A measurement system analysis (MSA) is just as mandatory for optical measurement systems as for tactile gauges. But the sources of variation it has to uncover are different. With hand gauges, the operator dominates reproducibility through gauge force, reading, and handling. In an automated optical system, that influence shrinks. Fixturing and part seating, lighting and ambient light, part surface condition, and the programming of the evaluation move to the front.
That is why applying a classic GR&R study unchanged to a vision-based system often misses the problem. Such a study can certify excellent repeatability and overlook the sources of variation that actually hit in series production. This article shows what changes for MSA and GR&R with optical systems, within the framework of AIAG MSA and IATF 16949.
Key Takeaways
- An MSA is just as mandatory for optical systems as for tactile gauges (IATF 16949, among others).
- Operator influence drops. Fixturing, lighting, surface, and programming become the main sources of variation.
- In a GR&R study, you replace "operator" with the real changing conditions, and each repeat has to include the full measurement cycle.
- For pass/fail systems, an attribute MSA takes the place of the classic GR&R.
- AIAG guideline: %GRR under 10% is acceptable, and 10% to 30% is conditionally acceptable.
Why Do Automated Optical Systems Need an MSA Too?
Proof of gauge capability is required regardless of the measurement technology. IATF 16949 calls for measurement system analysis on the gauges used in production control, and automotive customers expect proof under AIAG MSA whether the measurement is tactile, optical, or laser-based. The reason is basic. Every observed process variation includes measurement variation, and every pass/fail decision at a tolerance limit is only as reliable as the measurement system behind it.
An automated system also creates a particular risk. It always delivers a value, with the same confident look, whether the measurement was stable or not. Without an MSA, there is no proof that the values describe the parts and not the machine's mood that day.
For the basic terms, see KEYENCE's page on accuracy and repeatability in measurement.
Which Sources of Variation Replace Operator Influence?
In an optical system, variation doesn't go away. It moves from the person to the setup. This comparison shows where an MSA for optical systems has to look.
Operator (reproducibility)
Tactile handheld gauge
Dominant: gauge force, reading, measuring location
Automated optical system
Small, but not zero: part loading, cleaning, program selection
Fixturing and part seating
Tactile handheld gauge
Part is positioned by hand or in a fixture
Automated optical system
Central: seating points, dirty fixtures, residual play
Lighting and ambient light
Tactile handheld gauge
Almost irrelevant
Automated optical system
Central: ambient light, lighting aging, reflections
Part surface condition
Tactile handheld gauge
Low (mechanical contact)
Automated optical system
Central: gloss level, oil film, temper colors, lot differences
Temperature
Tactile handheld gauge
Part and gauge
Automated optical system
Part, optics, and mechanics (warm-up behavior)
Programming and evaluation
Tactile handheld gauge
Not present
Automated optical system
Central: edge finding, filter parameters, tolerance of the evaluation logic
Stability over time
Tactile handheld gauge
Wear of measuring faces
Automated optical system
Drift of optics and lighting, contamination
The consequence for study planning is this. A GR&R that only covers "three operators measure ten parts three times each" mainly tests the repeatability of the camera on an optical system. That is the smallest source of variation anyway. The study becomes meaningful only when the real influencing factors appear as factors in the study design. That means re-presenting the part instead of measuring it in an unchanged position, using different shifts and times of day instead of one lab session, and using parts with production surfaces instead of cleaned sample sets.
How Do You Structure a GR&R Study for an Automated Optical System?
The core idea is that "operator" is replaced by the real changing conditions, and each repeat has to cover the complete measurement cycle, not just the image capture. A proven setup has five steps.
- 1. Fix the feature and tolerance. The study is evaluated per feature, and the reference is the tolerance (and, for process evaluation, the process variation as well).
Fix the feature and tolerance. The study is evaluated per feature, and the reference is the tolerance (and, for process evaluation, the process variation as well).
- 2. Select parts across the real range of variation. Use ten parts from series production that cover the actual process variation and surface variance, including parts near the limits, not just "nice" samples.
Select parts across the real range of variation. Use ten parts from series production that cover the actual process variation and surface variance, including parts near the limits, not just "nice" samples.
- 3. Repeat as a complete cycle. Remove each part and load or feed it again for every repeat, so that fixturing and part seating enter the repeatability. Measurements in an unchanged position systematically underestimate the real variation.
Repeat as a complete cycle. Remove each part and load or feed it again for every repeat, so that fixturing and part seating enter the repeatability. Measurements in an unchanged position systematically underestimate the real variation.
