Key takeaways
  • A control chart separates normal process variation from something genuinely changing, so you react to real signals and not to noise.
  • Control limits come from the process data, not from the specification. Drawing them from the spec destroys the entire method.
  • A process can be in control and still produce out-of-specification parts. Control and capability are different questions.
The single most common error
Setting control limits equal to specification limits

Control limits describe what the process actually does. Specification limits describe what the customer will accept. Substituting one for the other means the chart can no longer detect change, which is the only thing it was for.

What the chart is doing

Every process varies. The chart's job is to distinguish that ordinary variation, which needs no reaction, from a change in the process, which does. Reacting to ordinary variation makes things worse, a behaviour known as tampering: adjusting a machine that was fine adds variation you introduced yourself.

Choosing a chart

  • X-bar and R. Measured values in small subgroups, the common factory default.
  • Individuals and moving range. When you get one measurement at a time, such as a batch property.
  • p or np. Proportion or count of defective units.
  • c or u. Count of defects, where one unit can have several.

Signals worth reacting to

  • A point beyond a control limit.
  • Seven or more consecutive points on the same side of the centre line.
  • Seven or more consecutively rising or falling.
  • Unusual patterns such as cycles, or points hugging the centre line too closely, which often means the subgroups are wrongly formed.

Getting started without software

Collect twenty to twenty-five subgroups under normal conditions. Calculate the average and the average range, derive the limits, then plot. Do not include known abnormal periods in the calculation, and recalculate after a deliberate process change, because the old limits describe a process that no longer exists.

In control is not the same as capable

A predictable process can sit predictably outside specification. Control charts answer 'is it stable', capability indices answer 'is it good enough'. You need stability first, because a capability figure calculated on an unstable process describes a period, not a process.