You finish a DMAIC project, document the savings, and hand the process back to operations. Three months later the gains have evaporated and nobody knows why. The root cause is usually the same: the baseline was never statistically valid, and the Control phase was a dashboard nobody looked at. Statistical process control fixes both problems, but only if you use it from the start, not as a footnote in the final phase.

What Statistical Process Control Actually Does

Statistical process control is a data-driven methodology for monitoring, controlling, and improving manufacturing processes using statistical analysis. It was invented in 1924 when Walter Shewhart at Bell Laboratories designed the first control chart. The core idea is simple: compare what is happening today with what happened when the process was stable, and signal when the difference is big enough to matter.

SPC is proactive. It monitors process inputs in real time so you catch drift before it produces scrap. Traditional quality inspection is reactive, detecting defects after production. The whole point of SPC is to move a company from detection-based to prevention-based quality controls. An operator watching a control chart can see a trend developing and adjust before the first non-conforming part is made.

The method was widely used during World War II for munitions and weapons quality, then standardised by W. Edwards Deming, who introduced it to Japan after the war. It became a core part of Six Sigma and, by extension, lean manufacturing.

Every process exhibits variation. SPC's job is to tell you which kind you are looking at.

Common cause variation is inherent process noise: slight differences in raw material, ambient temperature shifts, normal tool wear. These are built into the system. You cannot eliminate them without changing the process itself.

Special cause variation is a signal that something specific changed: a broken tool, a new operator, a material batch out of spec. Special causes are identifiable and removable.

Misidentifying either one is expensive. Adjusting a machine in response to common cause variation, chasing every point that lands slightly above average, actually increases variation. It is called tampering, and it makes the process worse. Ignoring a special cause because it looks like normal noise lets defects reach the customer. Control charts help visualise whether a process is stable or if a special cause has emerged, so you respond to the right one.

The Control Chart: SPC's Primary Tool

Shewhart's control chart plots data over time with three horizontal lines: a centreline at the process mean, and upper and lower control limits set at three standard deviations from the mean. That creates a six-standard-deviation spread around the target. When a point falls outside those limits, it is a signal, not noise.

Control limits are derived from the process data itself. They are not the specification limits set by an engineer on a drawing. This distinction trips up practitioners more than any other. A process can be in perfect statistical control, every point within the limits, and still produce out-of-spec parts if the process is not capable. A process can also be out of control while every single part still meets spec. The control chart tells you about stability. The specification tells you about acceptability. They answer different questions.

SPC implementation follows two phases. Phase I establishes the baseline: you collect historical data, calculate initial control limits, and verify the process was stable during that period. If special causes are found, you investigate and remove them, then recalculate the limits. Phase II is real-time monitoring using the limits from the end of Phase I. Skip Phase I and your control limits are built on noise, which means every signal they produce is suspect.

A typical X-bar and R chart setup uses 25 subgroups of 4 or 5 samples each, 100 measurements total. That is enough to establish a reliable baseline without paralysing the line with data collection.

Which Control Chart to Use (and When)

The chart you pick depends on your data type. Variable data, continuous measurements like diameter, weight, temperature, uses one family of charts. Attribute data, counts of defects or defective units, uses another. Using the wrong chart produces meaningless limits.

Variable data:

  • I-MR (Individual Moving Range) chart: use when your data is individual values, one measurement per time period.
  • X-bar and R chart: use when recording data in subgroups of 8 or fewer.
  • X-bar and S chart: use when subgroup size is greater than 8. The standard deviation gives better sensitivity with larger samples.

Attribute data:

  • P chart: records the number of defective parts in a group of parts.
  • C chart: monitors the count of defects in a single product unit where the opportunity area is constant.
  • U chart: records the number of defects in each part when sample sizes vary.

If you are measuring a dimension with calipers, you need a variable chart. If you are counting how many units fail a go/no-go gauge, you need an attribute chart. If you are not sure which data type you have, figure that out before you open Minitab.

How SPC Fits Into Six Sigma DMAIC

Six Sigma was born in an organisation that practised SPC as an ongoing management technique. At Motorola, where Bill Smith and Mikel Harry developed Six Sigma in 1986, inputs and outputs from most processes were monitored using control charts, and capability studies were used to assess quality.

In most organisations today, SPC is confined to the Control phase, the last step of DMAIC. That is a mistake. Wheeler identified this as one of the flaws in most Six Sigma DMAIC approaches: the failure to develop a well-defined baseline of performance in the Define phase leads to problems with quantifying benefits realistically, rework in later phases, and in some cases, project failure.

A quarterly number is not a baseline. An average from a process displaying statistical control is a coherent baseline, and any performance baseline derived without regard to statistical control is suspect, providing a poor basis for project justification.

