decides whether it is a defect. If it is, the part is marked as defective with its reason and linked to the batch.
No hype
What works today and what is tailored
Available today
- Defective part with reason, batch and optional photo, from the operator's tablet.
- OEE quality with declared data: parts made minus defective parts, marked as declared or estimated.
- Batch per order, which inherits from the order line and can be edited in the kiosk.
- Photo with ADRI's opinion, optional and off by default. A fixed camera on the machine can send the photo along with the sensor alert.
What already works is in production control.
Tailored: we set it up and test it at your plant
- Automatic detection of defects with a camera in real time. Today the product does not analyse video.
- Label reading and packaging checks.
- Counting by image where a sensor cannot fit.
- GPU server at your plant or in Europe.
Before installing a camera
The camera looks at the product, under clear rules
Even if the camera is looking at the part, people also walk through a plant. We explain how we approach it and what is up to you.
How we approach it
- It detects the workstation, the conveyor or the area, not the person.
- No facial recognition: the system does not know who anyone is.
- No ranking or scoring of workers, and no comparing workers with each other.
- No inferring emotions: at work this is prohibited by Article 5 of Regulation (EU) 2024/1689, except on medical or safety grounds.
- No sound and no cameras in changing rooms, toilets or canteens.
- Images are processed on your own server, at the plant or in Europe, and stored for a short time. What is retained are alerts and figures.
What's down to you, and we'll review it with you
- Inform the workforce beforehand, in an explicit, clear and concise way, and their representatives if there are any (art. 89 of Organic Law 3/2018).
- If there is a works council, its prior report on work organisation and control systems and the information on algorithms or AI systems that affect working conditions (art. 64 of the Workers' Statute).
- Carrying out the impact assessment on data protection where applicable (art. 35 of the GDPR).
- Careful with the purpose: an AI system to monitor and evaluate the performance and behaviour of people at work is considered high-risk (Annex III of the Regulation). That is why we don't set it up that way.
LOPDGDD art. 89ET art. 20.3 and 64GDPR art. 35Regulation (EU) 2024/1689 art. 5 and Annex III
We review it with you and with your employment and data protection adviser. This is not legal advice nor a guarantee of compliance: every plant is different. We cover it in more detail in Cameras and AI in the factory: what the law allows in Spain.
How we test it
One line first, then you decide
We choose a control station
A point where checks are already done by eye, so we can compare what the camera sees with what your operators see.
Examples and testing
We gather examples of good and bad parts, set up the camera with its lighting and test it with real production.
We tune it, you decide
We fine-tune what it flags and what it doesn't. If it works, we take it to other lines.
Frequently asked questions
What we usually get asked
Does it replace human quality control?
No. The camera flags what looks like a defect and the person decides. A bad part is always marked by whoever is at the machine.
How many types of defect can it distinguish?
The ones we teach it with your examples. The more real examples of good and bad parts, the better. We start with the defects that matter most.
Does it work for food?
Yes, for what can be seen: shape, colour, visible foreign body, labelling. It does not replace laboratory analyses or the checks required by your regulations.
What happens to the photos that are stored?
Photos of bad parts are private: they can only be seen inside the application with Production's permission, and they are deleted once the retention period is up.
Budget-friendly, no commitment
Tell us about your line and we'll tell you what we'd set up
A short video call to understand what is bringing your line to a standstill, where and why. Then a written proposal with what we would do, where the AI would run and how much it would cost. No commitment.