Statistical Process Control and Control Charts: How to Tell Whether Your Process Is in Control

Why variation is the central problem in quality control

No two products or process outputs are identical. A machining center that nominally produces a shaft of 25.00 mm diameter will produce shafts that range from 24.97 mm to 25.03 mm under normal conditions, because tool wear, temperature fluctuations, material variation, and small setup differences all introduce small, random shifts. This normal scatter is common cause variation, and it is the baseline noise level of the process. It can only be reduced by changing the process itself (different tooling, tighter temperature control, better material grading), not by reacting to individual readings.

Occasionally a reading falls well outside the normal range — say, 25.12 mm. This might be caused by a worn tool, an incorrect setup, a material substitution, or a machine fault. This is special cause variation: it has a specific, identifiable cause that can be found and corrected. Acting on it reduces the defect rate; ignoring it does not.

The problem is that without a statistical framework, operators cannot reliably distinguish between these two types of variation by eye. A reading of 25.06 mm might be the high end of normal scatter, or it might be the early signal of a drift that will produce out-of-specification parts within an hour. SPC provides the framework for making that distinction objectively.

How a control chart works

A control chart has three key elements: a center line (the process average), an upper control limit (UCL), and a lower control limit (LCL). Control limits are set at three standard deviations above and below the center line. Under a stable, normally distributed process, 99.73 percent of points will fall within these limits by chance. A point outside the control limits is a statistical signal that something has changed: the probability that a stable process produces a point beyond three-sigma is only 0.27 percent, or about 1 in 370 measurements.

The control chart is updated in real time as new measurements arrive. As long as all points fall within the control limits and no non-random patterns appear, the process is considered in statistical control and no adjustment is needed. When a signal appears (a point outside the limits, or a non-random run pattern), the process should be stopped, the special cause investigated, and the root cause corrected before production resumes.

Types of control charts

X-bar and R chart (variables data, subgroups). The most widely used control chart for continuous measurements. Measurements are taken in subgroups (typically 4 to 5 readings at regular intervals). The X-bar chart plots the average of each subgroup; the R chart plots the range (maximum minus minimum) within each subgroup. The X-bar chart detects shifts in the process mean; the R chart detects changes in process spread (consistency). Both charts must be examined together. An X-bar chart with all points inside the control limits but an R chart showing increasing ranges indicates that the process spread is growing even if the average is stable.

X-bar and S chart. Similar to the X-bar and R chart but uses the subgroup standard deviation (S) instead of the range. The S chart is more sensitive to changes in spread when subgroup sizes are larger than about 10. For small subgroups (2 to 8), the R chart and S chart perform similarly; for larger subgroups, the S chart is preferred.

Individuals and Moving Range chart (I-MR chart). Used when subgrouping is not practical because only one measurement is taken at each time point (a lab test, a batch process, a long-cycle operation). The I chart plots individual measurements; the MR chart plots the moving range between consecutive readings. I-MR charts are more sensitive to non-normality than subgroup charts, so the distribution of individual measurements should be checked before applying them.

p-chart (attribute data, proportion defective). Used for attribute data when the metric is the proportion of defective items in a sample of varying size. Each sample produces a proportion nonconforming (number defective divided by sample size), which is plotted on the p-chart. Variable sample sizes are accommodated by recalculating the control limits for each point based on that point’s sample size.

np-chart (attribute data, count of defectives). Similar to the p-chart but plots the count of defective items rather than the proportion, and requires a constant sample size. Easier to explain to operators because the charted number (count of defectives) is directly observable.

c-chart and u-chart (attribute data, count of defects per unit). The c-chart plots the count of defects per inspection unit when the sample size is constant; the u-chart plots defects per unit when the sample size varies. These charts are used when each unit can have multiple defects (solder joints on a circuit board, blemishes on a painted surface) and the metric of interest is the defect count, not whether the unit is defective or non-defective.

