AOI Metrics: False Call Rate, Escape Rate and First Pass Yield

The three numbers that describe how well an AOI performs are the false call rate (good parts the machine flags as defective), the escape rate (real defects the machine lets through) and first pass yield (the share of boards that pass without rework). False calls cost review time, escapes cost quality, and first pass yield tells you how the process upstream of the AOI is doing.

These metrics only mean something if everyone uses the same denominator. This guide gives plain-text formulas, shows how to collect the data on your own line, and explains why lowering one number by loosening thresholds usually raises another.

What are the four possible AOI outcomes?

Every inspection decision falls into one of four boxes. Each metric below is built from them.

The AOI only knows its own decision. The operator's review verdict tells you which calls were true or false. Escapes can only be found later, at in-circuit test, functional test, rework, or at the customer.

What is the AOI false call rate and how is it calculated?

A false call is a component, joint or region the AOI flags as defective that the reviewer confirms is acceptable. The false call rate is usually expressed in one of three ways, and quotes that use different ones cannot be compared directly.

• Per component or per joint: false call rate = false calls ÷ components (or joints) inspected. Often reported in ppm: multiply by 1,000,000.
• False calls per board: false calls per board = total false calls ÷ boards inspected.
• Per board (board-level): false call board rate = boards with at least one false call ÷ boards inspected × 100%.

The per-board count is the one operators feel, because it sets how often they stop the line or pull a board to review. The per-component rate is better for comparing machines across boards with different component counts. Always state which one you mean.

What is the AOI escape rate?

The escape rate measures what the AOI misses. It is the metric that protects your customer, and it is harder to measure because escapes are only found downstream.

• Escape rate = defects found downstream that the AOI should have detected ÷ (defects the AOI correctly called + those escapes) × 100%.
• Detection rate = 100% − escape rate.
• Escape ppm = escapes ÷ inspection opportunities × 1,000,000, when you want a figure normalized to volume.

Count only defects within the AOI's scope. A hidden BGA joint void found by X-ray is not an AOI escape, because an optical system cannot see it. Keep a separate list of defects that no inspection in your line covered.

What is first pass yield in SMT?

First pass yield (FPY) is the percentage of boards that go through a process step correctly the first time, with no rework or repair.

• FPY = boards that pass without any rework ÷ boards entering the step × 100%.
• AOI pass rate = boards with no AOI calls ÷ boards inspected × 100%.

The two are not the same. AOI pass rate is pulled down by false calls, so a machine that cries wolf makes your process look worse than it is. True FPY is calculated after review, counting only boards with confirmed defects as failures. When FPY drops, the AOI data should help you find the upstream cause: print, placement, or reflow.

How do I measure these metrics on my own line?

1. Fix the denominators. Decide whether you report per board, per component or per joint, and use that for every product and every machine.
2. Record every review verdict. Each AOI call needs an operator decision stored with it: real defect or false call, plus the defect type. Without this, the false call rate is a guess.
3. Trace escapes back. When test, rework or a customer return finds a defect, check whether it was on a board the AOI passed, and whether that defect type is in the AOI program.
4. Build a known-defect sample set. Keep boards (or images) with confirmed defects of each type you care about, and re-run them after program changes to check nothing that was caught is now missed.
5. Split by product and by component type. A line-wide average hides the one package that generates most of the false calls.
6. Watch the trend, not one day. Measure over weeks, and mark the dates of program changes, material changes and supplier changes.

What does good AOI performance look like?

We are not going to quote an industry benchmark, because the right figure depends on board density, component mix and your defect definitions. A useful test is qualitative:

• Operators actually look at every call. If they have started passing calls without looking, the false call rate is too high, whatever the number says.
• Escapes of critical defect types are rare and traceable. Each one has a known cause and a program change behind it.
• Watch escapes when false calls fall. If false calls decrease but more defects are missed, it may reflect looser thresholds rather than better inspection.
• New products reach a stable false call rate quickly. Wherever new boards arrive often, the tuning period on each one matters as much as the steady state.

Why does lowering false calls often raise escapes?

On a rule-based AOI, the usual way to stop a false call is to widen the acceptance threshold for that check. That widening applies to every component judged by the same parameter, so a real defect that falls inside the new band is now passed too. The false call rate drops and the escape rate quietly rises.

Our guide to reducing AOI false calls covers the other root causes, such as lighting, board warp and supplier variation.

How does feedback learning improve these metrics?

DaoAI's feedback learning changes the model instead of the threshold. When an operator marks a false call as OK at the review screen, that image becomes a labelled example. Running an AI update retrains the model on it, so the component that looked unusual is now recognized as normal, while a defect near the same score still falls outside what is normal.

The operator chooses the scope of the correction: one position, a part number, a package or a class. Anything set by hand is kept, and an update can be undone. The same mechanism handles acceptable variation between suppliers, such as a different marking font or body color on the same part number.

For measurement, DaoAI's production integration reports first pass yield per whole board and per sub-panel, stores every inspection with its images and verdict, and exports results to your MES as CSV. Per-position yield in a panel points to where a stencil or reflow problem sits.

FAQ

What is a false call in AOI?

A false call is a component or joint the AOI flags as defective that a reviewer confirms is acceptable. It costs review time and, when frequent, trains operators to ignore calls.

What is the difference between a false call and an escape?

A false call is a good part flagged as bad. An escape is a bad part passed as good, found later at test, rework or at the customer.

How is first pass yield calculated?

First pass yield equals boards that pass a step without any rework divided by boards entering that step, times 100%. For AOI, count only confirmed defects as failures, not false calls.

Should I report the false call rate per board or per component?

Report both if you can. Per board reflects operator workload; per component or per joint, in ppm, allows fair comparison across boards with different component counts.

Can I lower false calls without raising escapes?

Yes, if the fix teaches the system what an acceptable part looks like instead of widening a threshold. Feedback learning retrains the model on confirmed false calls and leaves the acceptance band for real defects in place.

The fastest way to see these numbers for your product is to measure them on your board. Send us one of your own boards and we will test it for you, free of charge, and review the false calls and detections with an engineer.