No process produces identical parts. Even the best-calibrated machine generates results that fluctuate within a range. The key question in quality is not whether variation exists—it always does—but what type of variation it is: the kind that belongs to the process by nature, or the kind that signals that something has broken. Statistical Process Control (SPC) is the set of statistical tools that answers that question in real time and allows you to act before producing defects, not after.
Developed by Walter A. Shewhart at Bell Labs during the 1920s and later popularised by W. Edwards Deming, SPC remains, a century on, the foundation of modern industrial quality and an explicit requirement in sectors such as automotive (IATF 16949), pharmaceuticals and medical devices.
Common-cause versus special-cause variation
Shewhart's conceptual contribution was to distinguish two sources of variation. Common-cause variation is inherent to the process: the background noise produced by a multitude of small, random factors (slight differences in temperature, in material, in measurement). A process that exhibits only common cause is said to be in statistical control: it is stable and predictable within its limits. Special-cause variation (or assignable cause) comes from something identifiable and external to normal operation: a worn tool, a defective batch of material, a change of operator.
The most costly mistake in process management is to confuse the two. Reacting to common-cause variation as if it were special—adjusting the machine every time a measurement drifts a little—introduces more variability, not less. Deming called it tampering and illustrated it with the famous funnel experiment. SPC offers the objective criterion to avoid falling into that trap.
The control chart: anatomy and reading
The central tool of SPC is the control chart. It plots the process data over time against three lines: the centre line (the process mean) and two control limits, upper (UCL) and lower (LCL), typically set at three standard deviations from the mean. It is crucial to understand that control limits are not specification limits: the former describe what the process does (the voice of the process); the latter, what the customer demands (the voice of the customer). Confusing them is one of the most frequent mistakes on the shop floor.
As long as the points fall within the limits and are distributed randomly, the process is in control. To detect special-cause signals, the Western Electric rules (or Nelson's rules) are applied, identifying patterns that are improbable under pure chance: a point outside the three-sigma limits; eight consecutive points on the same side of the centre line; sustained increasing or decreasing trends; or repetitive cycles. Each pattern suggests an investigable cause.
Types of charts according to the data
| Type of data | Chart | Measures | Example |
|---|---|---|---|
| Variables (continuous) | X-bar and R | Mean and range of subgroups | Diameter of a machined part |
| Variables (continuous) | X-bar and S | Mean and standard deviation | Large subgroups (n > 10) |
| Variables (individual) | I-MR | Individual value and moving range | Batch chemical processes |
| Attributes (defectives) | p / np | Proportion or number of nonconforming units | % of rejected parts |
| Attributes (defects) | c / u | Number of defects per unit | Defects per metre of fabric |
The practical rule is simple: if the characteristic is measured on a continuous scale, you use variables charts (X-bar/R is the workhorse); if it is counted (conforming/nonconforming, number of defects), you use attributes charts.
Process capability: Cp, Cpk, Pp and Ppk
A process being in control does not mean it meets the specifications. Process capability compares the natural spread of the process with the width of the required tolerance. The Cp index measures the potential (how many times the process variation fits within the specification), but it ignores whether the process is centred. Cpk does take centring into account and, for that reason, is the reference indicator: a Cpk of 1.33 is the minimum usually required by industrial customers, and 1.67 is demanded for critical safety characteristics. The Pp and Ppk indices are their long-term performance analogues, calculated with the overall variation instead of the subgroup variation.
An inescapable requirement: it only makes sense to calculate the capability of a process that is already in statistical control. If special causes are active, the process is unpredictable and any capability index will be a fiction.
Steps to implement SPC on the shop floor
- Select the critical characteristic to control, prioritising those with the greatest impact on the customer.
- Validate the measurement system with an MSA/Gage R&R study: if the measurement is not reliable, the chart is noise.
- Collect data in rational subgroups and calculate the control limits using data from the process itself.
- Stabilise the process by eliminating special causes until only common cause remains.
- Study the capability (Cp/Cpk) once stable and compare it with the requirement.
