A quality manager at an automotive supplier recently called me after a customer rejected 2,400 sealing rings. The customer claimed the dimensions were out of tolerance. The supplier's final inspection data showed every part was well within specification. The supplier shipped conforming product according to their instruments, but the customer received nonconforming material.
The contradiction points to a failure in Measurement System Analysis (MSA). When a gauge repeatedly confirms a dimension is correct, but the customer's incoming inspection disagrees, the gauge is not telling the truth. I asked the manager when they last verified that their measurement system actually measured reality, aside from standard daily calibration. The silence on the line was the answer.
Most automotive and aerospace plants treat MSA as a synonym for Gage R&R. They verify repeatability and reproducibility to satisfy IATF 16949 or AS9100 requirements, file the report, and assume the gauge is capable. But Gage R&R only measures variation. It does not tell you if the instrument is reading the correct absolute value, if that error shifts across the measurement range, or if the system drifts over six months.
Bias: The Systematic Lie
Bias is the difference between the observed average of measurements and a calibrated reference value. If you measure a 10.000 mm master gauge and your instrument repeatedly reads 10.012 mm, your bias is +0.012 mm. The instrument is not making a random error. It is consistently and predictably lying by the exact same amount.
This systematic error is dangerous because it mimics precision. If an operator measures twenty parts and the gauge returns identical readings, the data looks highly capable. But if every reading is shifted by +0.012 mm, you can systematically accept parts sitting just outside the tolerance band. The rejected sealing rings I investigated suffered exactly this failure mode.
To quantify bias, follow the AIAG MSA reference manual. Measure a calibrated reference standard at least ten times using the same operator and setup. Calculate the average of those readings and subtract the reference value. Verify the statistical significance of the result using a t-test to ensure the bias is real and not a sampling artefact.
I have audited plants where bias was not constant, but fluctuated with ambient temperature. In the morning, when the hall was cold, the bias was +0.005 mm. By afternoon, after the machinery had heated the environment, the bias shifted to +0.018 mm. That thirteen-micron shift is the difference between a conforming and nonconforming part when the total tolerance is ±0.025 mm. Implementing a temperature compensation protocol in the measurement procedure eliminated the drift.
Linearity: Error Across the Range
While bias describes systematic error at a single point, linearity asks how that error changes across the gauge's entire operating range. An instrument might exhibit a bias of +0.005 mm at 10 mm, +0.012 mm at 30 mm, and +0.025 mm at 50 mm. The instrument is not just inaccurate; its inaccuracy scales with the dimension being measured.
This pillar frequently catches out coordinate measuring machines (CMMs). A CMM calibrated with a 50 mm master block will show excellent bias at 50 mm. However, if the parts it actually measures in production fall between 10 mm and 20 mm, the CMM might suffer from compounded geometrical errors in that lower range, creating a cumulative error large enough to cause assembly failures.
To verify linearity, select five or more reference standards spanning the entire operating range of the instrument. Measure each standard at least ten times and calculate the bias for each point. Plot the bias values against the reference values and calculate the best-fit linear regression line. The slope of this line tells you exactly how the error magnifies across the range.
The acceptance criterion is strict. If the absolute value of the slope multiplied by the range exceeds 10% of the total tolerance, the linearity is unacceptable. You must recalibrate the instrument using masters that cover the full range, not just the midpoint, or restrict the gauge's qualified operating range to where it remains linear.
AIAG Acceptance Criteria for Location Metrics
Stability: Monitoring Drift Over Time
Bias and linearity are snapshot characteristics. You measure them on a Tuesday and assume they hold true on Friday. Stability tests that assumption. It tracks whether your measurement system remains consistent over weeks and months as components wear, sensors degrade, and environmental conditions shift.
You verify stability using an X̄-R or X̄-S control chart on a single master gauge. An operator measures the reference standard at the start of every shift, taking three to five readings. You plot the averages and ranges over time. If the points remain randomly distributed around the centreline with no special causes, the system is stable.
In one plant, I implemented a daily master gauge check on an optical sensor. For three months, the X̄-R chart showed perfect stability. Then the data began to shift. The control chart triggered a warning before any defective parts were produced. Investigation revealed a new supplier of cleaning chemicals was leaving a microscopic film on the contact surfaces, interfering with the optical reading.
Stability monitoring is the cheapest preventive maintenance tool in quality engineering. Catching sensor drift via a control chart costs nothing but operator discipline. Catching it through a customer 8D report costs significant engineering time, scrap, logistics, and lost trust.
The Compounding Effect of System Failure
When a measurement system fails an MSA audit, it is rarely just one pillar causing the problem. The danger lies in the compounding effect. A gauge might have an acceptable bias of +0.005 mm, but if it also suffers from a +0.008 mm non-linearity at the upper end of the range, and a +0.004 mm drift over three months, the combined error pushes parts outside tolerance.
This is why verifying only one or two characteristics leaves a blind spot. If you check bias but ignore linearity, you do not know if the gauge is accurate across the whole part family. If you check both but ignore stability, you do not know if the data is still valid six months after the initial study.
If you cannot trust your measurement system, you cannot trust any quality decision based on its data.
In the case of the rejected sealing rings, the full MSA revealed the compounding reality. The gauge had a marginal bias, a slight non-linearity at the upper tolerance band, and a slow upward drift that nobody monitored. Individually, each error seemed trivial. Combined, they generated enough false accepts to trigger a customer rejection.
Implementing a Comprehensive MSA Protocol
- 01System InventoryClassify gauges by criticality (A, B, C) based on product risk.
- 02Bias VerificationMeasure a calibrated master 10 times; calculate average offset.
- 03Linearity CheckPlot bias across 5 masters spanning the entire operating range.
- 04Stability MonitoringTrack daily master measurements using an X̄-R control chart.
- 05RequalificationRepeat full MSA annually, after crashes, or following relocation.
Building MSA Into the Quality System
Treating MSA as a one-time paperwork exercise for PPAP submission guarantees future measurement failures. Bias, linearity, and stability must be integrated into the routine quality management system. This means establishing hard boundaries for gauge acceptance before production begins, not after a customer files a complaint.
Categorise every measurement system by its impact on final product quality. Category A gauges directly determine final acceptance on critical safety or dimensional features. Category B gauges monitor in-process dimensions. Category C gauges provide purely indicative data. Apply the full rigour of bias, linearity, and stability testing exclusively to Category A and B systems.
When a gauge fails its linearity or stability check, do not simply recalibrate and move on. Open a nonconformance. Run an 8D investigation to understand why the system failed. Did a probe wear unevenly? Did a fixture loosen? Did a software update alter the algorithm? Treat measurement system failure with the same rigour you apply to a 9001 nonconforming product investigation.
MSA is the foundation of every metric we use in quality engineering. Your OEE calculations, Cpk reports, SPC charts, and PFMEA risk priorities all depend on the integrity of the underlying measurement. If the measurement system systematically lies, every downstream decision built on that data is compromised. Mastering bias, linearity, and stability is the only way to ensure your quality decisions are built on solid ground.