Control Charts
The method of Shewhart (1931) has data plotted as soon as possible
to have a running record of process performance with
lines for target (the set point) and for upper and lower control
limit. The process is deemed "in control" when the means
are normally distributed around the target and the variance is
constant. Placing the upper and lower control limits plus and minus
three standard deviations from the target leads to a tiny
probability that performance will stray beyond them when the process
is in control. The obvious next step is
to analyze the process to determine why control was not effective
and to take corrective action.
Shewhart Chart
The figure shows new data entering on the right and flowing across the screen. The middle yellow line is the target, and the other yellow lines are the upper and lower control limit. Each time you click, you see 100 more minutes of the record.
The first attempt at programming this demo used a random number generator. The figure looks alright, but real process data do not have this white noise distribution. When you click on White Noise, you alter the input to get a normal distribution. On this contrived exercise, these normal data cluster around the target and almost never generate an extreme value. The limits and scaling must be changed to get realism. This next applet works fairly well and has a normal distribution scaled better.
Control Charts with non-normal distributions