After separating absolute repeatability from the percentage denominator, the robot-loaded CMM result becomes explainable: the repeated flatness values are close, but the ten selected parts are also close enough that measurement variation still occupies 14.51% of the reported study variation. Treat this as a measurement-system study design and interpretation problem before treating it as a calibration failure.
Stop applying the usual quick fixes
Do not recalibrate a new CMM solely because analysis software prints “Your measuring equipment may need calibration or maintenance.” That message is a generic response to a statistical ratio. Calibration establishes traceability and geometric performance under stated conditions; it does not force a GR&R percentage lower when the sampled parts have little part-to-part variation.
| Quick fix or conclusion | Why it fails here | What to check instead |
|---|---|---|
| Calibrate the CMM again | The installation was completed the previous day, and probe validation was run after each ten-piece set. Neither action changes the study denominator. | Compare absolute repeatability with the specification for the actual CMM configuration and measurement program. |
| Accept the result because the three runs look close | GR&R is relative. Small repeat errors can produce a noticeable percentage when part-to-part spread is also small. | Review absolute repeatability, part variation, percentage of study variation, percentage contribution, and percentage of tolerance separately. |
| Try more spreadsheet templates | Several programs produced similar results. Recalculation cannot correct a narrow or poorly structured sample. | Confirm the statistical method, data layout, units, part mapping, and meaning of each reported percentage. |
| Use probe validation as proof of the complete process | Probe validation does not exercise robotic pickup, seating, fixturing, part cleanliness, thermal effects, alignment, scan path, filtering, or the flatness calculation. | Repeat the complete unload-load-measure cycle. |
Compare this result with a different feature's 0.39%
|
A feature with larger absolute variation can have a lower percentage if its part spread or tolerance denominator is much larger. | Compare absolute measurement error and denominators, not percentages alone. |
Read the 14.51% as a ratio
The reported 14.51% does not mean that the CMM is wrong by 14.51% of each flatness value. For a percentage-of-study-variation result, the governing relationship is:
%GR&R = 100 × measurement-system standard deviation / total study standard deviation
The ten parts span roughly 0.0189 mm to 0.0281 mm. Their part-to-part standard deviation is about 0.0025 mm, or 2.5 µm. The standard deviations of the three readings for each part average about 0.00035 mm, or 0.35 µm. A measurement component near 0.35 µm compared with a part component near 2.5 µm naturally produces a result in the neighborhood of the reported 14.51%; the exact value depends on the ANOVA model and variance-component calculation.
Check the output label before judging it. Percentage of study variation uses standard deviations. Percentage contribution uses variances, so it is approximately the square of the study-variation fraction. For example, 14.51% of study variation corresponds to about 2.11% variance contribution if both values come from the same model. Percentage of tolerance uses the drawing tolerance as its denominator and answers a different question.
The reported equipment variation of 11.27% is not added arithmetically to other GR&R components. ANOVA combines variance components and then converts them back to standard-deviation terms. Read the software's definitions and selected method before comparing fields.
Use the run data to locate the real signal
| Check | Observed result | Engineering meaning |
|---|---|---|
| Run A mean | 0.02329 mm |
Highest run mean |
| Run B mean | 0.02297 mm |
0.00032 mm below Run A |
| Run C mean | 0.02279 mm |
0.00050 mm below Run A |
| Largest within-part range |
0.0012 mm on part 6 |
Inspect this loading, seating, scan, and calculation record first |
| Next-largest within-part range |
0.0010 mm on part 10 |
Second priority for trace review |
| Overall observed range | 0.0092 mm |
The selected sample has a limited denominator |
The run means descend from A through C. The total shift is only 0.50 µm, but its direction matters because all ten parts were measured in the same order. Time, temperature, contamination, settling, probe state, and order are therefore confounded. A repeatable downward run shift can contribute variation even when each group looks visually tight.
Plot every part across A, B, and C. Parallel downward traces point toward a run-level bias. One or two isolated traces point toward part seating, robotic gripping, contamination, fixture contact, or local scan behavior. Irregular changes across many parts point toward general repeatability rather than a single run offset.
Treat the robot as part of the measurement system
Removing the human appraiser removes human-to-human reproducibility from the production cycle; it does not remove measurement-system variation. The robot, gripper, fixture, CMM, probe, measurement program, environmental state, scan strategy, data filtering, alignment, and flatness algorithm form one automated measurement system.
A one-robot, one-program study estimates repeatability of that system across repeated load-and-measure cycles. It cannot estimate differences between human appraisers because none exist in the design. If production uses multiple robot cells, fixtures, probes, programs, or CMMs, those are potential reproducibility factors and must be crossed or nested deliberately in the study.
