SPC 101 for discrete manufacturing

Statistical Process Control (SPC) has a reputation for being academic — the domain of quality engineers with statistics degrees who produce thick binders of control charts that nobody reads. In many shops, that reputation is deserved, because SPC was implemented as a compliance exercise rather than a production tool.

Done right, SPC is one of the highest-leverage quality tools available to a discrete manufacturer. This post covers the practical basics: what control charts are, how to read them, and when to act.

What SPC actually is

SPC is the practice of monitoring a process characteristic over time and using statistical rules to distinguish between:

• Common cause variation: the normal, inherent variation in any process. A CNC spindle will never produce exactly the same dimension twice; the variation follows a distribution. Common cause variation is managed by improving the process, not by reacting to individual points.
• Assignable cause variation: variation caused by a specific, identifiable event — a tool change, a new material batch, an operator change, a machine adjustment. Assignable cause variation should be investigated and resolved.

The control chart is the tool that separates these two types. The centerline is the process mean. The upper and lower control limits (UCL and LCL) are set at ±3 standard deviations from the mean. Points outside the control limits, or patterns within them, signal that an assignable cause is likely present.

Control chart types

For discrete manufacturing, the most common chart types are:

X-bar / R (Range) chart: tracks subgroup means and ranges. Best for subgroup sizes 2–9. Standard for most machining and assembly applications.

X-bar / S (Std Dev) chart: tracks subgroup means and standard deviations. Better than X-bar/R for larger subgroups (10+) because the standard deviation is a more efficient estimator of variation than the range.

Individuals / Moving Range (I-MR): tracks individual measurements, not subgroup averages. Used when subgroup size is 1 — chemical batches, one-piece flow cells, or when measurement cost prohibits multiple samples per subgroup.

P-chart: tracks the proportion of nonconforming units per subgroup. Used for attribute (pass/fail) data.

Reading a control chart: the Western Electric rules

A process is “in control” when all points fall within the control limits and no systematic patterns are present. Beyond the simple rule of “point outside limits,” the Western Electric rules identify subtler signals:

1. One point beyond 3σ (the control limit itself)
2. Two of three consecutive points beyond 2σ on the same side
3. Four of five consecutive points beyond 1σ on the same side
4. Eight consecutive points on the same side of the centerline (a run)
5. Six points in a row steadily increasing or decreasing (a trend)

Each of these patterns has a small probability of occurring by chance in a stable process. When they appear, they warrant investigation.

Capability indices: Cp and Cpk

Control limits and spec limits are different things. Control limits are calculated from the process data and represent what the process is doing. Spec limits are what the customer requires.

Cp measures the ratio of the specification width to the process variation width. A Cp of 1.33 means the process variation fits within the spec 1.33 times — considered minimum acceptable for automotive.

Cpk adjusts for the process not being centered. A perfectly centered process has Cp = Cpk. If the process is shifted toward one spec limit, Cpk will be lower than Cp.

A capable process (Cp ≥ 1.33, Cpk ≥ 1.33) that is also in statistical control will produce parts within spec at a very high rate without inspection screening every part.

The most common SPC mistake

The most common SPC mistake is setting control limits from the spec limits rather than from the process data. This produces control charts that show virtually everything as “in control” — because the control limits are wide enough to contain nearly any natural variation — while the process may actually be producing nonconforming parts.

Control limits must be calculated from actual process data using the appropriate control chart formulas. They should be updated when the process fundamentally changes.

Getting started without a statistics degree

You don’t need a statistics degree to start using SPC effectively. Start with:

1. Pick one characteristic on your highest-volume part that you know is close to a spec limit
2. Choose the appropriate chart type (usually I-MR for initial implementation if you’re not already grouping measurements)
3. Collect at least 20–25 subgroups of data to establish initial control limits
4. Plot the chart in real time going forward
5. Investigate every out-of-control signal — even if the part measures OK

The last point is critical. An out-of-control signal on a part that measured in spec is not a false alarm — it’s the process telling you that something changed. Finding and eliminating that assignable cause is what shifts the distribution away from the spec limits.

Qontiv includes built-in SPC charting with automatic Western Electric rule detection and alerting. See how quality works in Qontiv.