Control Limits Set Once and Never Changed? — A Five-Step Method for Establishing, Inspecting, and Updating Control Limits

During a process audit at an automotive parts company, an auditor pointed to the X̄–R control chart hanging on the wall and asked, "Where does this line come from?" The accompanying process engineer replied, "The system calculates it automatically." The auditor then asked, "When was the last time it was recalculated?" The engineer flipped through the records but couldn't provide an answer. In the same month, the plant manager of another company shared his experience: whenever the control chart showed an anomaly, he would have the inspector recalculate the control limits, and "the chart would immediately return to normal."

These two practices seem contradictory, but they are actually two sides of the same issue—control limits are treated as decorative lines on the chart rather than parameters that need to be defined, reviewed, and controlled for changes.

1. Clarify Three Distinct Concepts

Control limits, specification limits, and "historical maximum and minimum values" are three entirely different things.

Control limits are derived from the inherent variability of the process, calculated from the actual measurement data of the process, and answer the question, "To what extent can this process be stable?" Specification limits come from drawings, customer requirements, or technical agreements and define the allowable range for the product. The former describes the process, while the latter sets the threshold. There is no direct conversion between the two; they can only be linked through capability indices: the same control limits can be deemed capable or incapable depending on the tolerance.

Therefore, directly plotting specification limits on a control chart is a common mistake—once the chart takes on the form of "only crossing the line counts as an issue," the control chart degenerates into a mere out-of-tolerance alarm. It remains silent until the process has already deviated. Conversely, using the maximum and minimum values of actual measurements as control limits is equivalent to letting the control limits automatically align with the tails of the current distribution, making the chart increasingly quiet until it stops alarming altogether.

Another aspect of control limits is that they determine not just the appearance of the chart but also the effectiveness of the monitoring system. Loosening the limits reduces false alarms but increases missed alarms; tightening the limits leads to frequent false alarms, and the chart will soon be abandoned by the field. Control limits are a scale, not just a warning line for aesthetics.

2. Initial Establishment: Sufficient and Clean Data

Step 1: Select a baseline data segment. The baseline must be a segment of process data that has been confirmed as stable: after the trial run, after the first article inspection is passed, and during a period of several consecutive days without confirmed anomalies. The sample size should be at least 25 subgroups (for individual value charts, at least 100 individual values). If fewer than 25 subgroups are used, the estimated standard deviation itself is unstable, and the control limits will be "set" by the first batch of data, making it difficult to correct even with more data points later.

Step 2: Conduct a measurement system analysis first. Control limits include three types of variability: process true variability, measurement tool variability, and sampling variability. When the measurement tool variation is high (typically over 30%), the control limits are inflated by the measurement tool noise, which is like setting a red line with an inaccurate ruler. The sequence should always be to confirm the measurement system is usable before calculating the control limits.

Step 3: Use the correct estimation method. For X̄–R charts, use R̄/d₂; for X̄–s charts, use s̄/c₄; for individual value charts, use the average of the moving ranges divided by d₂. These methods reflect within-subgroup (short-term) variability. Do not use the standard deviation of the entire data segment, as it includes process drift, which will significantly widen the control limits. If the subgroup size changes, the formula's n must also change, and the control limits must be recalculated.

Step 4: Document the removal of outliers. Points in the baseline segment that have been identified and determined to be sporadic anomalies can be removed before calculation, but each removal must be recorded with the reason and basis, and only points that can be traced to a specific cause should be removed. Repeatedly removing points to reduce the standard deviation is one of the main sources of control limit distortion.

Step 5: Freeze the baseline. Write the data interval, sample size, mean, standard deviation, control limits, estimation method, and measurement tool information on a baseline card and bind it to the control chart version number. This card serves as the basis for all subsequent change judgments.

3. When Recalculation is Mandatory

Control limits must be recalculated when the process undergoes substantial changes. Common triggers include: changes in equipment, tooling, molds, material grades, or suppliers; adjustments in process parameters, cycle times, or line layouts, as well as changes in measurement tools or methods (from offline to online, from calipers to vision inspection); events that alter the variability structure, such as major equipment repairs or relocations; and adjustments in sampling frequency or subgroup size.

