Entry
Automation audit
Written report with prioritised next steps.
Learn more →Case study · recycling
An AI camera in sorting:
feasibility first, plant second
Whether an AI camera reliably tells apart what needs telling apart in a running material stream is not something a datasheet can answer. Using the Cognex starter kit for AI cameras, we tested it on real material before anyone decided on a plant.
Discuss a similar project
01
Background and test methodology
A recycling company was evaluating whether an AI-enabled camera could reliably distinguish fractions in a running material stream. Vendors had quoted recognition rates, but these referred to laboratory conditions and sample materials, not the customer's specific stream with its characteristic contamination, moisture and variance.
We tested feasibility with the Cognex starter kit for AI cameras on the real material stream. We captured training images on the running belt, trained the model and measured recognition performance under real conditions. The result is not a glossy presentation but a defensible figure: this camera recognises this fraction in this material stream with this accuracy.
The result was nuanced: for certain fractions the recognition performance was sufficient for automation deployment, for others it was not. This information saved the client a mistaken investment in a full installation and at the same time clarified which part of the sorting problem a camera is genuinely suited to.
Cognexstarter kit for AI cameras
Real materialtested on the running stream, not on samples
Up frontfeasibility before plant engineering
Questions about this project
For which sorting tasks are AI cameras appropriate?
AI cameras work well when visual features provide reliable distinguishing characteristics: colour, shape, texture. They reach limits when variance in the material is high, when lighting and moisture vary strongly, or when recognition speed and error rate are tightly specified. A feasibility test on real material gives the honest answer.
What does an AI camera feasibility test involve?
We capture training images on the real material stream, train a model with the starter kit and measure recognition performance under real conditions. The test takes one to three days depending on material variance and requirements. The result is a written report with measured recognition rates and a recommendation.
Can the feasibility test be extended to other camera types?
Yes. Cognex is one of the leading providers of industrial machine vision but not the only standard. Depending on the task, other camera systems may also be suitable. We advise vendor-independently and recommend the system best suited to your application.
AI camera: test first, invest second.
We test recognition performance on the real material stream and give you a defensible basis for the decision.
Discuss a projectRelated services for this project
Concept
Automation concepts
Options and target state when the report points to a larger modernisation.
Learn more →Data
iniationOS
One shared data layer for ERP, MES and quality.
Learn more →Interested, but not ready to enquire?
An e-mail address is enough. We come back once with a short read on it, no newsletter.