Control chart signals: what to look for

A single point outside the control limits is the most obvious signal, but several non-random run patterns within the control limits also indicate that the process has changed. The most commonly used detection rules (originally developed by the Western Electric Company) include:

• One point beyond three-sigma (outside the control limits)
• Two of three consecutive points beyond two-sigma on the same side of the center line
• Four of five consecutive points beyond one-sigma on the same side of the center line
• Eight consecutive points on the same side of the center line
• Six consecutive points trending steadily up or down
• Fourteen consecutive points alternating up and down (sawtooth pattern, often caused by mixing two process streams or alternating operators)

Each of these patterns has a low probability of occurring by chance in a stable process. When one appears, it is a signal to investigate the process, not to adjust the control chart limits.

Control limits versus specification limits

A distinction that trips up many practitioners is the difference between control limits and specification limits. Control limits are derived from the process itself (calculated from the actual variation of process data) and tell you whether the process is statistically stable. Specification limits are derived from customer or design requirements and tell you whether the product meets the requirement.

A process can be in statistical control (all points within control limits, no non-random patterns) and still be producing out-of-specification product, if the process spread is wider than the specification window. This is a process capability problem, addressed by process improvement to reduce variation or shift the mean. A process can also be out of statistical control (signals present) and still producing all conforming parts for the moment, if the process spread is much narrower than the specification window. The signals still need to be investigated because the instability will eventually produce non-conforming output.

SPC manages process stability; process capability indices (Cp, Cpk) measure how well the stable process meets the specification. Both tools are necessary and serve different purposes.

Setting up a control chart in practice

The setup sequence for a new control chart is: collect 20 to 30 subgroups of data from the process operating under normal, representative conditions; calculate the center line and trial control limits; examine the trial chart for signals; investigate and remove any special causes found in the baseline data; recalculate the limits from the clean baseline; and then use these limits going forward for real-time monitoring.

Control limits should be recalculated when a deliberate process change is made (new raw material, new tooling, equipment overhaul, process parameter change), because the baseline variation has changed. They should not be recalculated simply because the process has drifted, because that drift is exactly what the chart is supposed to detect.

Common errors in SPC implementation

• Using specification limits as control limits. Specification limits set by engineering represent what the customer needs; control limits represent what the process actually does. Using specification limits as control limits either misses signals (when the spec is wider than the natural process spread) or creates false alarms (when the spec is tighter than the natural process spread).
• Plotting individual measurements on an X-bar chart. An X-bar chart averages multiple measurements per subgroup, which narrows the expected variation. Plotting individual readings on a chart designed for averages produces control limits that are too tight, generating constant false alarms.
• Collecting rational subgroups incorrectly. A rational subgroup should represent measurements taken under essentially the same conditions (same shift, same machine, same operator, same material lot), so that common cause variation within the subgroup is minimized and special cause variation between subgroups can be detected. Mixing measurements from different shifts or machines within a single subgroup buries real shifts in within-subgroup noise.
• Treating SPC as a recording exercise rather than an action system. A control chart that is faithfully plotted but never acted on when signals appear provides no quality benefit. The chart only delivers value when the organization responds to signals promptly and closes the loop by documenting the special cause found and the corrective action taken.

Frequently asked questions

What is Statistical Process Control (SPC) and why is it used?

Statistical Process Control (SPC) is a quality management method that uses statistical techniques, primarily control charts, to monitor a process in real time and detect changes that could lead to defects. It is used because all processes have inherent variation, and the challenge is distinguishing between the normal random variation that is always present (common cause variation) and the unusual variation that signals a specific problem has occurred (special cause variation). Without SPC, operators tend to react to every fluctuation, which causes over-adjustment and actually increases process variability. With SPC, the control chart provides an objective statistical threshold: points within the control limits are normal noise, points outside (or non-random patterns within) are signals that warrant investigation. SPC is used across manufacturing, healthcare, financial services, and any other process where consistency and defect prevention matter.

What is the difference between common cause and special cause variation?

