How many bad parts left today unseen? An interactive AI quality inspection demo
Parts leave a press and pass under a vision camera. Each gets a defect score, rejects are pushed to a bin, and every image is kept against its batch. You set the reject line, then wear the press tool and watch the camera spot the pattern before a bin fills with scrap.
Every number, ID and event in this demo is simulated to show how the system behaves. None of it is customer data or a measured result.
What you are looking at
- Parts flow from the press to the camera on a conveyor.
- The camera inspects every part and the model scores it for scratches, dents, missing holes and burrs.
- Above the threshold, the pusher sends the part to the reject bin; below it, the part ships.
- Tune the threshold: watch defects caught, escapes and false rejects move against each other.
- Defect pattern: wear the press tool and dents start repeating; a cluster alert points upstream.
- Fix upstream: change the tool and the dent rate drops back.
How this maps to a real system
A real project starts with sample images of good and bad parts from your line, a check of whether the defects are visible to a camera at line speed, and lighting and optics chosen for your part.
Results, images and batch numbers go to your quality system and MES, so a customer complaint can be traced back to the exact part image.
Questions
How does AI visual inspection decide to reject a part?
A camera photographs each part and a trained model gives it a defect score. Parts scoring at or above a threshold are rejected. Where you set that threshold decides the balance between defects that slip through and good parts thrown away.
Why not set the threshold as low as possible?
A low threshold catches more defects but also rejects more good parts, which costs material and rework time. The demo lets you move the threshold and see both numbers change.
Can inspection find process problems, not just bad parts?
Yes. When the same defect repeats in a short window, the pattern usually points at a machine or tool upstream. Alerting on that pattern stops scrap at the source.
Is the data in this demo real?
No. The parts, defects, scores and rates are simulated to show how the system behaves. They are not the accuracy of any real model or customer line.
Why does the demo ask for my name and email?
So an engineer can follow up if you want to talk about your own parts. The explanation on this page is open to everyone; only the interactive view needs the form, once for all six demos.
Free scoping call. We start with one line, never change your machine logic, and you own the code and the data.