Yesterday the defect rate was 7.2 percent, today it is 8.1. Is the process getting worse? Most plants answer this question every morning by feeling, and react. A run chart and, one step further, a control chart replace the feeling with a rule.
The run chart
A run chart shows one value per day or per shift in time order, with the median as center line. For powder coating line 2, the parts with a defect per day in the 20 days of the baseline, out of 1,200 parts inspected each day:
| Week | Mon | Tue | Wed | Thu | Fri |
|---|---|---|---|---|---|
| 1 | 127 | 86 | 91 | 84 | 95 |
| 2 | 116 | 92 | 85 | 97 | 89 |
| 3 | 119 | 90 | 96 | 84 | 91 |
| 4 | 116 | 94 | 87 | 92 | 89 |
The median is 91.5 parts, or 7.6 percent. The daily values range from 7.0 to 10.6 percent.
Four patterns that mean something
- Shift. Six or more points in a row on one side of the median. The level has changed.
- Trend. Five or more points in a row all rising or all falling. Something wears, builds up or is being learned.
- Cycle. A peak or a dip at a regular interval. Something happens on a schedule.
- Outlier. One point far away from all others. A single event.
On line 2 there is no shift and no trend. There is a cycle: the four Mondays are the four highest values of the period. That fits the stratification by weekday in part two, where inclusions on Mondays ran at twice the rate of the other days.
From run chart to control chart
A control chart adds calculated limits to the run chart. For a defect rate with a known number of units per day, the right chart is the p chart:
- Center line = all defects divided by all units inspected.
- Standard deviation = square root of center line x (1 - center line) / units per day.
- Upper and lower control limit = center line plus and minus three standard deviations.
For the baseline of line 2:
| Element | Value |
|---|---|
| Center line: 1,920 / 24,000 | 8.0 % |
| Standard deviation | 0.78 % |
| Upper control limit | 10.35 % |
| Lower control limit | 5.65 % |
The first Monday with 127 parts, or 10.58 percent, lies above the upper limit. It was the first start after two weeks of plant holiday. The other 19 days are inside the limits.
Two kinds of variation, two reactions
Common causes are the many small influences that are always present. They make the points move at random inside the limits. The right reaction is to change the process itself: method, equipment, material. The wrong reaction is to adjust after every point, which adds variation instead of removing it.
Special causes are things that are not part of the normal process. They show as a point outside the limits or as one of the four patterns. The right reaction is to find out on the same day what was different, and to remove it. The wrong reaction is to ignore the signal.
One consequence is uncomfortable. A process that stays inside its limits is predictable, not good. Line 2 was stable at 8 percent, and it would have stayed at 8 percent until someone changed the process. Appeals to work more carefully do not move a center line.
The proof of a change
After the actions described in the next part, the same check sheet ran for another 20 days. All 20 daily values lay between 4.2 and 4.9 percent. Every one of them is below the old lower limit of 5.65 percent. That is not noise. The process has changed, and the limits are calculated again from the new data: center line 4.6 percent, limits 6.41 and 2.79 percent.
Plot the point on the day and write events on the chart: new powder lot, maintenance, color change. This part covers the chart for a defect rate. Charts for measured values and process capability are part of the Six Sigma series.