Jig Design #05: Repeatability and Acceptance — How Do You Prove a Jig Is Really Stable?

"It sits nice and tight" or "we ran it a few times without an error" is not enough to accept a jig.

A fixture can repeat well and still sit consistently off the nominal position. It can also be correct when one engineer operates it, yet scatter badly once several operators are involved. Worse, the measuring system used to judge the jig may itself produce more variation than the jig does.

To prove a jig is stable you need a protocol that clearly separates:

Trueness against a reference.
Repeatability when the part is removed and reloaded.
Reproducibility across people, shifts or conditions.
Workpiece variation.
Measurement system variation.
The pass/fail decision rule.

This article shows how to build that protocol for FAT, SAT and periodic requalification.

1. Do not call every kind of error "accuracy"

ISO 5725-1 distinguishes two groups of concepts:

Trueness: how close the mean value is to the reference value.
Precision: how close repeated results are to each other.

In fixture practice:

Trueness: is the average machined hole centre actually at the nominal position?
Repeatability: with the same part, the same person and the same conditions, does unloading and reloading return to nearly the same position?
Reproducibility: when the person, shift, machine or conditions change, are the results still equivalent?

A jig with high repeatability but a fixed offset can still be corrected. A jig whose mean is correct but whose scatter is large cannot be rescued by adjusting a single offset.

2. Define the acceptance question before choosing the measurement

"Measure the repeatability of the jig" is far too vague. Write it as questions:

Which datum does the jig locate the workpiece against?
Which output characteristic is the CTQ?
How much scatter is allowed?
Are we accepting the jig alone, the jig mounted on the machine, or the whole process?
Will the result release production, or only compare design options?
Who decides when a result sits close to the limit?

If the jig is used for machining, the CTQ may be the feature position after machining. If the jig is used for inspection, the CTQ is the measurement result and the uncertainty of the jig–sensor–software system.

Do not choose a measurement just because the equipment is available. Choose a measurand that reflects the function of the jig directly.

3. Freeze the system boundary

Before testing, draw the boundary:

The workpiece.
The jig.
Locators, supports and clamps.
The machine or baseplate.
The measuring equipment.
The method of establishing the datum.
The operator.
The software and the data processing.
Temperature, cleanliness and stabilization time.

If clamping force, the probe or the alignment algorithm changes between runs, the data no longer represents the same system.

Record the revision of the jig, the locators, the measuring program and the CTQ drawing. The protocol must allow the same conditions to be reproduced later.

4. Prove the measuring system before using it to judge the jig

NIST emphasizes measurement process characterization in order to understand and quantify the sources of error in a measurement result. If the measuring system is not stable enough, you cannot conclude that the variation comes from the jig.

Check at least:

The equipment is within a suitable calibration status.
The resolution is fine enough to see the variation being assessed.
The measurement setup does not add large errors of its own.
Datum alignment is clearly defined.
The probe can reach the feature repeatably.
Temperature and stabilization time are appropriate.
The people measuring follow the same work instruction.

A gauge R&R or measurement system study can separate the repeatability of the equipment and method from the reproducibility between operators. Do not apply one fixed percentage threshold mechanically to every CTQ; acceptance has to match the risk and the intended use.

5. Separate the three layers of variation

An observed result is usually a combination of:

The workpiece: tolerance, flatness, burrs, material and temperature.
The jig and process: locator clearance, clamping force, deflection, wear, chips and the machine.
The measurement: gauge, alignment, operator, environment and calculation.

If you only measure many different workpieces, you cannot tell whether the variation comes from the parts or the jig. If you only measure one workpiece without unloading it, you are mainly testing the measuring system.

Design separate experiments:

Repeated measurement in the same state, part not removed: estimates short-term measurement variation.
Unload and reload the same part several times: adds locating and clamping variation.
Several workpieces: adds part-to-part variation.
Several operators or shifts: adds reproducibility variation.

6. Choose a representative sample of workpieces

The "nicest" part does not prove the jig runs well with production material.

The sample should cover:

Dimensional extremes that affect the locators.
Realistic flatness or bow.
Locating hole positions.
Material, coating or heat-treat condition where relevant.
Different cavities, moulds or suppliers.
Models and revisions within the jig scope.

Do not deliberately select only out-of-spec parts to evaluate the jig, unless you are running a challenge test. The main goal is to represent the distribution of valid input.