Common cause variation (also called natural variation or noise) is the random variation inherent in a stable process. It results from many small, independent sources that are always present: minor tool wear, ambient temperature fluctuations, small raw material differences, and subtle operator-to-operator differences. Common cause variation can only be reduced by changing the process itself (better tooling, tighter environmental controls, improved materials), not by reacting to individual readings. Special cause variation (also called assignable cause variation) is unusual variation caused by a specific, identifiable event: a worn tool, a batch of non-conforming material, an operator setup error, a machine fault, or a change in process settings. Special cause variation can and should be investigated and corrected when it appears. The distinction matters because the correct response to each type is opposite: reacting to common cause variation (adjusting a process that is behaving normally) increases variability; ignoring special cause variation (doing nothing when something has actually gone wrong) allows defects to accumulate.

How are control chart limits calculated?

Control limits are calculated from process data collected during a baseline period of stable operation, typically 20 to 30 subgroups. For an X-bar and R chart, the center line of the X-bar chart is the grand average of all subgroup averages; the center line of the R chart is the average range across all subgroups. The upper and lower control limits for the X-bar chart are calculated as the grand average plus or minus a factor A2 (which accounts for subgroup size) multiplied by the average range. The upper control limit for the R chart is D4 times the average range; the lower control limit is D3 times the average range (for subgroups of 6 or fewer, D3 is zero, so only an upper control limit applies). The factors A2, D3, and D4 are tabulated constants that vary with subgroup size. The result is control limits set at approximately three standard deviations from the center line, which means a stable process will have 99.73 percent of its points within the limits by chance. Any point outside is a statistical signal.

What is the difference between an X-bar and R chart and an I-MR chart?

The X-bar and R chart is used when measurements can be grouped into rational subgroups (multiple readings taken under essentially identical conditions at each time point). The chart plots the subgroup average (X-bar) and the subgroup range (R). Averaging within the subgroup reduces the short-term variability in the plotted point, making the chart more sensitive to sustained shifts in the process mean. The I-MR (Individuals and Moving Range) chart is used when only one measurement is available at each time point, such as in batch processes, long-cycle operations, or destructive testing where only one sample per period is feasible. The I chart plots individual measurements; the MR chart plots the absolute difference between consecutive readings as a proxy for short-term variation. I-MR charts are less sensitive to small shifts than X-bar and R charts, and they are more sensitive to non-normality. When the measurement interval is short enough to allow multiple readings, subgrouping and an X-bar and R chart are preferred. When only one reading per time period is available, the I-MR chart is the appropriate choice.

How do control limits differ from specification limits?

Control limits and specification limits answer different questions. Control limits are calculated from the actual process data and answer the question: is this process behaving consistently, the way it has in the past? A point outside the control limits means the process has changed statistically. Specification limits are set by engineering or the customer and answer the question: does this product meet the requirement? A measurement outside the specification means the part does not conform. The two sets of limits can be very different. A process with tight natural variation (small common cause noise) will have control limits much narrower than the specification; it is capable of meeting the spec easily. A process with wide natural variation will have control limits that overlap or exceed the specification; it will produce non-conforming parts even when it is statistically in control. Process capability indices like Cp and Cpk measure how the control limits relate to the specification limits. Never substitute specification limits for control limits on a control chart, and never substitute control limits for specification limits in an acceptance decision.

When should control limits be recalculated?

Control limits should be recalculated when a deliberate, verified improvement has been made to the process: new tooling is installed, raw material specifications are tightened, a process parameter is permanently changed, or equipment is overhauled. After the change, a new baseline period of 20 to 30 subgroups should be collected under the new process conditions, and new control limits calculated from this data. Control limits should not be recalculated simply because the process has drifted or because an out-of-control signal has appeared. Recalculating limits to accommodate a drift destroys the chart’s ability to detect that drift going forward. Control limits should also be reviewed periodically (annually, or whenever the product mix or process conditions change materially) to ensure they still reflect current operating conditions. If the process has genuinely improved over time, tighter control limits will reflect that improvement; if the limits are still the same but the process has deteriorated, the chart will signal more frequently, which is the correct and intended behavior.