Choosing the right control chart
A decision guide by data type and subgroup size, with examples from the shop floor.
Measurements (variables data)
- One reading at a time (batch processes, slow processes, destructive tests): I-MR chart.
- Subgroups of 2 to 9: Xbar-R chart.
- Subgroups of 10 or more, or varying sizes: Xbar-S chart.
- Many part numbers with different nominals on one machine: short-run chart (DNOM or standardized).
- Small, slow drifts that a Shewhart chart misses: EWMA or CUSUM.
- Several related characteristics at once: Hotelling T².
- Mixed sources of variation within and between subgroups: I-MR-R/S (three-way) chart.
Counts (attribute data)
- Defective units, sample size varies: p chart.
- Defective units, sample size constant: np chart.
- Defects, constant inspection area: c chart.
- Defects per unit, area varies: u chart.
- Very large samples where the p chart limits are far too tight: Laney p' or u' chart.
- Rare events (days between incidents, units between defects): G chart or T chart.
Examples
- Bore diameter, 5 parts every hour: Xbar-R.
- Batch viscosity, one test per batch: I-MR.
- Percentage of late deliveries per week: p chart.
- Scratches per painted panel: c chart (same panel size).
- Days between lost-time injuries: G or T chart.
Questions and answers
Is it worth measuring instead of using pass/fail?
Almost always. A variables chart with 5 parts per subgroup detects shifts that a p chart needs hundreds of parts to see.
In RS-1: The SPC tab groups charts this way; Automatic Analysis suggests one when you open a file.
