- A good record ties a result to a specific part, time, person and photo.
- A photo and a heatmap show what the result was based on, not just the verdict.
- Recording the model version and any override reason keeps the history honest.
- Records help you answer questions faster. They do not make you compliant by themselves.
- Keep records easy to export so they are usable, not just stored.
The questions you get asked
Customers and auditors ask similar things. Was this lot inspected? By whom, and when? What did you find? What happened to the parts that failed? Did the method change between lots? A complaint adds sharper questions: was this particular serial number checked, and what did it look like then? Many small shops answer from memory, paper travelers or a spreadsheet. That works until the question is about a part from months ago. A signature on a traveler shows that someone signed. It does not show what the part looked like, and it is hard to search across lots.
The cost is time. Digging through folders for one answer can take a morning, and the answer may still be incomplete. Records that are kept per part, at the moment of inspection, turn that into a short search. A searchable record makes that search a matter of minutes.
What a per-part record holds
A useful record is small but complete. It identifies the part, shows who inspected it and when, states the result, and keeps the evidence. The goal is that someone who was not there can understand what happened and why. Each field answers a different question that someone may ask later.
Think of it as a chain: which part, which setup, which look, which decision. If any link is missing, the record is weaker. A result with no photo cannot be reviewed. A photo with no part number cannot be matched to a shipment. Missing any one of these weakens the record when you need it most.
- Time of inspection
- Part number, lot and serial number
- Operator who ran the inspection
- Result: pass or fail
- Score behind the result
- The photo that was inspected
- A heatmap showing where the system saw something unusual
- Model version in use at that time
- Override reason, if a person changed the result
Why photos, scores and versions matter
The photo is the strongest piece of evidence. If a customer later reports a scratch, you can pull up the image taken before shipping and see whether the scratch was there. That settles many disputes quickly, in either direction, and it protects you when the damage happened after the part left.
The score and heatmap show how close a call was. A pass with a very low score is different from a pass just under the threshold. Over time, scores let you see drift, such as a slow rise caused by tool wear or a change in raw material. The model version matters because the method changes. If you retrain or change a setting, results before and after are not directly comparable. Recording the version, and the threshold in force, lets you explain why two lots were judged differently. Without it, a change can look like an unexplained inconsistency.
Overrides, reviews and honest history
Sometimes a person disagrees with the system. They may pass a part the system flagged because the mark is a harmless tool line, or fail one the system passed. This is a normal and healthy part of operator-assist inspection, and it should be recorded, not hidden. Two rules keep the history honest. First, require a short written reason for every override. Second, never overwrite the original system result. Keep both, side by side: what the system said, what the person decided, who they were and why. An auditor can then see that the process has human control and a trail.
Overrides are also useful data. If operators keep overriding the same kind of flag, the threshold or the training set may need attention. If one person overrides far more than others, you may have a training or standards issue. None of this is visible without a record. Review overrides on a regular schedule, not only after a complaint.
Using records without overclaiming
Records support your quality system. They do not replace it. Having a per-part log does not by itself satisfy any standard or customer requirement. What a standard asks for, and what evidence is enough, depends on your customer, your industry and your quality manual. Check with your quality lead or auditor.
Make the data usable. Export to CSV so you can filter by lot, date or result, and attach it to a report when asked. Decide how long to keep photos, since images take space and retention rules differ by customer. Back them up like any other quality record. Finally, keep the records honest about their limits. A visual inspection record shows what a camera and a person saw on the visible surfaces. It does not prove dimensions, hidden features or material properties. State what the inspection covers, so no one reads more into the record than it holds.
Each VZSOR inspection stores the time, part number, lot, serial, operator, result, score, photo, defect heatmap and model version. An operator can override a verdict with a written reason, and the original verdict is never overwritten. Records export to CSV. The camera and AI model run on the station computer in your shop, and the dashboard is a web app. A per-shop photos-stay-on-station setting keeps part photos on the station, so only counts and dollar figures reach the cloud. The records are evidence for your quality system and do not make your shop compliant on their own.