Control Limits are indicated on a Statistical Process Control Chart as the UCL (Upper Control Limit) and LCL (Lower Control Limit). They serve as the decision criteria for identifying a signal that the process is being acted upon by special cause variation.
Control Limits are almost always set at 3 Standard Deviations above and below the overall mean of the data. At this point, most of the distribution will fall between the UCL and LCL unless special cause variation exists in the process. According to the Empirical Rule, this proportion is at least 99.73% for a symmetric bell-shaped (normal) distribution; for all other distributions, Chebyshev's Theorem states that the proportion is 89% or higher. Thus, any data points falling outside the control limits indicate a signal that special causes are acting upon the process.
In general, the Lower control limit is calculated as the Target or mean minus 3 times the estimated standard error of the mean. The Upper control limit is calculated as the target (or mean) plus 3 times the estimated standard error.
When to Recalculate Control Limits: - http://www.isixsigma.com/library/content/c021202a.asp