A control chart is only useful if someone reads it and acts on it. In regulated industries, the gap between having a control chart and having a functioning statistical process control program often comes down to exactly that: whether the data on the chart connects to a defined response workflow within your quality management system.
This guide covers the mechanics of building control charts, the rules for reading them, and how to embed them in a QMS so they drive real corrective action rather than sitting as paper records that satisfy an auditor’s checklist once a year.
What a control chart is and what it is not
A control chart is a statistical tool that plots process measurements over time against calculated control limits. It is designed to distinguish between two types of variation: common cause (inherent to the process) and special cause (attributable to a specific, identifiable event or condition).
A control chart is not a specification conformance chart. This confusion causes problems in regulated environments. Specification limits tell you what product must be to pass release. Control limits tell you what your process normally produces. The two sets of limits are calculated differently and serve different purposes. Plotting your data against spec limits tells you whether product passed. Plotting it against statistically derived control limits tells you whether the process is behaving consistently.
Walter Shewhart developed the control chart at Bell Laboratories in the 1920s specifically to address this distinction. His insight — that variation from common causes requires management action on the system, while special cause variation requires investigation of specific events — remains the conceptual foundation of SPC today.
Types of control charts and when to use each
Chart selection depends on two factors: whether the data is continuous (variable) or discrete (attribute), and how many measurements make up each time-ordered data point.
Variable charts for continuous data
X-bar and R chart (Xbar-R): The most widely used chart in manufacturing. The X-bar chart tracks the mean of each subgroup; the R chart tracks the range within each subgroup. Use this when you collect two to ten measurements per subgroup at regular intervals. Common applications in regulated manufacturing include tablet weight checks, fill volume monitoring, and dimensional measurements on machined components. Subgroup size is typically 2 to 5 for most pharmaceutical and device applications.
X-bar and S chart (Xbar-S): Similar to Xbar-R but uses standard deviation instead of range to track within-subgroup variation. More statistically efficient than the R chart when subgroup sizes exceed 8 to 10. Used in high-volume production where larger subgroup samples are practical.
Individuals and moving range chart (I-MR or XmR): Used when only one measurement is possible or practical per time period. The individuals chart (X) tracks each single measurement; the moving range chart (MR) tracks the range between consecutive individual values. This is the most relevant chart for pharmaceutical batch manufacturing, where one analytical result per batch is the norm. It is also used for slow processes where waiting to accumulate a subgroup would mean significant delay before a problem is detected.
Attribute charts for discrete data
p-chart: Tracks the proportion of nonconforming items in a sample when sample size varies. Used for visual inspection of labeled units, packaging integrity checks, and pass/fail acceptance testing where sample sizes change per inspection event.
np-chart: Tracks the number of nonconforming items in a sample of constant size. Simpler to communicate to operators than the p-chart when sample size is fixed.
c-chart: Tracks the count of defects per unit when each unit has the same opportunity for defects. Applied in inspection of device assemblies, printed labels, and packaging where defect counts per unit are meaningful.
u-chart: Tracks defects per unit of inspection area when inspection area varies. Common in film coating inspection, printed circuit board inspection, and surface defect monitoring.
How control limits are calculated
Control limits are set at three standard deviations above and below the process mean — written as the mean plus or minus three sigma (3σ). This threshold is not arbitrary. A process that follows a normal distribution will produce a point beyond three sigma limits by chance alone only 0.27% of the time — fewer than 3 points per 1,000. When you see a point beyond those limits, the probability that it happened by chance is very low, which makes it a credible signal of a special cause.
For Xbar-R charts, control limits are calculated from the process data itself using tabled constants (A2, D3, D4) that adjust for the average range and subgroup size. The formulas are standardized in reference texts including the Automotive Industry Action Group (AIAG) MSA manual and widely referenced quality engineering references.
The critical rule: control limits must be calculated from data collected while the process is in a known state of control. Using data from a period when special causes were present will artificially inflate the estimated process spread, producing control limits that are too wide to detect real problems. Most practitioners use a minimum of 25 subgroups to establish initial trial control limits, then review those limits for points attributable to known special causes before finalizing.
Reading a control chart: the detection rules
A point outside the three-sigma control limits is the most obvious signal, but it is not the only one. Processes can show special cause variation through patterns within the control limits. The most widely used detection rules in regulated industries are based on the Western Electric Statistical Quality Control Handbook rules, often called the Nelson rules or Western Electric rules. Your SPC procedure should specify which rules you apply.
The four most commonly applied rules:
- Rule 1: One point falls outside the three-sigma control limits. The probability of this occurring by chance in a stable process is 0.27%. Immediate investigation is warranted.
- Rule 2: Two of three consecutive points fall beyond the two-sigma zone on the same side of the centerline. This signals a sustained shift in the process mean even though no individual point breaks the three-sigma limit.
- Rule 3: Four of five consecutive points fall beyond the one-sigma zone on the same side of the centerline. Another indicator of a process shift.
- Rule 4: Eight consecutive points fall on the same side of the centerline. Even if all points are well within control limits, eight in a row on one side of the mean is statistically unlikely in a stable process and signals a sustained shift.
