- Attention and eyesight tire over a shift, so the same inspector can miss more late in the day.
- Different people also apply the standard differently, which is called inspector-to-inspector variation.
- Inspecting 100 percent of parts by eye does not guarantee that 100 percent of defects are found.
- A camera assist gives each part the same look, every time, and leaves the decision to a person.
- Records from the assist show where and when things drifted.
What 100 percent visual inspection asks of a person
Many shops inspect every part by eye before it ships. It sounds safe: nothing leaves without being looked at. In practice it asks a person to hold a high level of attention for a long time on a task where nearly every part is fine. Most of the work is confirming that nothing is wrong.
This is a known weak point of human attention. When defects are rare, the brain expects a good part, and a faint scratch is easy to read past. Small parts make it worse, since the inspector has to look closely, often under a magnifier, and often at a part with tiny features and shiny surfaces.
Be careful with claimed numbers. Figures for how effective manual inspection is vary a lot between sources and settings, and many are anecdotal. The safer statement is that 100 percent inspection by eye is not the same as 100 percent detection, and the gap is worth measuring in your own shop.
How performance drifts over a shift
Fatigue builds slowly, so it is hard to notice from the inside. Eyes tire from repeated focusing. Attention wanders after a long stretch of identical parts. Early in a shift, an inspector may be sharp. After several hours, or after a break that interrupts the rhythm, the same person may miss more.
Other things shift the result too. Lighting at the bench changes as the sun moves or overhead lamps age. A rushed schedule pushes people to speed up. A big order, a late delivery or a long weekend changes how carefully anyone looks. None of this is a lack of skill. It is how people work. The effect is that quality is not constant across a day. Parts inspected at the end of a long run may have a higher chance of an escape than parts inspected at the start. Because nobody records what each look found, the drift stays hidden until a customer complains.
Differences between inspectors
Two good inspectors can disagree. One passes a light scratch that the other rejects. Without a shared reference, the standard lives in each person's head. A new hire learns by watching someone else, so the unwritten standard can drift as people come and go. This is often what is behind arguments over what counts as a defect. Shops that have bought outside inspection systems report long disputes with vendors over what was or was not a scratch. The cause is usually that the standard was never written down in a way anyone could test.
A practical fix, with or without a camera, is a boundary sample set: a few parts that are clearly acceptable, a few clearly rejectable, and some borderline ones with a written decision. Everyone inspects against the same samples. It makes the standard visible and gives you a test set to measure against.
Where a camera assist helps
A camera assist looks at every part the same way. It does not tire, and it applies the same score at the end of a shift as at the start. When a part looks unusual, it flags it and shows where. The inspector then spends attention on the flagged parts, which is the work people are good at.
This changes the job from searching every surface to judging a short list. It can also act as a second pair of eyes at the point where human attention is weakest. And because the same setup sees every part, it gives you a consistent baseline to compare against. It works best with fixed lighting and a fixed position, so the picture is the same each time. It is a poor fit where the part cannot be shown to the camera, or where the defect is something the camera cannot see, such as a dimension or an internal flaw.
What a camera should not replace
A camera does not replace the inspector. People handle the unexpected: a damaged fixture, an odd finish, a defect type nobody planned for. They also make the final call on borderline parts and know the customer's expectations. A good assist keeps the person in charge. That means designing for review. The system should flag, show why, and let the person accept or reject with a written reason. It should keep the original verdict visible instead of overwriting it, so you can later see where people and the system disagreed.
It also should not be sold as a guarantee. Any inspection method misses some defects and flags some good parts. The honest question is whether the combination of camera and person beats the person alone, measured on your own parts with a test set you trust. Write down every disagreement and review it on a regular schedule.
VZSOR is an operator assist. The camera flags suspect parts and shows where on a heatmap, and the operator decides. An operator can override a verdict with a written reason, and the original system verdict is kept. It is pre-pilot, and the pilot measures how it performs on your parts.