- An escape is a bad part marked pass. A false reject is a good part marked fail.
- Escapes cost you customers. False rejects cost you time and slow the line.
- A threshold trades one error for the other. It cannot remove both.
- Measure on a test set you and the buyer agree on before you start.
- Zero escapes on a small set does not prove zero escapes in production.
The two errors
An escape is a defective part that the inspection marks as good. It moves on, ships, and may reach your customer. A false reject is the opposite: a good part that the inspection marks as bad. It gets pulled aside, re-checked, reworked or scrapped for no reason. People sometimes use other words. Escape is also called a miss or a false negative. False reject is also called a false alarm or a false positive. The names differ, but the two outcomes are the same. Pick one set of terms for your shop and use it every time, so that a number always means the same thing.
These errors apply to people as much as to cameras. A tired inspector can miss a scratch, and a cautious inspector can reject parts that are fine. When you compare a new method with the current one, you compare both error rates, not just one. Write the definitions down where everyone can see them.
Why both matter
Escapes are the visible failure. A bad part reaches a customer, a complaint arrives, and a corrective action follows. In regulated work, one escape can damage a supplier rating. So most shops say first that they cannot let defects through. False rejects are quieter but real. Every good part flagged as bad needs someone to look at it again. If many parts get flagged, the reviewer stops trusting the flags and starts waving them through. Then the system behaves like it has no checks at all, and real defects slip by with the false alarms.
So the aim is not zero false rejects, and it is not catching everything at any cost. The aim is a mix that your shop can live with: no escapes on the defects you care about, and a false-reject rate low enough that reviewers still take each flag seriously. Review the flags regularly so trust in them stays high.
How a threshold trades one for the other
Most automated inspection gives each part a score, where higher means more suspicious. You pick a threshold. Scores above it fail, and scores below it pass. Move the threshold down, and more parts fail. You catch more defects, but you also flag more good parts. Move it up and the reverse happens.
Picture two groups of scores. Good parts cluster low, and bad parts cluster higher, but the groups overlap. Wherever you draw the line, some bad parts fall under it and some good parts rise over it. Better lighting and a better model shrink the overlap. Nothing removes it. That makes the threshold a business decision, not just a technical one. A shop that ships safety-critical parts may accept more false rejects to push escapes toward zero. A shop making low-risk parts may choose the other way. It helps when the setting is visible and a change to it is logged.
Measuring both on an agreed test set
To measure, you need parts whose true status is known. Gather good parts, and gather defective parts that cover the defects you care about, at the sizes and severities that matter. Have your quality lead confirm each label. Then run every part through the inspection and count the results in a simple two-by-two table.
Escape rate is the bad parts marked pass, divided by all bad parts. False-reject rate is the good parts marked fail, divided by all good parts. Agree on the test set with the buyer or auditor before you run it. Otherwise a disagreement later about what counts as a defect will undo the whole exercise.
Keep the test set separate from anything used to set up or tune the system. If you tune on the same parts you test on, the result looks better than real life. Hold some parts back, and add new ones over time, especially the awkward borderline cases that people argue about.
Reading small numbers carefully
Be careful with small test sets. Say you test 40 defective parts and the system catches all 40. That is a good sign, but it does not prove the escape rate is zero. A common rule of thumb, the rule of three, says that with zero misses in n tests, the true miss rate could still be as high as about 3 divided by n, which is roughly 7.5 percent for 40 parts.
The same holds for false rejects. Two false rejects in 50 good parts is 4 percent, but a different batch might give 1 or 5. Treat early numbers as an estimate with a wide margin, and keep measuring as volume grows. Also ask whether the test defects look like real ones. Defects made in a hurry with a hard tool can differ from real wear and handling marks. A number measured on easy defects flatters the system. Include subtle ones, and say plainly how the set was made when you report the result.
The VZSOR pilot is built around these two numbers. Success is defined as zero escapes at no more than 5 percent false rejects, measured on your own parts against a defect set that you and VZSOR agree on up front. Flagged parts go to a person for review. VZSOR is an operator assist, not the final sign-off.