You change the process, the Cpk does not move, and everybody assumes the process is stable. Nobody checked whether the gage measuring that part can be trusted with itself.
Measurement System Analysis, or MSA, checks whether your gage, your method and the person measuring produce a number you can trust. Before you believe a Cpk or a control chart point, MSA asks whether that data reflects the part or reflects the measurement system.
Two questions MSA keeps separate
A Gage Repeatability and Reproducibility study, known as Gage R&R, splits measurement system variation into two pieces. Repeatability is the variation you get when the same person measures the same part multiple times with the same gage. Reproducibility is the variation that shows up when the person measuring changes. If neither one is controlled, you do not know whether the variation you see in your process belongs to the process or to the gage.
The number that decides whether you trust the gage
The result is expressed as %GRR, the percentage of total variation that belongs to the measurement system rather than to the parts themselves. Per AIAG guidance, a %GRR under 10% is considered acceptable. Between 10% and 30% may be acceptable depending on the application, the cost of the gage and how critical the characteristic is. Above 30% the measurement system is not acceptable, and no data coming out of it should be used to decide anything.
The example I run in the workshop
Picture a study with three operators, ten parts and two repeats each. If the part to part variation is wide and the variation from repeating the same part is small, %GRR comes out low and the gage earns the benefit of the doubt. If instead the variation between repeats from the same operator is nearly as large as the variation between different parts, the gage is reacting to its own noise more than to the part.
Why this comes before Cpk
There is no point calculating process capability on a gage that did not pass MSA. A Cpk calculated on data from an unstable measurement system describes the gage that produced it, and any decision made off that number inherits the same error. That is why MSA runs first in the sequence, not after it and not as an audit formality.
What usually fails, and it is not the gage
When a Gage R&R comes back bad, suspicion usually lands on the instrument first. In most studies I have run, the problem sits in the method: the operator holds the part differently each time, the criteria for where to touch the surface was never written down, or the gage was never calibrated before the study. Swapping gages without fixing the method buys a new problem with the same name.
Three mistakes I keep seeing in an MSA
- Running the study with only the most experienced operator, because "they measure best." The study needs the shift’s real variation, not one operator’s best version.
- Using parts that do not cover the real production range. If every part in the study is nearly identical, part to part variation reads artificially low and %GRR looks worse than it is.
- Accepting a 25% GRR without asking what the data will be used for. What is acceptable on a reference measurement is not always acceptable on a safety characteristic.
Where this fits
MSA is the first filter before any process capability or statistical control exercise, and it is part of what I check in any problem solving workshop where the data itself is in question, not yet the cause.
Start here
Before you trust the next Cpk someone hands you, ask when that gage last ran a Gage R&R and what %GRR it produced. If nobody knows the answer, that is the first problem to solve, before arguing about any other cause.
What does %GRR mean?
It is the percentage of total variation that belongs to the measurement system (gage, method and operator) rather than to the parts themselves. The lower it is, the more you can trust that the number you measure describes the part and not the gage.
How good does %GRR need to be?
Per AIAG guidance, under 10% is considered acceptable. Between 10% and 30% may be acceptable depending on the application and how critical the characteristic is. Above 30% the measurement system is not acceptable.
How often should MSA be repeated?
There is no universal rule. It depends on the plant’s control plan and the customer’s requirements. What stays consistent is that it gets repeated after any gage change, any out of range calibration or any quality problem that points back to the measurement system.