SPC should start in Define. Put the primary metric on a control chart before you do anything else. That gives you a valid baseline, an operational definition of breakthrough, and a rational basis for the project charter.

In Measure and Analyse, control charts serve a second purpose: checking data homogeneity. The hypothesis tests used in these phases, t-tests, ANOVA, tests of normality, are sometimes completed on data that has not been checked for homogeneity. Results of such tests are irrelevant if the data come from an out-of-control process. A quick run chart before you run a hypothesis test tells you whether the comparison is even valid.

In Control, the chart monitors that gains hold. By the time you reach Control, you should already have charts running on the critical Xs and Ys. The control plan documents what is already in place.

SPC vs SQC: What Practitioners Need to Know

The terms are used interchangeably in many organisations, but they describe different activities. SPC monitors process inputs in real time. SQC checks finished outputs after production.

SPC is your early warning system. SQC is your final verification gate. SPC asks: is my process in control right now? SQC asks: does this finished product meet the specification?

SQC includes acceptance sampling, which SPC does not. Typical SQC tools include lot acceptance sampling plans, skip lot sampling plans, and Military (MIL) Standard sampling plans. These are probabilistic decision tools. You draw a sample from a finished lot, test it against acceptance criteria, and decide whether the lot ships or is held.

Both use the same seven quality control tools compiled by Dr. Kaoru Ishikawa in 1974: cause-and-effect diagram, check sheet, control chart, histogram, Pareto chart, scatter diagram, and stratification. The difference is where you point them. SPC looks at independent variables, the inputs you can adjust. SQC looks at dependent variables, the results. A facility that only runs SQC is constantly playing catch-up. A facility that only runs SPC may miss the downstream verification that a customer or auditor expects to see.

Process Capability: Cp, Cpk, and What the Numbers Mean

Process capability indices tell you whether your process can meet the specification limits, not just whether it is stable. A process can be in perfect statistical control and still produce 30% scrap if the natural variation is wider than the tolerance band.

Cpk reflects short-term capability within subgroups. Ppk reflects overall process performance. A Cpk below 1.33 is a conversation worth having with your team.

Any capability study requires a stable process to be valid. Running a capability study on an out-of-control process produces numbers that look precise but mean nothing. Get the process stable first, then assess capability. The sequence matters.

SimplicityHub's statistical process control page covers the theory, the 14 tools, and how SPC pairs with poka-yoke to close the loop from detection to prevention. The calculators and templates handle the computation so you can focus on what the chart is telling you.

Beyond the Basic Chart: CUSUM and EWMA

Standard Shewhart charts are sensitive to large, sudden shifts. A point outside 3-sigma limits is unambiguous. But they are less effective at detecting small, sustained drifts. A process that shifts by half a sigma per day might run for weeks before a single point crosses the control limit, while producing thousands of marginally off-target parts.

CUSUM (Cumulative Sum) charts solve this. Each plotted point represents the algebraic sum of the previous point and the most recent deviation from the target. They are sensitive to small, sustained shifts that a standard Shewhart chart might miss. Tool wear is the classic use case. A cutting tool dulls gradually, and the part dimension creeps upward by microns per shift. A CUSUM chart catches it days before the Shewhart chart does.

EWMA (Exponentially Weighted Moving Average) charts give more weight to recent process history and decreasing weights for older data. Each plotted point represents the weighted average of current and all previous subgroup values. EWMA charts are useful when you need to be responsive to recent changes without being thrown off by every individual outlier.

Neither replaces the Shewhart chart. They complement it. Use a Shewhart chart for day-to-day monitoring. Add a CUSUM when you suspect a slow drift. Add an EWMA when you want a smoothed view that weights recent data more heavily.

Getting Started: A Practitioner's Checklist

  1. Identify the critical-to-quality characteristics. Which process outputs matter most to the customer? Start there, not with the easy-to-measure ones.
  2. Determine your data type. Continuous variable or attribute? This dictates your chart selection.
  3. Collect Phase I data. Aim for 25 subgroups of 4 or 5 samples each.
  4. Calculate control limits from the data, not from the specification.
  5. Verify stability. Are there points outside the limits? Run rule violations? If so, investigate and remove special causes, then recalculate limits.
  6. Move to Phase II monitoring. Collect data at regular intervals and plot against the established limits.
  7. Act on signals. A control chart that nobody responds to is wallpaper. Every out-of-control point should trigger a documented corrective action.
  8. Assess capability only after stability is confirmed. Running capability on an unstable process wastes everyone's time.

The tools exist to make this straightforward. The same calculators and templates that handle the computation free you to focus on what matters: reading the signals and acting on them before the shift becomes a pile of scrap.