Additional rules address trends (six consecutive points steadily increasing or decreasing), cycles, and other patterns that indicate systematic process behavior. The number of rules you apply is a tradeoff between sensitivity to real shifts and the rate of false alarms. More rules increase sensitivity but also increase the frequency of false positives that consume investigation resources.
What to do when a chart signals a problem
This is where many SPC programs in regulated environments break down. The chart signals a problem, and the response is either ignored, informally handled without documentation, or documented so long after the fact that the investigation loses value.
In a properly integrated QMS, a control chart signal triggers a defined response sequence:
- Immediate containment: Depending on the nature of the signal, this may mean placing the affected product or batch on hold, stopping production until the cause is identified, or implementing interim measures to prevent further nonconforming output. The action taken at this step should be documented with a timestamp.
- Deviation initiation: A deviation report is opened to formally record the out-of-control condition, the affected product or process parameters, and the initial assessment of potential impact.
- Root cause investigation: A structured root cause investigation is conducted. Tools commonly used include fishbone (Ishikawa) diagrams, 5-Why analysis, fault tree analysis, and process mapping to identify what changed in the process that led to the special cause signal.
- CAPA determination: Based on the root cause, a corrective and preventive action is determined and implemented. Corrective action addresses the specific instance; preventive action addresses whether other processes or products face the same vulnerability.
- Effectiveness verification: After the CAPA is implemented, return to the control chart and confirm that the process has returned to a stable state. The control chart data from after implementation is the primary evidence of CAPA effectiveness for this type of quality event.
All steps must be captured with a complete audit trail showing who took each action, when, and based on what information. During an FDA inspection or ISO 13485 audit, this chain of evidence from chart signal to closed CAPA is exactly what investigators look for.
Setting up control charts in a regulated QMS: practical considerations
Document your chart parameters in an SPC procedure
Every controlled SPC chart in a regulated environment should be backed by a procedure or work instruction that specifies: the parameter being monitored, the chart type, the subgroup size and sampling frequency, the method for calculating control limits, the detection rules applied, the required response for each rule violation, and the frequency for reviewing and updating control limits. Without this documentation, your SPC program is not defensible in an audit, even if the charts themselves look correct.
Separate control limits from specification limits on the chart
If you plot both control limits and specification limits on the same chart, clearly distinguish them visually and label them unambiguously. An auditor seeing a chart where control limits and spec limits are indistinguishable — or worse, where they appear to be the same — will flag it as a documentation deficiency. Many organizations choose to maintain separate charts for process monitoring (control limits only) and for conformance to specification, which avoids the confusion entirely.
Review and recalculate control limits at defined intervals
Control limits based on data from six months ago may no longer reflect current process reality, particularly if process improvements, equipment changes, or raw material changes have occurred. Your procedure should specify when control limits are recalculated — at minimum after any validated process change, and on a defined periodic schedule such as annually or semi-annually. Document the rationale for any limit revision.
Train operators and reviewers on chart interpretation
A control chart reviewed only by the quality department has limited value. Operators who understand what the chart is telling them can identify special causes at the source before product is affected. Training records documenting that operators understand how to read the chart and what to do when a rule violation occurs are part of the quality system record.
Common mistakes in regulated-industry control charts
Calculating control limits from mixed data: If the dataset used to establish control limits includes periods when special causes were present and uncorrected, the resulting limits will be too wide. Purge known special cause data points before finalizing control limits, with documentation explaining each removed point.
Applying control charts to non-normal data without adjustment: The standard control chart formulas assume normally distributed data. Measurements that are inherently skewed — certain analytical results, count data — may require transformation or a different chart type to produce valid control limits. Applying standard X-bar or I-MR charts to heavily skewed data generates false alarm rates much higher than the expected 0.27%.
Updating control limits too frequently: Recalculating limits after every few data points defeats the purpose of control charts. Limits should reflect a stable baseline period. Frequent recalculation disguises process drift by constantly adjusting to it.
Not connecting the chart to the QMS workflow: A control chart that generates signals with no defined response process is a compliance liability. The analytical report documenting control chart review must show that signals were evaluated and responded to, not just that charts exist.
Control charts and eQMS integration
In paper-based or spreadsheet-driven SPC programs, the connection between a control chart signal and the deviation/CAPA system relies entirely on a person manually initiating the next step. That handoff fails more often than it should, and when it does, the failure is documented in the audit trail by its absence — there is no deviation record corresponding to the out-of-control event the chart clearly shows.
An electronic quality management system that integrates control chart monitoring with the deviation, CAPA, and management review workflows eliminates that manual handoff. Cloudtheapp’s platform connects process monitoring data directly to deviation initiation and CAPA management across its 60+ quality applications, with a validated audit trail that captures every event from chart signal to corrective action closure. The management review module can pull SPC trending data directly, giving quality leaders the process performance visibility they need without assembling data from multiple disconnected sources.
To see how Cloudtheapp connects statistical process monitoring to your full QMS workflow, schedule a demo.
Conclusion
Control charts work when three things are in place: the chart is built correctly from valid baseline data, the detection rules are defined and consistently applied, and every signal triggers a documented response within the QMS. In regulated industries, all three are required. A chart that meets two out of three — even if the statistics are perfect — is still a compliance gap. Building that third element into your quality system, not as an afterthought but as a designed workflow, is what separates a functioning SPC program from one that exists only to satisfy a checkbox.
]]>
