A capability report shows a single number: Cpk 0.75. Management asks for a better machine. The machine would not have helped, because the number hid two separate problems.
Collect a baseline over time
Capability starts with a data collection plan: the characteristic, the gage, the sample, the frequency and the context. In the bore example the team measured five consecutive parts every hour, 25 subgroups in total, and recorded time, shift, tool count and spindle temperature with every subgroup.
Look at a run chart or control chart before calculating anything. Capability from an unstable process predicts nothing, because the process will not behave the same way tomorrow.
Two indices
- Cp compares the width of the tolerance with the spread of the process: (upper limit minus lower limit) / (6 × standard deviation). It ignores where the process sits.
- Cpk measures the distance from the mean to the nearer limit: the smaller of (upper limit minus mean) / (3 × standard deviation) and (mean minus lower limit) / (3 × standard deviation).
The example
Bore diameter 24.000 mm ± 0.030 mm. Baseline: mean 24.012 mm, standard deviation 0.008 mm.
- Cp = 0.060 / (6 × 0.008) = 1.25
- Cpk = (24.030 minus 24.012) / (3 × 0.008) = 0.75
Cp of 1.25 says the spread would almost fit. Cpk of 0.75 says the process sits too close to the upper limit. About 12,300 parts per million fall outside the tolerance, roughly 1.2 percent.
Two numbers, two messages: first move the process to the center, then reduce the spread. A new machine would have addressed the spread, which was not the main problem.
Turn it into money
With 180,000 housings per year and EUR 38 per reject, 1.23 percent rejects cost about EUR 84,000 per year. Ask the controller to confirm the cost per reject. A benefit confirmed at the Measure tollgate carries more weight than one estimated at the end.
Targets
A Cpk of 1.33 is a common minimum for important characteristics, 1.67 for safety relevant ones. Many customers set their own values, so ask before you set the target.
After the project, the bore reached a mean of 24.001 mm and a standard deviation of 0.006 mm: Cp 1.65, Cpk 1.60, about one part per million outside the tolerance.
Three cautions
- State the standard deviation you use. The overall standard deviation of all values and the short-term estimate from subgroup ranges give different results.
- Check the shape. The calculation assumes a roughly normal distribution. Flatness, roundness and other one-sided characteristics need other methods.
- Use defects when you must. When only good or bad can be counted, express capability as defects per million opportunities.
Capability is not a grade for the process. It is a diagnosis that tells the team where to look next.