Record the ID of each part so that part-to-part variation can be distinguished from run order.

7. Design the repeatability trial

One repeat cycle has to include the real operations:

Release the clamps.
Remove the part from every locator.
Clean according to the procedure.
Reload the part.
Confirm seating.
Clamp in the correct sequence and conditions.
Measure the CTQ.

Only opening and closing the clamps without taking the part off the locators evaluates clamping force more than the ability to relocate.

Randomize the part order so that drift over time is not mistaken for part-to-part difference. If the operator sees the result beforehand and can "tweak by hand" until it passes, blind the result during data collection where appropriate.

8. Test reproducibility across people and conditions

A good jig reduces dependence on individual skill.

Test with:

Several representative operators.
The first and the last shift.
After warm-up and from cold.
After cleaning and after enough cycles to reach a representative level of contamination.
Manual and robot loading, if both are in scope.
Supply pressure across the allowed operating range.

Do not change many factors at once without recording it. The experiment has to allow a difference to be traced back to its source.

If one operator consistently produces different results, review:

The loading sequence.
Where the part is gripped.
Force or speed of handling.
How seating is confirmed.
The work instruction.
The guiding geometry of the jig.

The goal is not to "fix the person", but to make the jig less sensitive to reasonable variation in handling.

9. Watch for drift over time

Repeatability over ten minutes does not prove stability over a whole shift.

Sources of drift:

Temperature of the machine, jig and workpiece.
Locator wear.
Chip build-up.
Pressure changes.
Bolts working loose.
Sensors or brackets moving.
The probe warming up.

Design a trial that extends across the important states, or measure a check standard periodically. Plot the results over time to reveal trends, step changes and cycles, instead of reporting only one overall standard deviation.

If the data drifts, the assumption of "one stable distribution" no longer holds.

10. Measure the real CTQ, but add diagnostic points

The CTQ decides pass or fail. Adding a few diagnostic measurands, however, helps find the cause:

The position of each locator.
Clamp stroke.
Pressure at the mechanism.
Lift at the supports.
Temperature.
Clamping force or force signature.

Do not turn the protocol into hundreds of purposeless measurements. Every diagnostic channel should be tied to a specific failure mode.

For example: the CTQ is off but all three seating points are stable, while clamp-stroke data changes from part to part. That points the investigation in a different direction than an air check reporting that the part is not seated.

11. Data analysis: look at the plots before the summary number

The basic charts:

Results in time order.
Distribution by part.
Distribution by operator.
Results by unload/reload cycle.
Range within each repeat group.
Operator × part interaction.

Ask:

Is one part always different?
Does variation grow with the size of the input feature?
Does one operator create an offset, or extra dispersion?
Do the results drift over time?
Are there two clusters corresponding to two seating states?

A single overall R&R figure can hide all of this structure. NIST likewise recommends examining repeatability by check standard, operator or gauge when evaluating the data.

12. Trueness needs an independent reference

To evaluate trueness or bias, you need a suitable reference value:

A master part measured by a method of higher capability.
A calibrated artefact.
A CMM or an independent reference system.
An alternative measurement method with appropriate traceability.

Never use the jig being accepted to assign reference values to the parts and then use those values to prove the jig is correct.

If the jig has a correctable offset:

Confirm that repeatability is good enough.
Estimate the bias against the reference.
Correct it according to a procedure.
Rerun an independent validation.

Do not adjust between measurements inside the same study without recording it; that destroys the information about drift and bias.

13. Fix the criteria before looking at the results

Acceptance criteria must be defined in advance:

Which CTQ is evaluated.
The limit on mean position or bias.
The limit on dispersion or repeatability.
Requirements per operator.
Test conditions.
The rule for handling outliers.
The number of repeats and how the statistics are calculated.
The decision rule near the limit.

If the metric is chosen — or an "ugly" sample dropped — only after seeing the data, the conclusion is easily biased.

Do not assume a ratio such as "the jig must consume 10% of the tolerance" for every application. The allocation depends on the total tolerance budget, the measurement uncertainty, the severity of the failure and the process capability.

14. Decision rules and the zone near the limit

ISO 14253-1 provides rules for proving conformity or non-conformity while taking measurement uncertainty into account for GPS characteristics.

