SPC With Inspection and Measurement Data: Which Control Charts and Which Responses

Which control chart is the right one comes down to two questions. Do you have variable measured values or attribute inspection results? And how does the data arise: in sample subgroups, as individual values, or as a gap-free stream from 100% inline measurement?

For measured values in subgroups, the x̄-R chart is the standard. For individual values, it is the I-MR chart. For pass/fail and defect count data, you use the attribute charts (p, np, c, u).

The response logic matters as much as the chart choice. A control chart separates common from special causes of variation, and only special causes justify stepping into the process. This article walks through chart selection, response rules, the basics of Cp and Cpk, and what changes when every value is available instead of samples.

Key Takeaways

Choose the control chart by data type (variable or attribute) and by how the data is collected (subgroup, individual value, or 100%).
The core of SPC is separating common (random) causes from special (actionable) causes.
A control chart signal calls for a search for the cause, not a tolerance decision.
Cp and Cpk connect process variation to the tolerance. 100% inline measurement requires a deliberate adjustment of the classic sample-based charts.

Why Is the Difference Between Common and Special Causes the Core of SPC?

SPC rests on one basic idea. Not every deviation is a problem, and not every deviation justifies an intervention. Every process varies from common causes. These are the sum of many small, built-in influences such as material variation, play, and environment. This variation is stable and predictable, and it defines the "voice of the process."

Special causes, such as a broken tool, the wrong lot, or a shifted fixture, change the process abruptly or systematically. The control chart is the tool that tells the two apart statistically. Its control limits are calculated from the process variation, not from the tolerance.

The most important behavior rule follows from this. If you react to common cause variation and keep readjusting the process, you make the variation worse (overcontrol, or "tampering"). If you ignore special causes, you let avoidable defects run. SPC is therefore less a charting technique than a discipline for making decisions.

Which Control Chart Fits Which Inspection and Measurement Data?

Chart selection follows the data type and how the data is collected. The overview below covers the cases that are common in manufacturing.

Data situation	Suitable control chart	Typical example	Notes
Measured values in subgroups (n = 2 to 9 parts per draw)	Suitable control chart x̄-R chart (mean and range)	Typical example Hourly sample of 5 turned parts, diameter	Notes The classic standard. A subgroup should contain only common cause variation.
Measured values as individuals	Suitable control chart I-MR chart (individuals and moving range)	Typical example Batch processes, long cycle times, costly inspections	Notes More sensitive to non-normality. Trends are easy to see.
Fraction of defective units, variable sample size	Suitable control chart p chart	Typical example Fraction of failed parts per shift from final inspection	Notes Needs large samples to detect small changes.
Number of defective units, constant sample size	Suitable control chart np chart	Typical example Failed parts in a fixed inspection lot of 200	Notes Easier to communicate than a p chart.
Number of defects per unit or area	Suitable control chart c or u chart	Typical example Paint defects per car body, shrinkage voids per casting	Notes Counts defects, not defective parts.
100% measurement stream from inline sensing	Suitable control chart I-MR on individual values, or x̄ charts on rational time windows	Typical example Every part on a line is measured optically	Notes Rethink how you form subgroups (see below).

One basic rule runs through all of this. Prefer variable measurement data to attribute data whenever both are available, because a measured value tells you how close the process is to the limit, and a pass/fail result does not.

Which Response Fits Which Control Chart Signal?

A control chart signal doesn't call for a tolerance decision. It calls for a search for the cause. That is the most commonly misunderstood point in SPC practice. A tiered response logic has proven itself, documented in advance as a reaction plan (OCAP, out-of-control action plan).

A point outside the control limit. Stop the process or separate the parts, identify the last change (tool change, lot change, setup), remove the cause, and confirm the effect on the chart.
Systematic patterns inside the limits, such as several points in a row on one side of the center line or a steady trend. This points to drift (tool wear, temperature). Prepare a planned intervention before the limit is reached.
Unusually low variation. Don't ignore it. It points to measurement problems, rounded values, or a real improvement that you should understand and possibly standardize.
No signal. Don't intervene. Adjustments without a signal are overcontrol and raise the variation.

