Statistical process control: reading a process while it runs rather than judging it afterwards

Control limits come from the process, specification limits come from the drawing

The most common misuse is charting against the tolerance. Limits calculated from the process describe what it does when nothing unusual is happening, which is why a point outside them means something changed. Drawing limits describe what the customer will accept and have no relationship to the process behaviour. Plotting tolerance lines on a chart produces a display that stays quiet while the process moves substantially, then reacts only once parts are already unacceptable, which throws away the entire benefit and leaves the plant with the cost of collecting data for nothing.

How the samples are grouped decides what the chart can see

Pieces taken together should be as alike as the process allows, so that differences between groups reveal change over time. Taking one part from each of several cavities and treating them as a group hides the difference between cavities inside the group variation and produces limits so wide that nothing ever signals. The same error appears with parallel spindles, multiple heads and several operators. Where sources genuinely differ, chart them separately. Getting this wrong is the difference between a chart that finds problems and one that merely occupies an operator.

A signal creates an obligation, and the obligation must be written

Charting is worthless unless a signal produces a defined action from the person watching it. That means naming what the operator does: stop or continue, adjust within stated bounds or call for help, quarantine material back to the previous good group or not. Without it, points are marked, trends are noticed, and everyone waits for a review meeting that happens after the batch has shipped. Overreaction is the mirror failure. Adjusting after every point that looks slightly high adds variation to a stable process, which is why the reaction rule matters as much as the chart.

The process has to be capable of being controlled at all

Charting suits processes that produce continuously, drift gradually and can be corrected while running: machining with tool wear, filling, extrusion, plating thickness, temperature-driven operations. It fits poorly where output is one-off, where the characteristic cannot be measured until long after the operation, where the only adjustment available is to stop, or where defects arrive as isolated events rather than as movement in a distribution. Applying charts to those situations produces impressive-looking paperwork and no control, and it discredits the method with everyone who had to fill the sheets in. Before introducing a chart, it is worth asking what the operator would actually do differently as a result, and abandoning the idea if the honest answer is nothing.

Charts that nobody reads are a running cost with no return

Walk any factory and you will find charts filled to the edge of the sheet with no annotation, no signal investigated and no adjustment recorded. They cost operator minutes every hour and return nothing. The test worth applying is whether the last few signals can be traced to a documented action. If they cannot, either the chart is on the wrong characteristic, the limits are wrong, or nobody has authority to act, and the honest response is to remove the chart rather than to run a training session about completing it properly.

Frequently asked questions

Why should control limits not be set to the drawing tolerance?

Because the two describe different things. Limits calculated from the process tell you when the process has changed, which is the early warning charting exists to provide. Tolerance lines tell you when parts become unacceptable, by which point the change happened some time ago and material is already affected. A chart drawn against tolerance sits quiet through a substantial shift and then signals too late, so the plant carries the cost of data collection without receiving the benefit.

Which characteristics are worth charting?

Ones that move for reasons you can act on, that can be measured soon enough for the action to matter, and where the consequence of drift justifies the effort. Features held comfortably by the process with plenty of room in the tolerance rarely repay charting. Features that wander with tool wear, temperature or material lot are the natural candidates. Charting everything guarantees that nothing gets attention, which is the usual reason enthusiastic programmes collapse within a year.

Can control charting work in a low-volume job shop?

Not in its usual form, because the method needs a run long enough to establish how the process behaves. What often transfers is charting the process rather than the part: spindle temperature, plating bath chemistry, machine offsets between jobs, or a repeated feature that appears across many part numbers. That shifts the subject from a product characteristic to a piece of equipment, which is stable across jobs even where the components are not.