Control Chart Fluctuations Increase, Don't Adjust Parameters Yet — A Five-Step Method to Separate Process and Measurement Variations

In a machining company, the control chart for the cylinder bore diameter, which was previously stable, suddenly started fluctuating up and down last week, with a few points nearly touching the control limits. The process engineer's first reaction was that the cutting tool or cutting parameters were at fault, so they re-calibrated the tool, adjusted the feed rate, and replaced the cutting insert. After three days of effort, the fluctuations on the chart remained unchanged. It turned out that the line had switched to a new internal diameter gauge, and the new gauge had a repeatability error of a few micrometers.

This is the most common type of misjudgment in the field. When the chart alarms, most people assume that only the process has changed, forgetting that each number on the control chart is the result of both the "part and the measurement system." The measurement system itself can also fluctuate, and its variations will be fully reflected in the chart.

1. Fluctuations on the Control Chart Always Have More Than One Source

The variation in a single measurement can be broken down into two parts: the true difference in the parts (process variation) and the difference introduced by the measurement system (gauge, fixture, measurement method, and operator reading habits). These two parts are not simply added together but are combined using the root sum square method. This means that when measurement variation is small, it doesn't matter much, but once it increases to the same magnitude as the process variation, the total variation will significantly increase, creating a false impression on the control chart that "the process has deteriorated."

MSA (Measurement System Analysis) serves one simple purpose: to isolate the portion of variation attributable to the measurement system, leaving the rest to the process. The industry-standard criterion is that if the GR&R (Gage Repeatability and Reproducibility) ratio to the total process variation or tolerance is within 10%, it is acceptable. A ratio between 10% and 30% may require improvement based on the importance and cost of the gauge, while a ratio exceeding 30% must first address the measurement system. In other words, if the ratio exceeds 30%, about half of the fluctuations on the control chart are due to the measurement system itself. Adjusting the process in such a situation is like trying to straighten your tie in a funhouse mirror.

2. Four Signals Indicating "The Problem is on the Measurement Side"

Signal One: Repeated measurements of the same part show significant scatter. Take a stable part (preferably a standard part or a retained sample), and have the same person measure it ten times with the same gauge. If the range of these ten measurements is comparable to the range of the subgroups on the chart, the measurement system is likely the culprit. This is the least expensive verification, which can be completed in five minutes.

Signal Two: Fluctuations follow shifts or operators. Stratify the chart by shift or operator. If the variation is stable within each stratum but there are clear steps between strata, it is a reproducibility issue—differences in reading habits, measurement point selection, and clamping force, which are unrelated to the process.

Signal Three: Fluctuations follow the gauge or method. If the chart starts to fluctuate after events such as purchasing a new gauge, returning a gauge from repair, changing the inspection tool, upgrading software, or altering the sampling frequency, and the timing aligns, suspect a change in the measurement system first.

Signal Four: Process-side characteristics do not match. Tool wear issues typically show a gradual, one-directional drift, not a sudden increase. Material batch issues concentrate on a specific batch when stratified. Machine-specific issues become apparent when stratified by machine. If none of these characteristics match, and the chart just becomes "noisier" overall, the likelihood of a measurement system issue increases.

3. Five Steps to Complete the Separation

Step One: Identify the issue before taking action. Confirm which out-of-control rule has been triggered—single point out of bounds, multiple consecutive points on the same side, continuous rising or falling, or a significant increase in variation. At the same time, rule out the most basic errors: incorrect data entry, unit conversion errors, missing or duplicate data points in automatic data collection, and decimal point shifts when entering data. These errors are not uncommon in the field and can be fixed with a single correction.

Step Two: Stratify the control chart. Draw stratified control charts by shift, operator, gauge, machine, and batch. If a single stratum is normal when isolated, the issue is likely with the corresponding factor in that stratum. Stratification is the most cost-effective step in the separation process because it requires no additional testing, only reorganizing existing data.

Step Three: Conduct a targeted MSA. Use the same batch of samples to test repeatability (multiple measurements by the same person with the same gauge) and reproducibility (cross-measurements by multiple people with multiple gauges). In destructive testing or small batch scenarios, nested designs or short-term repeatability can be used as alternatives. Simultaneously, verify the stability of the gauge: measure a standard part at regular intervals over a period of time to check for systematic drift. Many "new gauges causing fluctuations" issues stem from this, not from the gauge itself being faulty.

Step Four: Quantify and allocate, then decide which side to address. Use GR&R or ANOVA to calculate the proportion of measurement variation to process variation. If the proportion is high, all corrective actions should focus on the measurement side: repair or replace the gauge, standardize and document the measurement method (measurement point, clamping method, number of measurements, whether to take the average), retrain and re-evaluate personnel, and add auxiliary fixtures or switch to more suitable gauges if necessary. If the proportion is low, focus on the process side: equipment, tools, process parameters, and incoming inspection. The action lists for these two paths have almost no overlap, and choosing the wrong direction is a waste of effort.

Step Five: Recalculate control limits after changes and incorporate the changes into management. Once the measurement system is changed (new gauge, new inspection tool, new measurement method, new inspection software), the original control limits are no longer applicable and must be recalculated using data from the new conditions. Additionally, include "measurement system changes" in the control chart maintenance trigger conditions: evaluate upon change, decide whether to recalculate control limits, and annotate the change points on the chart. Failing to do this can lead to two consequences: wide old control limits masking real issues (false negatives) or narrow control limits with measurement errors causing frequent alarms (false positives).

4. Four Common Pitfalls

Calibration does not equal a qualified measurement system. Calibration only answers whether the gauge's indication error is within the allowable range, not whether the measurement system can distinguish the process variations you want to detect. Insufficient resolution, poor repeatability, and inconsistent readings by different operators are all unrelated to the calibration certificate.

A single GR&R pass does not ensure long-term reliability. Measurement systems can change over time due to usage, wear of clamping parts, and personnel changes. For critical gauges, the MSA re-evaluation cycle should be included in the plan, rather than just being done before PPAP or customer audits.

Calculating process capability with measurement errors. Using data with significant measurement variation to calculate Cpk will underestimate the capability, leading to unnecessary process modifications. Conversely, if sampling inspection only takes single values instead of averages, the impact of measurement variation will be even greater. This is one of the sources of the discrepancy between the two methods of calculating similar data.

Subgroup division is decoupled from the measurement system. Rational subgroups should only contain variations caused by common causes. If different gauges or operators' data are mixed within the same subgroup, the within-subgroup variation will include reproducibility, skewing the control chart's judgment criteria from the start.

5. One Sentence Summary

When a control chart alarms, first prove that the "ruler" is stable before suspecting the measured object; without isolating measurement variation, any process adjustment cannot be verified.

The control chart cannot answer "who is changing." First, conduct an MSA to eliminate the measurement variation, and then the remaining variation can be considered the process voice.

Knowledge code: 6.2.1

Version: v20260929

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