Why does a PPM figure need an interval?

Because it is computed from a count, and counts vary. Twelve events could have been eight or seventeen with the process completely unchanged, and the resulting PPM figures span nearly two to one. Publishing the point estimate alone presents a draw from a distribution as though it were a measurement, and everything downstream, target comparisons, trend charts, corrective actions, inherits that error.

My rate went from 180 to 250 PPM. Did it get worse?

On these counts, there is no evidence that it did. 180 sits well inside this period's interval of roughly 109 to 391, so the two periods are consistent with an unchanged process. That is not a claim that nothing changed. It is a statement that twelve events cannot tell you either way, which is the honest position and the one that avoids chasing noise.

How do I get a tighter estimate?

More events, which usually means aggregating more periods. Relative precision goes as 1/√k, so quadrupling the count halves the width. More parts at the same rate does exactly this because it produces more events; more parts at a lower rate does not help at all, which is the counter-intuitive part: a very good process is a hard one to measure.

Is my PPM target even observable?

Check the expected count. A 200 PPM target on 48,000 parts means about 10 expected defects, and a rate that low cannot be resolved from a single period. The target may be perfectly reasonable while the reporting cadence is not; aggregating to a quarter, or accepting that the monthly figure is an event log rather than a measurement, both fix it.

Why can the lower bound go negative?

Because the normal approximation is symmetric and a small count leaves it no room. A negative lower bound is not a defect rate. It is the approximation announcing that the count carries almost no information about the rate. It is published rather than clamped because clamping it to zero would hide exactly that message.

Does this account for defects that escape inspection?

No, and the bias runs one way. Undercure is the cure failure most likely to pass inspection. A part soft beneath a cured skin looks finished and fails later at a customer, so the detected rate is a lower estimate of the true one. Any interval on this page is an interval on what was found, not on what occurred.

UV Curing calculator

UV Cure Defect Rate Calculator: PPM With Its Uncertainty

Turn a cure defect count into PPM and see how much of it is real. Enter the defects, the parts cured, your PPM target and the previous period's figure. The calculator returns the rate with a 95% interval around it, the distance from target, and the change since last period. The interval is the point: a PPM figure computed from a dozen events carries uncertainty of nearly two to one, and a comparison that ignores it turns ordinary counting noise into a corrective action.

What this calculator does

Convert a defect count into PPM and report the interval around it, so a comparison against a target or a previous period can distinguish a real change from counting noise.
Use it for reporting a cure defect rate honestly in a quality review, deciding whether a period-on-period change is real, testing whether a PPM target is observable at your volume, deciding how many periods to aggregate before drawing a conclusion, pushing back on a corrective action raised against counting noise.
Convert a defect count into PPM and report the interval around it, so a comparison against a target or a previous period can distinguish a real change from counting noise.

Formula used

Cure defect PPM = defects ÷ total cured parts × 1,000,000
95% interval on the count = defects ± 1.96 × √defects
Lower bound (PPM) = (defects − margin) ÷ total parts × 1,000,000
Upper bound (PPM) = (defects + margin) ÷ total parts × 1,000,000
Change from previous period = this period's PPM − previous period's PPM

Inputs explained

Cure-related defects this period: Defects attributable to cure: undercure, tack, adhesion failure, embrittlement from overcure. Other defect modes belong to their own processes.
Total cured parts this period: Parts that passed through the cure station in the same period.
PPM target: The rate the process is expected to hold.
Previous period PPM: The last period's figure, so the change can be judged against the interval rather than read as a trend.

How to use the result

Best suited to reporting a cure defect rate honestly in a quality review, deciding whether a period-on-period change is real, testing whether a PPM target is observable at your volume, deciding how many periods to aggregate before drawing a conclusion, pushing back on a corrective action raised against counting noise.
The normal approximation is poor below about ten counts, where the true interval is wider and asymmetric. The page warns when the count is in that range rather than pretending otherwise. Assumes a constant rate. Real cure defects cluster. A drifting lamp produces a burst, not a sprinkle, and clustering makes the true uncertainty wider than Poisson. Assumes all cure defects were detected, which is optimistic: undercure is exactly the failure mode most likely to pass inspection and appear at a customer. The lower bound can go negative at small counts. It is published rather than clamped, because a negative bound is a legible statement that the count carries almost no information. Says nothing about severity. One catastrophic defect and one cosmetic one count the same here.

Common questions

Why does a PPM figure need an interval? Because it is computed from a count, and counts vary. Twelve events could have been eight or seventeen with the process completely unchanged, and the resulting PPM figures span nearly two to one. Publishing the point estimate alone presents a draw from a distribution as though it were a measurement, and everything downstream, target comparisons, trend charts, corrective actions, inherits that error.
My rate went from 180 to 250 PPM. Did it get worse? On these counts, there is no evidence that it did. 180 sits well inside this period's interval of roughly 109 to 391, so the two periods are consistent with an unchanged process. That is not a claim that nothing changed. It is a statement that twelve events cannot tell you either way, which is the honest position and the one that avoids chasing noise.
How do I get a tighter estimate? More events, which usually means aggregating more periods. Relative precision goes as 1/√k, so quadrupling the count halves the width. More parts at the same rate does exactly this because it produces more events; more parts at a lower rate does not help at all, which is the counter-intuitive part: a very good process is a hard one to measure.
Is my PPM target even observable? Check the expected count. A 200 PPM target on 48,000 parts means about 10 expected defects, and a rate that low cannot be resolved from a single period. The target may be perfectly reasonable while the reporting cadence is not; aggregating to a quarter, or accepting that the monthly figure is an event log rather than a measurement, both fix it.
Why can the lower bound go negative? Because the normal approximation is symmetric and a small count leaves it no room. A negative lower bound is not a defect rate. It is the approximation announcing that the count carries almost no information about the rate. It is published rather than clamped because clamping it to zero would hide exactly that message.

Last reviewed 2026-08-25.