A trial run comes back with a Ppk of 1.14 on the flange angle against a target of 1.33. The natural reaction is to ask the shop to tighten the process. Sometimes the process is fine and the gauge is the problem.
What measurement variation does to Ppk
Every recorded value is the true part dimension plus measurement error. The spread you calculate from trial data is therefore the process spread and the gauge spread added together. A noisy gauge makes a capable process look incapable, and near the limits it sorts good parts as bad and bad parts as good.
The example below uses a 90° ± 1° flange, so the tolerance is 2.0°.
| Quantity | Value | Note |
|---|---|---|
| Trial mean | 90.25° | 30 parts |
| Observed standard deviation | 0.220° | process + gauge |
| Observed Ppk | 1.14 | (91 − 90.25) / (3 × 0.220) |
| Gauge standard deviation (from study) | 0.120° | %GRR = 6 × 0.120 / 2.0 = 36% |
| Process-only standard deviation | 0.184° | √(0.220² − 0.120²) |
| Ppk the process would show with a good gauge | about 1.36 | (91 − 90.25) / (3 × 0.184) |
The calculation shows why the number is low. It is not a way to approve the part. The right response is to improve the gauge or fixture, repeat the study, re-measure the trial parts and recalculate.
Three checks before you trust a number
1. Resolution
The gauge should resolve no coarser than about one tenth of the tolerance. For a 2.0° tolerance, a protractor that reads to 0.1° is acceptable; one that reads to 0.5° is not.
2. Repeatability and reproducibility
DSQP scales the study to the risk level of the part:
- Level C: 5 parts × 1 operator × 2 repeats as an initial repeatability screen.
- Level B: 10 parts × 2 operators × 2 repeats, using parts across the feature range.
- Level A: 10 parts × 3 operators × 3 repeats.
Report measurement variation as a percentage of tolerance with a stated method. As a guide, 10% or less normally passes; 10–30% needs a written engineering decision and a control plan; above 30% needs improvement before the gauge is used for qualification. For critical features, also look at bias, stability and the risk of a false accept, even when the percentage is low.
3. Go/no-go fixtures
Pin fixtures and check gauges need their own study. Start with 20 known parts, including at least five borderline or nonconforming ones, and have two operators make two blinded passes. A false accept on a critical feature is a stop: correct the fixture and test it again.
Then stability, then capability
Once the gauge is trusted, plot the trial values in production order. Points beyond control limits, steps at coil changes or long runs on one side of the mean are signals to investigate before any index is calculated. Ppk describes a stable process; on an unstable one it describes nothing in particular.
Two more cautions:
- For a one-sided characteristic such as burr height or minimum yield, use only the applicable side.
- For a clearly non-normal distribution, choose and document a suitable method instead of forcing a normal two-sided Ppk.
A short trial is an initial estimate in any case. Keep checking the first production lots, and repeat the measurement study whenever the gauge or method for a critical or significant characteristic changes.
Atlas Insights are general technical guidance. Customer, safety and regulatory requirements, and your engineer’s decisions, always govern.