It is important to keep responsibilities apart. The control limit belongs to the process (response: look for the cause). The tolerance limit belongs to the product (response: hold the parts). A process can be out of control while all parts are in tolerance, and the reverse.

What Do Cp and Cpk Really Tell You?

Cp and Cpk connect the process variation to the tolerance. They answer whether a stable process can reliably meet the specification. Cp compares the tolerance width to the process spread, and it describes the potential with perfect centering. Cpk also takes the position of the mean into account, and it describes the actual capability.

Here is an illustrative example. Say the tolerance width is 0.0060 inch and the process spread (six standard deviations) is 0.0040 inch. Then Cp = 0.0060 / 0.0040 = 1.5. Now say the mean sits off center, so only 0.0020 inch remains to the nearer limit. With a standard deviation of about 0.00067 inch, that is three standard deviations, and Cpk falls to 1.0. The difference between Cp and Cpk is therefore the direct price of off-center running.

Two prerequisites are often skipped. Capability figures are only meaningful for stable processes, so the control chart comes first and the Cpk second. They also assume a capable measurement system, because the observed variation always includes measurement variation. A measurement system analysis (MSA/GR&R) of the gauge belongs before the capability study. For basic terms, see KEYENCE's page on accuracy and repeatability in measurement.

How Does 100% Inline Measurement Change Classic SPC Assumptions?

When every part is measured, the sampling logic that classic control charts are built on goes away. That calls for deliberate adjustments, not just carrying over the x̄-R routine. Three points are decisive.

First, subgroup formation. A classic subgroup ("5 parts per hour") should contain only common cause variation. With a gap-free data stream, you have to redefine subgroups, for example as rational time windows, per cavity, or per spindle. Otherwise they mix systematic differences into the variation estimate, and the control limits come out wrong.

Second, signal flood. With a very large number of data points, even stable processes regularly produce statistical false alarms. Your response logic has to allow for that through a suitable choice of rules and through prioritizing, or the organization gets numb to signals.

Third, the opportunity. The data stream reveals patterns that samples never show, such as differences between cavities, periodic effects, or settling behavior after setups.

Where SPC Reaches Its Limits

SPC is not a cure-all, and there are situations where control charts contribute little or are used wrongly. In small lots and one-off production, there often isn't enough data per feature for reliable control limits. Standardized deviation charts across part families, or a focus on process parameters instead of product features, tend to help more.

Autocorrelated processes, such as slow thermal processes where successive values depend on each other, violate the independence assumption of classic charts and create false signals. Control charts also only show that something has changed, not why. Without root cause analysis and a maintained reaction plan, SPC stays plain documentation.

Finally, SPC doesn't replace capable gauges. If the measurement variation is too large, the chart monitors the measurement system instead of the process.

Frequently Asked Questions

Q Which control chart suits data from 100% inline measurement?

Two approaches have proven themselves. One is an I-MR chart on individual values, when each value is to be judged on its own. The other is x̄ charts on sensibly formed time or structure windows, such as per cavity or per production lot. What matters is that the subgroups contain only common cause variation, or the control limits won't be right.

Q What is the difference between control limits and tolerance limits?

Control limits are calculated from the actual process variation and show when the process has changed. Tolerance limits come from the drawing and decide whether a part is usable. The two must not be mixed up. A signal at the control limit triggers a search for the cause, and a tolerance violation triggers holding the parts.

Q At what Cpk value is a process considered capable?

A common industry minimum for series production is Cpk = 1.33, and many customers, especially in automotive, require higher values for critical features. What counts is always the customer requirement or the applicable standard. Every Cpk figure assumes a stable process and a capable measurement system.

Q Can you run SPC with attribute (pass/fail) inspection data?

Yes, with p, np, c, or u charts, but with much lower sensitivity. Attribute charts need large samples to detect changes, and they give no early warning before the tolerance is violated. Wherever a feature can be measured, prefer variable data, because it shows drifts before defective parts are made.

Next Steps

The value of SPC stands or falls with reliable measurement data from the line. KEYENCE can help you check which features can be captured inline with optical measurement systems, laser profilers, or vision systems. To discuss it, ask an expert about your application, including how results can be passed to your SPC environment, or request a demo or test on your real parts.