A template to put this into practice
Save one worksheet per trial and system version. Edit it in a text editor. The file does not send data to Adaris.
The part is fine, but the system rejects it
Fictional example: a workshop inspects caps with a camera. Some acceptable caps are marked as defective when reflections change, so staff inspect them again manually. Keep examples and determine whether the problem involves image capture, the defect definition or the system's decision.
The opposite error matters too: a defective part may be accepted. Reducing rejects without measuring missed defects can hide a larger problem. Record these outcomes separately.
Agree what makes a part acceptable
The person responsible for quality should define the visible defect and acceptance criterion. An allowed surface mark and a crack are different cases. Record disagreements between reviewers before treating an image as a reference.
Separate acceptable, defective and uncertain samples. Include normal variations in finish, position and batch. A demonstration using a few easy examples does not validate an entire production process.
- Part reference and surface to inspect.
- Specific defect and acceptance criterion.
- Person responsible for uncertain cases.
- Capture conditions and situations outside the trial.
Check the image before changing the model
Lighting can cause reflections or dark and overexposed areas that obscure detail. Basler describes how different lighting techniques highlight different surface features. Select a setup for the defect being examined.
Compare images under real workstation conditions. Check framing, focus, motion, dirt and lighting changes. If the defect is not distinguishable, improve capture or reconsider the inspection approach. A single view cannot be assumed to reveal hidden damage.
Count errors with clear denominators
Fictional recording example: the system rejects 10 of 200 independently verified good parts, giving a false reject rate of 5%. It accepts 2 of 40 verified defective parts, giving a missed-defect rate of 5%. These are separate measures even when their values match. Always report sample sizes and conditions alongside percentages.
Record unusable images and referrals for manual review separately. Keep a checking set that is not used to adjust the system. Avoid placing nearly identical images of the same part in both sets, which can make results look better than they are.
Test without controlling the production reject mechanism
An initial scope can analyse saved images or run in observation mode: record proposed decisions and compare them with the agreed review without controlling machinery. Investigate disagreements before planning industrial integration.
After changing a threshold, lighting or a rule, check both error types again and record the version. Define how missing and uncertain images are handled. System acceptance and actions on the line require agreed conditions and responsible people; trial results do not replace that decision.
What to share when discussing a trial with Adaris
Describe one part reference, one visible defect, process speed and the capture point. Identify your camera or software and whether the scope involves software review, capture or integration. Start with fictional or anonymised examples; we can agree a suitable channel if production images are needed.
Adaris works from Elche with businesses in Alicante and across Spain, with commercial communication in Spanish. We can scope a computer vision evaluation with samples, criteria and a results report. Cameras, lighting, installation and machine connections require separate scoping. Proposals and invoicing are through Adaris.
Related services
Sources
Which good parts are being rejected?
Describe the inspection and when it fails. We can scope an initial evaluation with acceptance criteria, deliverables and a quote. Commercial communication is in Spanish.