Another often overlooked but crucial condition is when improvement projects are verified to be effective and result in substantial improvements in process centering or variability. If the old control limits are still used, the chart will show long-term consecutive single-sided points or continuous out-of-limits, leading the field to either investigate the improvements as new anomalies or be forced to relax the anomaly rules. Both outcomes will gradually erode the newly established improvements. Conversely, if the process clearly deteriorates but the old limits are still used, the chart will remain "in control" until a customer complaint arises.

4. When Recalculation is Not Appropriate

First, and most importantly: when an anomaly occurs. The meaning of a control chart alarm is "this process may have an additional cause that needs to be identified." The correct action is to re-measure, check the material batch and parameters, and handle it according to the response rules, not to move the control limits. Recalculating the limits immediately after an anomaly is essentially like dismantling the alarm—this is the most common way control charts fail, and it is often not malicious but rather an attempt to "make the chart cleaner."

Second, recalculating to reduce false signals. The root cause of frequent false signals is usually elsewhere: unreasonable subgroup divisions, insufficient measurement tool resolution, autocorrelation in the data, overuse of anomaly rules, or incorrect estimation of control limits. Without addressing the root cause, recalculating the limits multiple times only pushes the problem from one period to the next.

Third, "cleaning up the chart" before a customer audit or annual review. This erases months of process records and is essentially data manipulation, with costs far exceeding the original recorded anomalies.

Fourth, when the process has a long-term single-sided drift and the specific cause cannot be found. At this point, the process should be analyzed as a capability degradation issue, and necessary adjustments to the process or equipment should be made. The new control limits should be recalculated based on the stable data of the new process, not simply widened.

5. Make Recalculation a Controlled Activity

Recalculation is a parameter change and should have an application, basis, approval, record, and synchronization.

The control limits on the control chart header, the baseline in the SPC system, the monitoring parameters referenced in the control plan, and the anomaly explanations in the work instructions must all be consistent. A common deviation seen during audits is that the system has been recalculated three times, but the chart hanging on the wall is still the version from two years ago. Alternatively, the system may default to "automatically recalculate with each new point," causing the control limits to drift with the process, gradually losing their alarm capability. The latter situation is particularly worth checking, as many SPC software options for recalculation are set to default.

Handling automated recalculations can be simplified into two rules: once the baseline is confirmed, it should be frozen, and only approved changes can modify it. Recalculation permissions should be restricted to the quality engineer level, and the field should not perform recalculations without authorization. A record form should be provided for each change, documenting the old limits, new limits, effective date, change reason, data interval, and approver.

It is recommended to review the control limits quarterly or when a change event occurs: count the number of alarms on the statistical control chart and the proportion of confirmed true anomalies. If alarms are consistently zero or almost all false signals, it indicates that the control limits no longer reflect the process state and need to be reassessed.

6. An Example

In a plastic injection molding workshop, a key dimension was monitored using an X̄–R chart. The control limits were established using 30 subgroups of data from the trial run phase and were not reviewed for the next two years. During these two years, the mold was changed once, the raw material supplier was changed once, and the sampling frequency was adjusted from every two hours to every four hours.

Upon reviewing, two opposite deviations were discovered: after the mold change, the process center shifted by about 0.8σ, but the control limits were not rebuilt. Most points on the chart remained within the old limits, and anomalies were deemed controlled. After a minor improvement, the process variability significantly narrowed, but the old control limits were still used, leading to multiple consecutive points on one side of the center line, which were investigated as anomalies for two weeks.

The corrective actions were not complex: a process change trigger list was established, listing mold changes, material changes, measurement tool changes, and frequency adjustments as mandatory review items, and a record form was provided for each recalculation. One month later, the control charts for both lines resumed their normal alarm rhythm—neither silent nor constantly crying "wolf."

Control limits manage the baseline, while anomalies manage the causes—the actions for these two cannot be interchanged.

Knowledge code: 6.3.1

Version: v20260919

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping companies continuously improve their quality capabilities.