Control
Standardize and monitor
Control makes the gain permanent. Deliverables include a control plan, documented standard work, updated procedures, training, a response plan naming who does what when a signal fires, a monitored metric with an owner, and project closure with lessons learned and formal handoff to the process owner.
Choosing an SPC chart
For continuous data use I-MR for individual values, Xbar-R for subgroups of two to nine, and Xbar-S for larger subgroups. For attribute data use p (proportion defective, variable sample size allowed), np (count defective, constant sample size), c (defect count per unit with constant opportunity), and u (defects per unit with varying opportunity).
Control limits versus specification limits
Control limits come from the process itself, the mean plus or minus three sigma, and never from the customer. Specification limits do not belong on a control chart. Out-of-control signals include a point beyond three sigma, a run of seven or eight on one side of the centerline, a trend, and two of three points beyond two sigma. Reacting to common-cause variation is tampering and actually increases variation.
Traps
Using an np chart when sample size changes, treating Cpk of 1.00 as fine for a safety characteristic where 1.33 or 1.67 is expected, and closing the project before the control plan has a named owner.
Worked example
A line produces one part every hour and each part is measured individually: use an I-MR chart. If five consecutive parts are measured hourly and averaged, use Xbar-R (n = 5). If inspectors count defective invoices per day with a varying daily volume, use a p chart.