Measurement System Analysis, Repeatability and Reproducibility Exercises

These exercises practise measurement system analysis as a process decision. The goal is to decide whether variation comes from the parts being measured or from the measurement system itself: equipment repeatability, operator reproducibility, fixture method, bias, stability, linearity and resolution.

Assume simplified MSA formulas unless an exercise states otherwise. Real studies should define part selection, operator training, randomization, environmental control, appraiser instructions, instrument calibration, fixture condition, acceptance criteria and the consequence of false acceptance or false rejection.

How to Use These Exercises

For each problem, separate part-to-part variation from measurement variation. State whether the decision uses tolerance, process variation, control limits or release risk. A measurement system can be calibrated and still be poor for production decisions if repeatability or reproducibility is too large.

Release Evidence Notes

An MSA study should be designed around the decision the measurement system must support. A system used for final acceptance against tolerance needs different evidence from a system used for process monitoring, setup adjustment, troubleshooting or rough sorting. The study should state how parts were selected, how operators were trained, whether trials were randomized, whether fixtures were reset between readings and whether the sampled parts represent the real process spread.

Repeatability and reproducibility results are not merely pass/fail labels. High repeatability error points toward instrument, fixture, contact, resolution, environment or method instability. High reproducibility error points toward appraiser technique, ambiguous work instructions, visibility, force, alignment, interpretation or data-entry differences. The corrective action should match the variance source.

When MSA fails, publishing more inspection results does not solve the problem. The organization may need a better fixture, clearer method, improved resolution, automated capture, operator retraining, restricted use, wider decision guard bands or a different measurement technology. If product has already been accepted with a weak system, impact review should consider false acceptance and false rejection risk.

A good MSA closeout records the allowed use of the measurement system. It may be approved for trend monitoring but not final acceptance, approved for one range but not another, approved for trained operators only, or approved only with a specific fixture and work instruction. Without that boundary, a conditional MSA result can be misused as a general approval.

Engineering Boundary Notes

These exercises use compact MSA calculations. They do not replace a full crossed, nested, destructive, attribute or stability study when the measurement process requires one. They also do not prove product conformity by themselves; they only assess whether the measurement system is fit to support the intended decision.

Common Release Mistakes

• selecting parts with too little process spread and then approving a weak system;
• mixing repeatability, reproducibility, bias and stability into one vague error label;
• approving a gauge for final acceptance when it only supports rough sorting;
• ignoring fixture resets, operator sequence, randomization and environmental control;
• reporting percent Gage R&R without stating whether the denominator is tolerance or process variation;
• using MSA evidence after the measurement method, range or appraiser instruction changed.

Scenario Map

Scenario	Exercises	Primary check	Engineering decision
Repeatability and reproducibility	1, 2, 3, 4	equipment variation, appraiser variation and total Gage R&R	Decide whether the measurement system is stable enough for use.
Capability against tolerance	5, 6, 7, 8	percent tolerance, percent study variation, ndc and part variation	Decide whether measurement noise hides real process differences.
Bias, stability and linearity	9, 10, 11, 12, 13	reference bias, drift, slope and attribute agreement	Decide whether the method is accurate across time and range.
Release decisions	14, 15, 16, 17, 18	fixture variation, range method, observed capability and MSA gate	Accept, restrict or redesign the measurement system.

Validation Package Checklist

• study purpose, measurement function and acceptance criterion are defined;
• part selection represents the process or tolerance region being judged;
• operators, trials, randomization and fixture resets are documented;
• repeatability, reproducibility, bias, linearity, stability and resolution are separated where relevant;
• percent tolerance, percent study variation and ndc use the correct denominator;
• restrictions on range, appraiser, fixture or decision type are stated;
• corrective action and impact review are recorded when the MSA gate fails.

Exercise 1: Equipment Repeatability

One operator measures the same part five times. The sample standard deviation is:

Estimate equipment variation using 6s_e.

Solution

Engineering Comment

Equipment variation includes instrument noise, fixturing, contact force and short-term method variation.

Plausibility Check

Six standard deviations should be several times the observed one-sigma scatter.

Exercise 2: Operator Reproducibility from Mean Difference

Two operators measure the same part group. Their average results differ by:

Estimate appraiser variation as this difference.

Solution

Engineering Comment

Operator differences often reveal unclear instructions, fixture technique, reading interpretation or training gaps.

Plausibility Check

If the mean difference is the only operator-effect estimate, the appraiser variation should match it.

Exercise 3: Total Gage R&R

Using:

estimate total Gage R&R by RSS.

Solution

Engineering Comment

When equipment variation is much larger than appraiser variation, improving training alone will not solve the measurement problem.

Plausibility Check

The total is slightly larger than EV, which is expected when EV dominates.

Exercise 4: Repeatability Share

Using the previous result, find the share of Gage R&R variance caused by equipment variation.

Solution

So:

Engineering Comment

The measurement system improvement should focus on equipment, fixture, contact condition or resolution before operator effects.

Plausibility Check

Since EV is more than twice AV, it should contribute most of the variance.

Exercise 5: Percent Tolerance

A feature tolerance is \pm 0.25\ \text{mm}, so total tolerance width is 0.50\ \text{mm}. If GRR=0.078\ \text{mm}, find percent tolerance.

Solution

Engineering Comment

Lower percent tolerance means less of the allowed product variation is consumed by measurement variation.