Flatness is especially sensitive to how the surface is represented. Changes in scan coverage, point rejection, filtering, datum treatment, alignment, surface contamination, clamping, and part support can alter the calculated extreme separation even when the machine coordinates repeat closely. Probe validation tests the probing configuration; it does not validate all of these influences.
Rebuild the study around production conditions
- Identify the exact response being analyzed. Record the drawing characteristic, flatness tolerance, units, CMM program revision, scan strategy, filtering, alignment, fixture, gripper, probe configuration, and environmental records.
- Select parts that represent the process range the measurement system must discriminate. Include normal low, middle, and high process output rather than ten nearly interchangeable pieces. Do not introduce deliberately defective parts unless the study procedure permits them and they remain representative of the measurement task.
- Keep permanent part identities. Verify that every row in the analysis maps Run A, B, and C to the same physical part.
- Run the complete production cycle for every replicate: robot pickup, placement, seating, scan, removal, and storage. A rescan without unloading tests the CMM program but omits the loading system.
- Randomize part order between runs when the process allows it. If the automated sequence must remain fixed, log time and order and use a stable check part to separate drift from part identity.
- Inspect and clean the fixture contacts and relevant part surfaces using the approved production method. Record corrective cleaning; unrecorded cleaning can hide the source of variation.
- Retain the raw point data and calculated result for parts 6 and 10. Compare seating, scan coverage, rejected points, alignment, and flatness construction across their three cycles.
- Analyze the study as an automated repeatability design unless a second appraiser-equivalent factor is actually included. Report absolute repeatability in millimetres or micrometres alongside the statistical percentages.
- Compare the absolute result with the documented performance of the installed CMM configuration and with the measurement requirement for this flatness characteristic. Do not substitute the capability demonstrated on another feature with a maximum tolerance of
0.013 mm.
Separate equipment variation from sample selection
Run two controlled checks without confusing their purposes. First, repeatedly scan a stable part without unloading it. This isolates short-term CMM, probe, program, and calculation repeatability. Second, repeat the complete robotic unload-load-measure cycle. The increase between those tests identifies the contribution associated with handling, placement, seating, and fixture interaction.
Next, inspect the denominator. If the production process genuinely makes parts within a very narrow flatness band, a part-based percentage may remain elevated even when absolute repeatability is suitable for the drawing decision. In that case, report both the process-based result and percentage of tolerance, using the actual flatness tolerance and the organization's approved multiplier and analysis method. A narrow process spread is not permission to change the formula or substitute another feature's tolerance.
Use the actual installed CMM specification, not a typical machine rating. CMM performance depends on configuration and stated test conditions. Read the acceptance report, calibration documentation, probing specification, and applicable performance table for the installed system, then compare like quantities and units.
Verify the correction before releasing production
- Repeat the study with representative parts and randomized order, while preserving full robotic loading between measurements.
- Confirm that the analysis identifies ten parts and three repeated measurements per part, all in millimetres.
- Check that run means no longer show an unexplained monotonic shift. If a shift remains, correlate it with temperature, elapsed time, probe checks, cleaning, fixture state, and robot placement records.
- Confirm that parts 6 and 10 no longer dominate the within-part ranges, or document the physical reason if they do.
- Review absolute equipment variation, total GR&R percentage of study variation, variance contribution, and percentage of tolerance as separate outputs.
- Apply the acceptance rules required by the plant, customer, or measurement-system procedure. Do not create a pass limit from the software's generic maintenance message.
Get production moving only after the measurement system can distinguish the decisions required by the drawing. Then preserve the program, fixture condition, robot placement settings, cleaning method, environmental limits, and study setup as the controlled baseline.
FAQ
What happens if a robot removes the human appraiser?
Human-to-human reproducibility drops out of the design, but repeatability remains. The robot, gripper, fixture, probe, CMM, program, scan processing, and environment still contribute variation.
What happens if the GR&R parts are too similar?
The part-to-part denominator becomes small, so even 0.35 µm average within-part standard deviation can produce a noticeable percentage against about 2.5 µm part variation. Select parts spanning the normal process range and retain the absolute repeatability result.
What happens if probe validation passes but GR&R stays high?
Test the full unload-load-measure cycle against repeated scans without unloading. The difference exposes robotic placement, seating, fixture, contamination, or handling effects that probe validation does not test.
What happens if every run uses the same part order?
Part identity becomes confounded with time and sequence. Randomize later runs when possible; otherwise log the fixed sequence and insert a stable check part to detect drift.
What happens if the absolute variation exceeds the installed CMM specification?
Stop changing templates or accepting parts from the suspect measurement process. Preserve the program, raw scan data, calibration and acceptance records, probe-validation results, fixture details, environmental log, and robotic loading sequence. Stop here and escalate through Hexagon's official support channel when repeat scans of a stable part still exceed the applicable installed-system specification or when a run-level drift cannot be tied to the fixture, environment, program, or loading cycle.