How do you calculate usable computer-vision inspection capacity?

Multiply inspections per cycle by available cycles for gross capacity, then multiply by uptime and first-pass yield. With 120/cycle over 480 cycles at 94% uptime and 97% yield, gross is 57,600 and usable is about 52,520 inspections.

What is first-pass model decision yield?

It is the share of inspections the model resolves confidently as pass or fail without escalating to a human reviewer. At 97% yield, 3% of parts are uncertain — here that diverts about 1,624 inspections to manual review.

Why subtract uptime and yield separately?

They are different losses. Uptime loss (here 3,456 inspections) is capacity the station never produced because the camera or inference engine was down. Yield loss is capacity produced but not auto-decided. Tracking them separately tells you whether to fix reliability or the model.

What is a good camera and inference uptime?

Well-integrated vision cells run 95-99% uptime. The default 94% is slightly conservative and already costs 3,456 inspections of capacity over the horizon, which is why reliability engineering pays off quickly.

How do I know if one vision station is enough?

Compare usable capacity to the inspections your line will actually present. If production needs more than the 52,520 usable inspections here, you need a faster cycle, higher uptime, better model yield, or a second station.

Gross vs usable capacity — which should I plan with?

Always plan with usable capacity. Gross (57,600) is a theoretical ceiling; usable (52,520) reflects real downtime and uncertain calls, so it is the number that should be measured against takt and volume.

AI & Digital Manufacturing Analytics calculator

Computer Vision Inspection Capacity Calculator

Computer-vision inspection capacity is the number of usable, confident pass/fail decisions an automated visual-inspection station can deliver over a planning horizon, after camera and inference downtime and uncertain model calls are removed. Quality engineers and automation integrators use it to size a vision cell against takt time and to decide whether one station can cover a line or a second camera is needed. It matters because the headline 'images per second' on a spec sheet ignores the reality that cameras drop offline and models punt low-confidence parts to humans. This calculator converts raw throughput into the capacity you can actually plan around.

What this calculator does

Estimate usable computer vision inspection capacity from inspections per cycle, available cycles, camera uptime, and first-pass model decision yield.
a quality or automation engineer needs to size AI vision inspection capacity against line output
It computes usable inspection capacity by derating gross cycle throughput for camera/inference uptime and first-pass decision yield.

Formula used

Gross vision capacity = inspections per cycle × available inspection cycles
Usable vision capacity = gross capacity × camera and inference uptime × first-pass model decision yield

Inputs explained

Inspections per vision cycle:
Available inspection cycles:
Camera and inference uptime:
First-pass model decision yield:

How to use the result

Use it when sizing a vision station, validating it against production volume, or diagnosing why an inspection cell is falling behind.
It assumes uptime and first-pass yield are stable averages; in practice model yield drifts as new defect types and lighting conditions appear.

Common questions

How do you calculate usable computer-vision inspection capacity? Multiply inspections per cycle by available cycles for gross capacity, then multiply by uptime and first-pass yield. With 120/cycle over 480 cycles at 94% uptime and 97% yield, gross is 57,600 and usable is about 52,520 inspections.
What is first-pass model decision yield? It is the share of inspections the model resolves confidently as pass or fail without escalating to a human reviewer. At 97% yield, 3% of parts are uncertain — here that diverts about 1,624 inspections to manual review.
Why subtract uptime and yield separately? They are different losses. Uptime loss (here 3,456 inspections) is capacity the station never produced because the camera or inference engine was down. Yield loss is capacity produced but not auto-decided. Tracking them separately tells you whether to fix reliability or the model.
What is a good camera and inference uptime? Well-integrated vision cells run 95-99% uptime. The default 94% is slightly conservative and already costs 3,456 inspections of capacity over the horizon, which is why reliability engineering pays off quickly.
How do I know if one vision station is enough? Compare usable capacity to the inspections your line will actually present. If production needs more than the 52,520 usable inspections here, you need a faster cycle, higher uptime, better model yield, or a second station.

Last reviewed 2026-05-12.