I think most spec sheets for a machine vision system bury the numbers that matter under the ones that sell well. Resolution gets top billing. Frame rate comes right after. Neither number tells you whether the system will still be catching defects on a Tuesday afternoon six months after installation, once the line has run through a dozen minor changes nobody documented.

Resolution Only Matters Against Your Smallest Defect

A camera’s megapixel count alone tells you little. The number that actually matters is pixels per millimetre across your smallest defect, at your actual field of view. A 12-megapixel camera scanning a two-metre panel can have worse effective resolution on a hairline scratch than a 2-megapixel camera scanning a fifty-millimetre part. Work backwards from the defect size you need to catch, then size the sensor and lens to hit that pixel density with real margin above the minimum. A vendor who quotes resolution without asking your smallest defect size first is guessing at a number that happens to sound impressive.

Frame Rate Has to Match Real Line Speed

A camera rated for 200 frames per second is irrelevant if your exposure time at that speed produces motion blur on a part moving past at three metres per second. Frame rate and exposure time trade against each other, and the trade gets worse as lighting drops. Ask for the exposure setting the vendor actually used in their demo footage. The maximum frame rate printed on the box is a separate number that rarely applies to your line, since real production rarely runs at the theoretical ceiling a data sheet quotes.

Where Camera Choice Actually Shows Up

The gap between a cheap sensor and a properly specified smart camera machine vision setup rarely shows up in a vendor demo. It shows up on reflective metal, on parts with variable surface finish, and on anything moving fast enough that exposure time becomes tight. A demo booth has controlled lighting and a hand-picked part. Your line has neither. A well-built smart camera machine vision setup earns its price on the parts a demo never shows you, the ones with glare, dust and inconsistent orientation.

The Spec Nobody Puts on the Sheet

Lighting consistency does more to determine accuracy than almost any camera spec, and it never appears on a data sheet. Ambient light from a nearby window, a shift change that turns on different overhead fixtures, or a forklift headlight sweeping past a station can all shift readings enough to move a system from stable to unreliable. If a vendor’s proposal never mentions how they are isolating the inspection zone from ambient light, ask them directly. A camera with a strong spec sheet cannot compensate for light that changes hour to hour.

Total Cost of Ownership Beyond the Spec Sheet

The purchase price on a quote rarely includes recalibration labour, replacement optics after a bump, or the engineering hours needed every time a product variant changes. A cheaper camera with a longer calibration cycle can cost more over three years than a pricier one that holds its settings. Ask each vendor how often their recommended setup needs recalibration under normal plant vibration, and who performs that recalibration once the installation team leaves the site.

What to Ask Before Comparing Two Spec Sheets

Before ranking another machine vision system on paper, ask the vendor for footage of it running on parts that resemble yours, under lighting that resembles your plant floor, at your actual line speed. Ask what happens when a part arrives slightly out of position, since real production tolerances are looser than a controlled test rig. Ask how long recalibration takes after a lens gets bumped, because on a real floor, something eventually gets bumped. Any machine vision system that only performs well under laboratory conditions is not the one you are buying for your actual line.

The spec sheet works as a starting filter. It never makes the decision by itself. The vendor who agrees to test on your worst parts, at your real line speed, is telling you more than any column of numbers ever will. Keep that footage on file too, since it becomes the reference point the next time a spec conversation with a different vendor starts from scratch.