Industry and manufacturing
Visual inspection, maintenance, process data: where AI holds up in industry and when it becomes a safety component.
Where it holds up
| Task | Value | Condition |
|---|---|---|
| Visual inspection with pre-sorting | very high | fixed capture situation |
| Completeness of assemblies | high | fixed camera position |
| Analysing maintenance reports | high | free text exists |
| Narrowing fault causes | high | plant data exists |
| Predicting maintenance needs | medium | at least 30 documented failures |
| Analysing process parameters | medium | clean time series |
Introduction
- 01
Fix the capture situation mechanically
Camera, distance, angle, lighting, background. Only then does data collection begin.
- 02
Collect defect images
Good parts are plentiful, bad ones rare. A stock of a few hundred per defect type is the target.
- 03
Pre-sort rather than decide
Good, bad, inspect. The value comes from the third group being small.
- 04
Collect the corrections
Every human re-inspection is a label. Collected systematically, a year yields a dataset nobody can buy.
- Quantify the cost per error type in advance: a defect that gets through against a false alarm.
- Document the threshold and do not change it quietly in operation.
- Track the calibration state of camera and lighting; a shift there changes quality.
When Annex I applies
An AI system used as a safety component of a product that is itself subject to conformity assessment counts as high risk. In practice:
| Application | Classification |
|---|---|
| Optical inspection for scratches | routinely not Annex I |
| Release of a safety-relevant component | Annex I |
| Distance monitoring on a machine | Safety function, machinery regulation |
| Predictive maintenance without intervention | not Annex I |
| Automatic intervention in the control system | Safety function |
The Machinery Regulation adds requirements for safety functions with self-learning behaviour. In practice this is handled with frozen model versions whose change is a documented act.
Employee protection
- Camera systems in production capture staff, even when they are meant to inspect parts. The field of view belongs technically limited.
- Analysis per workstation or shift is capable of performance monitoring and engages co-determination.
- Emotion recognition in the workplace is prohibited under Art. 5 AI Act, regardless of consent.
Entry sheet: manufacturing
FREE ACCOUNT
Entry sheet for manufacturing
This sector's tasks by risk class, the points to settle beforehand, and the metrics the benefit shows up in.
Checklist3 items