What Do Buyers Actually Ask About Vision Inspection?

The first question is almost always the same: how small a defect can you find? It is a reasonable question and it is the wrong one, because the answer depends entirely on the defect and the surface, and because a station that finds everything and rejects a fifth of the parts will be switched off within a month.

These are the six questions that actually predict whether a station survives its first year, roughly in the order they should be asked.

Can It Replace Our Checking Fixture?

Sometimes, and the distinction matters because it changes what you are buying.

A checking fixture answers one question — does this part fit within the gauge — and it answers it with a physical constraint. A vision station measures the features and compares each against a tolerance, so it can tell you that a hole is 0.15 mm off position as well as that it is out of tolerance.

For a part where the gauge is the specification, the fixture is faster and needs no recipe. For a family of variants, or where you need the value rather than the verdict, vision replaces a rack of fixtures with one station. The honest answer for most buyers is that vision replaces fixtures for families and not for a single high-volume part.

What Is The False-Call Rate?

Ask this before asking about detection, because it is the number that decides whether the station is used.

On real surfaces, a false-call rate between one and seven percent is normal depending on the defect class. A supplier who claims zero is either measuring an easier requirement than yours or has not measured on production parts.

What matters more than the figure is who reviews the flags. A station whose rejections go into a queue nobody clears is overridden within weeks, and then the miss rate becomes whatever the process produces without inspection.

What Happens When The Lamps Age?

Every light source dims, and on a surface inspection station a dimming lamp reduces contrast rather than producing an error message.

The symptom is a slow rise in missed defects on one class while everything else looks normal, and it is invisible unless someone is checking. Ask what the maintenance interval is, whether intensity is logged, and whether the station carries a reference target that can be run per shift.

A station with a reference target and a recorded baseline can prove it is still working. One without either cannot, and the argument about whether it has drifted will be settled by opinion.

How Long Does A Variant Change Take?

On a station covering sixty to eighty variants, changeover is what the operators experience and what determines whether they trust the station.

The answer has two parts: how long the program switch takes, and how the station confirms it is running the right recipe. A station that relies on the operator selecting correctly will eventually run the wrong program, and the result is a pass on the wrong part.

Identifying the variant from the part — a code if one is legible, the geometry if not — costs a second and removes that failure entirely. It is worth asking for explicitly.

What Accuracy Are You Promising, And Against What?

An accuracy figure without a reference is a number without a meaning. The useful statement names the reference instrument, the feature class and the working distance.

For dimensional stations the usual reference is a coordinate measuring machine, with agreement around 0.1 mm on hole positions and repeatability better than that. For inside-hole work the resolution is quoted in hundredths of a millimetre, and for surface defect work the criterion is usually a defect size rather than a dimension.

Ask which of those three you are being quoted, because they are not interchangeable and a station quoted on one will be judged on another if the specification is not explicit.

Who Can Change The Program After You Leave?

This is the question almost nobody asks and the one that decides the station's life.

A new defect class appears, a supplier changes a casting, a fixture is modified. Each of those requires someone to adjust the station, and if the answer is a service call, the station's coverage erodes because nobody extends it.

The good answer is that the images are retained, the labels are documented, the training pipeline can be reproduced, and someone on site has run it once. That is what a handover should include, and it is worth more than an extra half-millimetre of resolution.

And The Question Nobody Asks Until It Is Too Late

Where do the rejected parts go?

A station is complete when a rejection is routed to a defined position, recorded against the part identity, and reviewed by someone with authority to release or scrap. A station that only displays a red box on a screen is a demonstration, and demonstrations are the ones that get bypassed.