What is a control chart?
A control chart, invented by Walter Shewhart in the 1920s, is the core tool of statistical process control (SPC). It plots a quality characteristic in time order with a centre line and control limits at about three standard deviations. Variation inside the limits with no pattern is common-cause variation, built into the process. Points outside the limits or non-random patterns indicate special causes to find and fix.
Choosing the right control chart
| Data | Chart | Example |
|---|---|---|
| One measurement per period | I-MR | Assay result per batch, daily temperature |
| Subgroups of 2–10 measurements | X̄-R | 5 shafts measured every hour |
| Larger subgroups (about 8–10) | X̄-S | 8 fill weights per sample |
| Defective units, varying sample size | p chart | Defective labels per day out of n inspected |
| Defects per constant inspection unit | c chart | Surface defects per panel |
| Defects, varying number of units | u chart | Weld defects per assembly, by shift |
Out-of-control rules
The tool applies the Nelson rules you select to the main chart (and rule 1 to the range or standard deviation chart):
- Rule 1: one point beyond a control limit.
- Rule 2: nine points in a row on the same side of the centre line: a shift.
- Rule 3: six points in a row steadily increasing or decreasing: a trend.
- Rule 5: two of three points beyond 2σ on the same side.
- Rule 6: four of five points beyond 1σ on the same side.
Once the process is stable, check whether it meets the specification with a histogram and Pp/Ppk. When a signal appears, find the cause with 5 Whys or a fishbone diagram.
Control chart example
The I-MR example charts the assay of 30 consecutive batches, with limits calculated from the first 20 (shaded). From batch B121 the results move upward. B122 is already above the upper control limit (rule 1), and rules 5, 6 and finally rule 2 (nine points above the centre line) confirm a sustained shift rather than a one-off, even for batches that are still inside the limits. That is the value of run rules: they catch a shift before batches fail their specification. Switch the chart type to see an example of each chart.
How to use this Control Chart tool
- Choose the chart type. Pick I-MR, X̄-R, X̄-S, p, c or u depending on your data. Each type shows the data it needs.
- Enter or paste data in time order. Paste your data from Excel in the order it was produced, with a label or date for each point.
- Set the baseline. Calculate limits from all points or from the first N points of a stable period.
- Look for signals. Points beyond the limits and patterns flagged by the Nelson rules are marked in red with the rule number.
- Investigate and annotate. Click a signal to record the assignable cause as a note, and exclude it from the limits if the cause has been removed.
- Export the chart. Export the chart, limits and signal list to PDF, PNG or Excel for your batch record, review or CAPA.
Frequently asked questions
What is a control chart?
A control chart (or Shewhart chart, SPC chart) plots a process measure over time with a centre line and upper and lower control limits calculated from the process data. Points beyond the limits or non-random patterns signal special-cause variation that should be investigated.
Which control chart should I use?
For one measurement at a time, use I-MR. For subgroups of 2 to 10 measurements, use X-bar R (or X-bar S for larger subgroups). For the proportion of defective units with varying sample sizes, use a p chart. For defect counts on a constant inspection unit, use a c chart; with a varying number of units, use a u chart.
What is the difference between control limits and specification limits?
Control limits come from the process itself (about ±3 standard deviations around the mean) and describe what the process actually does. Specification limits come from the customer or design and describe what is required. A process can be in control but not capable, or capable but out of control.
What are the Nelson rules?
The Nelson rules (an extension of the Western Electric rules) detect non-random patterns: a point beyond the limits (rule 1), 9 points on one side of the centre line (rule 2), 6 points steadily rising or falling (rule 3), 2 of 3 points beyond 2 sigma (rule 5) and 4 of 5 beyond 1 sigma (rule 6), among others. Each signals likely special-cause variation.
How many points do I need to calculate control limits?
Use at least 20 to 25 points (or subgroups) from a period when the process was stable. Then freeze those limits and use them to monitor new data. In this tool, set "Calculate limits from first N points" to do that.
Should I remove out-of-control points?
Only when the special cause has been found and eliminated. In that case exclude the point from the limit calculation and record the cause as a note. Points without an explanation should stay, because removing them makes the limits narrower than the real process.