Plausibility Check

0.078 is about one sixth of 0.50, so about 16\% is reasonable.

Exercise 6: Percent Study Variation

Total observed study variation is:

with:

Find percent study variation.

Solution

Engineering Comment

Percent study variation depends on the parts selected for the study. If the part sample is too narrow, the measurement system can look worse than it is for the process.

Plausibility Check

The Gage R&R is about one quarter of total variation, so 26\% is plausible.

Exercise 7: Number of Distinct Categories

Part-to-part variation is:

and:

Estimate:

Solution

Rounded down:

Engineering Comment

Low ndc means the measurement system cannot reliably separate enough part categories for process control.

Plausibility Check

Part variation is a little over three times GRR, so ndc near four is credible.

Exercise 8: Part-to-Part Dominance

Total variation is TV=0.300\ \text{mm} and GRR=0.078\ \text{mm}. Estimate part variation:

Solution

Engineering Comment

If part variation dominates, the measurement system can still show process differences, but percent tolerance and decision risk must also be acceptable.

Plausibility Check

Because GRR is much smaller than total variation, part variation should be close to total variation.

Exercise 9: Bias Against a Reference

A reference part is 25.000\ \text{mm}. The average measured value is 25.018\ \text{mm}. Find bias.

Solution

Engineering Comment

Bias can create false decisions even when repeatability is good. It should be corrected or included in the decision rule.

Plausibility Check

The measured average is higher than the reference, so bias is positive.

Exercise 10: Bias Relative to Tolerance

Using the bias 0.018\ \text{mm} and total tolerance width 0.50\ \text{mm}, find bias as percent tolerance.

Solution

Engineering Comment

Small bias may be acceptable for screening but not necessarily for final acceptance near a limit.

Plausibility Check

0.018 is small relative to 0.50, so a few percent is expected.

Exercise 11: Stability Shift

A check standard average was 10.002\ \text{mm} last month and 10.026\ \text{mm} this month. Estimate shift.

Solution

Engineering Comment

Stability shift can indicate drift, fixture wear, environmental change or changed operator method.

Plausibility Check

The second average is higher, so the shift is positive.

Exercise 12: Linearity Slope

Bias is 0.005\ \text{mm} at 10\ \text{mm} and 0.025\ \text{mm} at 50\ \text{mm}. Estimate bias slope per millimeter.

Solution

So bias increases by:

Engineering Comment

Linearity problems mean a single bias correction cannot serve the whole range.

Plausibility Check

Bias changes by 0.020\ \text{mm} across 40\ \text{mm}, giving a small positive slope.

Exercise 13: Attribute Agreement Rate

Two inspectors classify 80 parts. They agree on 72 parts. Find agreement rate.

Solution

Engineering Comment

Agreement rate is not enough if the disagreements are concentrated near critical limits or if both inspectors agree on the wrong decision.

Plausibility Check

72 out of 80 is nine tenths.

Exercise 14: Fixture Contribution

A measurement system has total repeatability standard deviation:

Instrument-only standard deviation is:

Estimate fixture contribution by RSS difference.

Solution

Engineering Comment

Fixture contribution can be as large as the instrument contribution, especially with flexible parts, contact probes or inconsistent clamping.

Plausibility Check

The fixture term is positive and similar in size to the instrument term, which fits the total.

Exercise 15: Range Method from Average Range

For repeated readings, average range is:

For two trials, use d_2=1.128 to estimate standard deviation.

Solution

Engineering Comment

Range methods are convenient but depend on subgroup size and assumptions about stable variation.

Plausibility Check

The estimated standard deviation is slightly below the average range, as expected for d_2>1.

Exercise 16: Measurement Variation and Process Capability

Observed process standard deviation is 0.060\ \text{mm}. Measurement standard deviation is 0.020\ \text{mm}. Estimate corrected process standard deviation:

Solution

Engineering Comment

Observed process variation includes measurement variation. Capability studies should not ignore measurement contribution when it is material.

Plausibility Check

Because measurement variation is smaller than observed variation, the corrected value is only slightly lower.

Exercise 17: False Rejection from Measurement Scatter

A part is exactly at target, and the acceptance half-width is 0.10\ \text{mm}. Measurement standard deviation is 0.050\ \text{mm}. How many standard deviations is the limit from target?

Solution

Engineering Comment

If the measurement system alone can move results two standard deviations to the limit, false rejects are plausible.

Plausibility Check

The half-width is twice the measurement standard deviation, so z=2.

Exercise 18: MSA Release Gate

A measurement system has:

Check	Result	Gate
Percent tolerance	16\%	\le 10\% preferred, \le 30\% conditional
ndc	4	\ge 5
Bias	3.6\% tolerance	\le 5\%
Stability shift	0.024\ \text{mm}	\le 0.010\ \text{mm}
Attribute agreement	90\%	\ge 95\%

Decide whether it can be released for final acceptance.

Solution

Percent tolerance is conditional, not strong. ndc fails:

Bias passes, but stability fails:

Attribute agreement fails:

The system should not be released for final acceptance.

Engineering Comment

The measurement system may be usable for rough screening after controls are added, but final acceptance needs better stability, clearer classification and more part categories.

Plausibility Check

Several gates fail and the strongest pass is only bias, so rejection is consistent.