- Monitor and improve: keep the chart alive, react only to signals and attack the common cause to reduce background variation.
SPC in real time: from paper to sensor
SPC was born with a pencil, graph paper and manual measurements taken by the operator every so many parts. That logic remains valid, but the connected industry has transformed its scope. Today, sensors embedded in the line capture measurements continuously and feed dashboards that update the charts instantly, without waiting for periodic sampling. This capability changes the nature of the response: instead of discovering a deviation when reviewing the sheet at the end of the shift, the system triggers an alert the moment a special-cause signal appears, while there is still room to correct before generating scrap.
This evolution, integrated into manufacturing execution systems (MES) and into connected quality platforms, does not eliminate the statistical fundamentals; it amplifies them. The risk, precisely, is letting the volume of data dilute judgement: measuring more does not equate to controlling better if the limits are wrongly calculated or if you react to every micro-oscillation. Automation must respect the same Shewhart and Western Electric rules that govern the manual chart, and keep the quality engineer in the decision loop. The sensor detects the signal; the person investigates the cause. Confusing an abundance of data with knowledge is the modern version of the old mistake of tampering with the process.
SPC, Six Sigma and the data culture
SPC does not exist in isolation. It is the statistical backbone on which broader improvement methodologies rest. In the DMAIC cycle of Six Sigma (Define, Measure, Analyse, Improve, Control), control charts are the leading tool of the Control phase: once a process has been improved, SPC ensures that the improvement holds over time and does not degrade as soon as the project team withdraws its attention. Without that statistical control phase, improvements tend to revert and the effort invested dissipates. The connection with Deming's PDCA cycle is equally direct: the control chart provides the objective evidence that closes the Check phase and decides whether a corrective action has truly worked or only appears to.
Beyond the technique, implementing SPC successfully requires a cultural change. A plant where the operator understands the difference between common and special cause stops chasing every oscillation and begins to trust the process when it is in control. That shift in mindset—reacting to signals, not to noise—is usually harder to achieve than mastering the formulas, because it clashes with the natural instinct to "do something" in the face of any deviation. Training shop-floor staff in reading charts and in the meaning of control limits is, therefore, as important as choosing the right chart. Data without interpretive judgement becomes a source of erratic decisions; data with judgement becomes the basis of a stable and predictable process.
Common mistakes
- Using specification limits as control limits: SPC loses all of its detection power.
- Readjusting the process in response to common-cause variation (tampering), which worsens variability.
- Calculating Cpk on an out-of-control process, obtaining a meaningless number.
- Poorly formed subgroups, which mix different sources of variation and distort the limits.
- Not validating the measurement system before trusting the data.
Frequently asked questions
What is the difference between control limits and specification limits? Control limits are set by the process itself (its natural variation); specification limits are set by the customer or the design. A process can be in control and still fail to meet the specification.
How much data do I need to set the limits? As a guideline, between 20 and 25 subgroups before considering the limits reliable. With fewer, the limits are provisional.
Does SPC work in service processes, not just manufacturing? Yes. Any process with a measurable, repetitive output (response times, billing error rate, delivery lead times) can be controlled with attributes or individuals charts.
How does SPC relate to Six Sigma? SPC is one of the tools of the Control phase in the Six Sigma DMAIC cycle, and the statistical basis on which the goal of reducing variation rests.
Conclusion
Statistical Process Control changes the underlying question of quality: it stops being "is this part good?" and becomes "is the process capable and stable?". That difference shifts the effort away from final inspection—expensive, late and which does not prevent the defect—towards prevention in real time. The organisation that distinguishes common cause from special cause stops tampering with its machines blindly, reacts only when there is a real signal and, above all, knows in advance whether its process can meet what the customer demands. In a setting where industrial customers require documented Cpk as a condition of supply, SPC is not an optional technique but the common language in which it is demonstrated that quality is under control. At Summum Calidad we implement statistical control systems that turn shop-floor data into reliable decisions.