A gage stability study tracks one master part measured on the same gage over days or weeks, plots the results on a control chart, and watches for drift or shifts that mean the gage's error is changing with time. Stability is the change in bias over time, and the control chart is how you catch it before it corrupts a decision.
Gage R&R and a bias study are snapshots, they tell you the gage is good today. Stability is the movie. A micrometer that passed every study last quarter can pick up wear, its zero can creep, a fixture can loosen, and none of that announces itself. It shows up as a slow slide in what the gage reads on a part that has not changed. If you are not watching, the first sign is a customer complaint or a pile of scrap. This guide covers what a stability study is, why gages drift, how to run one with a control chart, and how it fits alongside gage R&R and the rest of measurement systems analysis.
What is a gage stability study?
A gage stability study is a periodic check: you keep one stable master part aside, measure it with the working gage on a set schedule, daily or weekly, and plot those readings on a control chart in time order. Because the part itself does not change, any real movement on the chart is the measurement system, not the part. Stability, in the measurement-systems sense, is the amount of change in the gage's bias over time. A stable gage stays in statistical control around one level; an unstable gage drifts, jumps, or wanders.
This is different from mechanical stability of the part or from the everyday sense of "steady." It is a specific statistical idea: the measurement process is in control with respect to location. If you took the same master and measured it this month, last month, and next month, a stable system would give you readings that look like the same distribution every time. The control chart is the tool that tells you whether that is true, using the same logic as any process control chart, a center line, control limits, and out-of-control rules.
Why does a gage drift over time?
Gages drift because they are physical things that wear and shift. Contact surfaces on a micrometer or snap gage wear down with use, slowly changing the zero. A dial indicator's spring and gear train fatigue. A fixture that locates the part works loose a few thousandths. Electronic gages have their own version: sensor response ages, and thermal cycling nudges the electronics. None of these are dramatic on any single day, which is exactly why they are dangerous, the change is small enough to hide inside normal noise until it accumulates.
Environment adds to it. Temperature is the big one: steel expands and contracts, and a gage or master that spends the afternoon near a furnace reads differently than it did in the cool morning, so what looks like drift can be a shop that swings ten degrees between shifts. Handling matters too, a gage that gets dropped, or a master that picks up a burr or a film of oil, changes without warning. A stability study catches all of it the same way, by watching one unchanging part and letting the chart flag when the readings stop behaving.
How do you run a gage stability study?
You need a master part that will not change, a schedule you can actually keep, and a control chart. The AIAG method is to measure a master several times on each occasion and plot the subgroups over a stretch of time. Here is the procedure:
- Pick a stable master part. Choose a part near the middle of the measuring range that will not wear, corrode, or change, ideally one with a reference value traceable to a standard. Store it protected so the only thing that moves is the gage, not the master.
- Set a measurement schedule. Decide how often you will measure, commonly daily or at each shift start, based on how heavily the gage is used and how much a wrong reading would cost. Heavier use and tighter tolerances mean more frequent checks.
- Take a small subgroup each time. On each occasion, measure the master 3 to 5 times and record every reading. The repeats within an occasion capture the short-term noise; the occasions capture the drift.
- Collect enough subgroups to set limits. Gather roughly 20 or more subgroups before you fix the control limits, so the limits reflect real variation and not a lucky first week.
- Plot an average-and-range chart in time order. Put the subgroup averages on an X-bar chart and the ranges on an R chart, both in the order the readings were taken, and calculate the center line and control limits from the data.
- Apply the out-of-control rules. Watch for points beyond a control limit, long runs on one side of the center line, and steady trends. Any of these says the gage is no longer stable.