- 4. Replace "operator" with real condition factors. Depending on the system, use different loaders or shifts and different times of day (ambient light, temperature), and where possible a warm and a cold start. With fully automatic feeding, time stability takes the place of reproducibility, and you capture it with measurements spread over time.
Replace "operator" with real condition factors. Depending on the system, use different loaders or shifts and different times of day (ambient light, temperature), and where possible a warm and a cold start. With fully automatic feeding, time stability takes the place of reproducibility, and you capture it with measurements spread over time.
- 5. Evaluate and document. Report %GRR relative to the tolerance, add a stability chart for the reference part, and file the result in the gauge record as proof of capability for audits.
Evaluate and document. Report %GRR relative to the tolerance, add a stability chart for the reference part, and file the result in the gauge record as proof of capability for audits.
The established AIAG guidelines apply to the evaluation. A %GRR under 10% is considered acceptable. Between 10% and 30%, it is conditionally acceptable, depending on the importance of the feature, the cost, and customer approval. Above 30%, it is not acceptable.
Some customers ask for an uncertainty-based approach instead or in addition. There, you work out an expanded measurement uncertainty and compare it to the tolerance. Which proof is required is set by your customer or your quality management system. Where an uncertainty budget is requested, it usually relies on references traceable to NIST and, for outside calibration, on ISO/IEC 17025 accredited labs.
What Should You Do for Vision Systems With a Pass/Fail Result?
If the system doesn't deliver measured values but attribute decisions, an attribute MSA takes the place of the GR&R, such as an attribute agreement study under AIAG. You assemble a sample set of clearly good, clearly bad, and deliberately borderline parts, and inspect it several times in random order. The study rates how well the system agrees with the reference decision and with itself.
Two practical points matter. First, the study stands or falls with the limit samples. A sample set with no borderline parts certifies perfect results for any system. Second, check whether the inspection task can be solved by measuring after all. A continuous measured value with a limit is methodologically better than an attribute decision, because it allows a regular GR&R, trend monitoring, and SPC.
Where Standard MSA Recipes Reach Their Limits on Optical Systems
The established MSA methods remain valid, but you should know three limits before you over-interpret results.
First, transferability is limited. A %GRR applies to the studied combination of feature, part range, and setup. If the surface changes (a new supplier, a different mold release in injection molding) or the inspection program changes, repeat the proof. With optical systems, that happens more often than with tactile ones, because they are more sensitive to surface.
Second, a GR&R that passes formally can hide a systematic error. The study evaluates variation. Linearity and bias across the measuring range need their own studies with traceable references.
Third, the classic figures come from the world of sample measurement. With 100% inline systems, you also face questions of long-term stability, recalibration strategy, and how to handle drift alarms, and a one-time study doesn't answer them. A recurring short check, using a stability part with a control chart, has become a common addition.
Frequently Asked Questions
Under common AIAG practice, a %GRR under 10% is acceptable, 10% to 30% is conditionally acceptable (depending on feature importance, cost, and customer approval), and above 30% is not acceptable. The requirement agreed with your customer always governs. Some customers add or substitute an uncertainty-based proof.
Classic operator reproducibility largely goes away, but it is replaced, not dropped. Changing conditions such as part feeding, shifts, ambient light, temperature, and machine warm-up take its place. A meaningful study therefore spreads the measurements over different times and conditions instead of capturing all repeats in one session.
AIAG MSA rates measurement systems with figures such as %GRR from standardized studies, and it is the reference in the North American automotive supply chain. An uncertainty-based approach identifies the individual uncertainty components, combines them into an expanded measurement uncertainty, and compares it to the tolerance. In international supply chains, both kinds of proof are often relevant, and your customer or quality system defines which one you need.
Yes, when conditions that affect capability change. Examples are a new inspection program or changed evaluation parameters, a changed fixture, a clearly different part surface (supplier, material, coating), or a move of the machine to another site. With optical systems, surface dependence is the most common trigger. A shortened repeat proof with a stability part and a sample GR&R is common practice.
Next Steps
Whether an optical measurement system reaches the required gauge capability on your parts is decided on the real part with its production surface, not on the data sheet. To discuss it, ask an expert, or request a demo or test on your original parts. KEYENCE offers laser profilers, optical measurement systems, and vision systems to evaluate.
- Image Dimension Measurement System IM-X1000 series
Image Dimension Measurement System
IM-X1000 series
- 2D/3D Laser Profiler LJ-X8000 series
2D/3D Laser Profiler
LJ-X8000 series