In a contract or an acceptance test, both parties need to agree:

Which results count as a pass.
How measurement uncertainty is taken into account.
Whether a guard band applies.
Who carries the risk of decisions near the limit.
When a re-measurement with a reference method is required.

Do not simply compare the displayed value with the limit and ignore uncertainty. But equally, do not introduce your own guard band after measuring if the requirement was never agreed.

15. FAT and SAT are not the same test

FAT — Factory Acceptance Test

Usually performed where the jig is built:

Demonstrates the design and its functions.
Checks geometry, repeatability, interlocks and challenge tests.
Uses the parts and environment available, but the differences from production must be recorded.

SAT — Site Acceptance Test

Performed where the jig is used:

Mounted on the real machine, baseplate and utilities.
With real parts, operators and conditions.
Confirms MES, PLC and robot interfaces where applicable.
Checks drift, cycle time and recovery.

Passing FAT does not guarantee passing SAT if the machine, foundation, temperature, air supply or loading method differ. The protocol should state which items are repeated at SAT.

16. An acceptance protocol template

A. Configuration information

Jig number and revision.
CTQ drawing and revision.
Locators, clamps and change parts fitted.
Machine, program and software.
Measuring equipment and calibration status.
Environmental conditions.

B. Sample scope

Part IDs.
Model, cavity, supplier.
Relevant input characteristics.
The reason for the sample selection.

C. Test matrix

Operators.
Randomized order.
Number of unload/reload cycles.
Clean/contaminated and cold/warm conditions.
Pressure or operating mode.

D. Data

CTQ.
Diagnostic points.
Timestamps.
Alarm states.
Notes on anomalies, without editing the raw data.

E. Analysis

Repeatability.
Reproducibility.
Bias and trueness.
Drift.
Significant interactions.
The measurement uncertainty and decision rule applied.

F. Conclusion

Pass.
Conditional pass with actions.
Fail requiring a design change.
Items that need retesting.

17. Requalification after entering production

Create the baseline when the new jig is accepted:

Initial CTQ results.
Locator positions.
Clamp stroke.
Check standard.
Photographs of the condition and key torque or setting values.

Re-evaluate when:

A locator or support is replaced.
The jig body is repaired.
A collision occurs.
The part model or revision changes.
The measuring program changes.
The jig moves to another machine or line.
SPC or drift results exceed a trigger.
The risk-based periodic interval is reached.

Not every event requires rerunning the whole protocol. Define a test subset according to the scope of the change, but with a documented reason and traceability.

18. Common acceptance mistakes

Measuring only one part

Does not cover the variation of the input.

Not removing the part between runs

Evaluates the gauge more than the jig.

Using an unverified measuring system

You cannot tell whether the variation belongs to the jig or to the measurement.

Letting the operator adjust freely

Turns the jig into a skill-dependent process.

Reporting only the mean

Hides dispersion, drift and interactions.

Choosing the criteria after seeing the data

Increases the risk of a biased conclusion.

Running FAT under ideal conditions

Not representative of production, and creates surprises at SAT.

19. Checklist before signing acceptance

Scope

CTQ, datums and system boundary are defined.
Jig, part, machine, program and gauge revisions are clear.
FAT and SAT objectives are distinguished.

Measuring system

The measurement system is capable enough.
Alignment and work instruction are consistent.
An independent reference exists for evaluating bias.

Study design

Parts represent the full range of valid input.
Full unload and reload operations are included.
Several operators and conditions are covered when in scope.
Test order is randomized where needed.
Warm-up, temperature, chips and pressure are considered.

Analysis and decision

Plots by time, part and operator are reviewed.
Repeatability, reproducibility and bias are separated.
Criteria and outlier rules were fixed in advance.
Measurement uncertainty and the decision rule are agreed.
Actions and retest scope are clear.

Life cycle

The baseline is stored.
Requalification triggers are defined.
Data, reports and approvals are traceable.

Conclusion

Jig acceptance is not a demonstration. It is a controlled experiment designed to separate the variation of the workpiece, the jig, the operator and the measuring system.

Start from the CTQ and the question you need to answer. Prove the measuring system is good enough, unload and reload for real, use representative parts, look at the data over time, and agree the decision rule before you measure.

When the protocol is designed properly, the result does not just say "the jig passed". It tells you how stable the jig is, under which conditions, and when it has to be re-evaluated.

View all MINATA technical articles