- Act on a signal, then keep charting. When the chart signals, stop and investigate, recalibrate, service, or replace, then reset and continue. Stability is not a one-time pass; the chart runs as long as the gage is in service.
| Study element | Typical practice | Why it matters |
|---|---|---|
| Master part | Stable, mid-range, protected | Any movement on the chart is the gage, not the part |
| Frequency | Daily or per shift, by risk | Catches drift before it reaches production |
| Readings per occasion | 3 to 5 | Separates short-term noise from long-term drift |
| Subgroups before limits | ~20 or more | Gives control limits that reflect real variation |
| Chart | X-bar and R, in time order | Reveals trends and shifts the eye would miss |
By the numbers. The AIAG Measurement Systems Analysis reference manual defines stability as the change in bias over time and treats a stable measurement process as one that is in statistical control with respect to location, the same standard a production control chart uses (AIAG, Measurement Systems Analysis). AIAG's stability method is to measure a reference master periodically, commonly three to five readings per occasion across roughly twenty or more occasions, and evaluate the results on an average-and-range chart. The control-chart mechanics, how the center line and limits are calculated and which out-of-control patterns count as signals, are documented in the NIST/SEMATECH engineering statistics handbook's process monitoring section (NIST/SEMATECH e-Handbook). Both are recognized primary references; the exact frequency and subgroup size are set by how the gage is used, not by a single universal number.
How do you read the stability control chart?
Read it the way you read any control chart. Points hugging the center line inside the control limits mean the gage is stable, its bias is holding steady, and you can trust it. A single point past a control limit is a signal the gage jumped. A run of seven or more points on one side of the center line, or a steady climb or fall across several occasions, is drift, the gage is slowly changing even though no single point has broken a limit yet. That slow drift is the whole reason you chart in time order rather than just averaging everything together, because an average would smear the trend away.
The R chart matters as much as the X-bar chart. If the ranges within each occasion suddenly widen, the gage's short-term repeatability is getting worse, a loose fixture, a sticking spindle, which is a different failure from a slow shift in the average. A stable gage keeps both charts quiet. When either one signals, the gage is telling you something changed, and the chart even hints at what: a jump in the average points at bias, a jump in the range points at repeatability.
How often should you check gage stability?
Base the frequency on risk and use. A gage that runs a tight, safety-critical dimension all shift should be checked at the start of every shift or every day. A gage that measures a loose feature a few times a week can be checked weekly. The question to ask is: how many parts would ship on a bad gage before the next check catches it, and can you live with that many. The tighter the tolerance and the higher the volume, the shorter the interval.
Tie the schedule to your calibration program, but do not confuse the two. Calibration verifies the gage against a standard on a fixed cycle, often months apart, while a stability check is the cheap, frequent watch in between that tells you if the gage went out of true before the calibration due date. Many plants fold the stability reading into the start-of-shift routine so it costs almost nothing and gives an early warning that a calibration alone would miss.
Where does stability fit in the wider quality system?
Stability is the maintenance layer of measurement systems analysis. A bias and linearity study establishes that the gage is accurate now, a gage R&R establishes that it is precise now, and the stability study is what keeps both of those true tomorrow. Without it, every other MSA result has a shelf life you cannot see. It is the difference between "the gage was good in March" and "the gage is good today," and only the second one protects the parts you are shipping this shift.
Because a stable gage is a precondition for trustworthy data, stability sits upstream of statistical process control and control charts on the product itself. If your production control chart signals, the first question is whether the process moved or the gage did, and a running gage stability chart answers it in seconds instead of sending you chasing a phantom process shift. It is the same discipline that shows up in a first article inspection: the measurement has to be trustworthy before the numbers on the report mean anything.
The hard part of stability is not the statistics; it is remembering to take the reading and actually seeing the trend before it becomes a problem. That is where live data helps. When each stability check is logged at the point of inspection and charted automatically, a gage that is starting to drift shows up as a rising run on the screen while it is still inside the limits, early enough to service the gage during planned time instead of discovering it after a reject. That live feedback is what Harmony gives a plant, turning a stack of clipboard readings nobody plots into a chart the floor can actually watch. CLS made that shift, from measurements found the next morning to measurements visible during the shift, which is exactly what a stability study needs to work.