Engineering & Technology · Quality Engineering, Experiments & Process Evidence

Process monitoring, control charts and special causes

A process produces an unusual sequence of readings even though the sampled values appear acceptable for use. A monitoring chart asks whether the process still behaves like its reference model. That question differs from whether a particular product meets its requirement.

Review measurement and uncertainty first. Means, ranges, square roots and percentages are used directly. The inference lesson defines its paired-difference model and supplies its Student-t quantile; mathematics develops probability and inference further. These examples teach interpretation rather than prescribe production sampling, monitoring or release rules.

Back to the chapter

After reading, you can

Use a monitoring model to identify changes worth investigating while distinguishing signals from proof and specifications.

The limit belongs to the statistic

A control chart plots a defined statistic in time or production order against reference limits. It might show individual values, subgroup means, ranges or defect counts. The model and limits must match that statistic and the sampling arrangement. A limit for individual lengths cannot be transferred unchanged to averages of several lengths.

In our original teaching model, subgroup size is 4 and the underlying independent individual-length standard deviation is supplied as 0.50 mm. Then the standard deviation of the subgroup mean is 0.50/√4=0.25 mm. Assume a stable normal reference model with mean 50 mm and known parameters. These are supplied assumptions, not estimates obtained from the next six values.

Limits organise an investigation

Three-sigma limits for those means are 50±3(0.25), or 49.25 and 50.75 mm. The six constructed subgroup means are 49.9, 50.1, 49.8, 50.2, 50.0 and 50.9 mm. The last exceeds the upper reference limit. It is a signal to investigate the process, measurement and relevant operating history.

The signal does not prove a particular fault or guarantee that the process changed. Unusual values can occur under a reference model, and a different model or dependent observations can change signal behaviour. A lack of signals also cannot prove that every relevant property is stable. This tiny exercise illustrates the rule; it is not a fitted or validated production monitoring scheme.

Six original subgroup means plotted in order. With supplied centre 50 mm and standard deviation of the mean 0.25 mm, limits 50±3(0.25) give 49.25 and 50.75 mm. The sixth mean 50.9 exceeds UCL. The first five points cannot establish stability, and a signal cannot identify its cause or establish an individual part is out of specification.

Distinguish common variation from a changed condition

Common-cause variation is variation represented by the established process model. A special cause is a specific changed influence that may explain a departure. Investigate candidates with observations and physical knowledge. A chart can direct attention to a time window, but it cannot decide whether material, setup, measurement or another interaction was responsible.

Keep monitoring limits separate from specifications and avoid moving limits merely to absorb an inconvenient signal. A justified process change or improved reference estimate may require a new baseline, with the old record retained. Additional pattern rules can improve sensitivity while increasing false signals. Choose and document the monitoring and response approach for the actual process instead of adding every rule after seeing the plot.

Check a signal without changing the question

The sixth mean differs from the reference centre by 0.9 mm, which is 0.9/0.25=3.6 standard deviations of the mean. It crosses the supplied three-sigma limit. The reference parameters came from the teaching assumptions; recomputing them from these six values would answer a different question and could conceal the intended comparison.

Suppose a separate item specification is 48 to 52 mm. That interval does not replace the chart limits, and subgroup means alone do not establish every individual item’s conformity. Investigate the signal while retaining the relevant item observations, sampling conditions and measurement evidence needed for any separate product decision.

Try the reasoning

Why is the standard deviation of the mean smaller in the stated model?
Show the answer and reasoning
A mean of four independent equal-variance observations has standard deviation σ/√4. Dependence or another sampling model can change this.
Does a point outside a control limit identify the root cause?
Show the answer and reasoning
No. It identifies an observation inconsistent enough with the chosen reference rule to warrant investigation.
Can a specification boundary be used as the chart limit without analysis?
Show the answer and reasoning
No. Product requirements and statistical monitoring limits have different meanings and derivations.

What to carry forward

Match the chart to its statistic and reference model, and use signals to guide evidence-based investigation rather than automatic causal conclusions.

References

Teaching resources and research supporting this lesson’s explanations and examples.

National Institute of Standards and Technology (NIST)3 sources