MSA vs. Measurement Capability Analysis - What are the differences?

[Jim Wynne]

The 10:1 rule and GR&R results are not directly related. The former says that the device used should be 10x as sensitive as the measurement required. In other words, if you have need to measure to .001", the device should be capable of .0001" discrimination. The 10% GR&R target takes other factors into consideration, such as operator error and part-to-part variation. Whether or not 10% is appropriate is dependent on the criticality of the measurements and the amount of tolerance available. If the process is already using 95% of the tolerance, 10% might be too much. On the other hand, if the process is using only 10% of the tolerance, more gage error is tolerable.

 

[sushant_kulkarni]

By MSA analysis

We have decided that

Our system is capable of either checking or analysing the particular data.

Through its not giving exat value like cpk or cp but still MSA & Measurement

Capability Analysis is the same thing.

 

[Potter John]

Welcome to the Cove, Errol!

Yes, they are basically the same. The whole exercise of MSA helps you quantify the two types of variation (position variation - bias, linearity and width variation-GRR) for measuring a particular characteristic.

Various indexes can be computed.

A. Based on a Bias/Repeatability Study

(Ref: My old notes based on the booklet "Capability of Measuring Instruments": Robert Bosch)
Take repeated measurements (25-50) on a production master, calculate average and Repeatability StDev.

The two indexes calculated are:
1. Cgm = (0.2 * Tolerance) / (6 * Rpt_Stdev)
2. Cgmk = Least of
( RefValue + (0.1 * Tolerance) - XBar) / (3 * Rpt_StDev)
-OR-
( XBar - (RefValue - (0.1 * Tolerance)) ) / (3 * Rpt_StDev)

Both should be >= 1.33; and bias (XBar-RefValue) not significant.

B. Based on GRR, typically two indexes are recommended:

1. Measurement Capability Index (MCI_I) based on Process Variation is defined as:
MCI_I = 100 (Sigma_GRR/Sigma_process)
Sigma_Process is calculated from ongoing SPC studies: RBar/d2

2. The other index, (MCI_II) is based on tolerance:
MCI_II = 100 (5.15 * Sigma_GRR/ Tolerance )
In place of 5.15 you can also use 6 to cover 99.73% spread.

Acceptance criteria for both these are:
<=20% :Good
>20 - <=30% :Marginal
>30% :Unacceptable

Atul,

I am familiar with performing a graphical MSA and developing control charts for Range and X-bar, as well as a Bias chart. I calculate the difference between the UCL and LCL on the X-bar chart to determine a GR&R as % of tolerance similar to your MCI_II above:

GR&R as a % of Tolerance = 100X (UCL - LCL)/Tolerance

In your opinion, is this method similar to your MCI_II method in terms of results?

I use the following Acceptance Criteria:
<=10% :Good
>10 - <=30% :Marginal
>30% :Unacceptable

Thank you very much for your input.

 

[Atul Khandekar]

.... I calculate the difference between the UCL and LCL on the X-bar chart to determine a GR&R as % of tolerance similar to your MCI_II above:

GR&R as a % of Tolerance = 100X (UCL - LCL)/Tolerance

In your opinion, is this method similar to your MCI_II method in terms of results?

I use the following Acceptance Criteria:
<=10% :Good
>10 - <=30% :Marginal
>30% :Unacceptable

Not quite, IMO. The XBar chart limits are based on repeatability. Also the (UCL-LCL) will give you 6*Sigma-XBar which will be scaled down from Repeatability stDev by a factor of sqrt, where n=subgroup size.

You might want to try out with a few data sets to see what kind of results this equation yields- esp if there is a high reproducibility error.

 

[potdar]

The 10:1 rule and GR&R results are not directly related. The former says that the device used should be 10x as sensitive as the measurement required. In other words, if you have need to measure to .001", the device should be capable of .0001" discrimination. The 10% GR&R target takes other factors into consideration, such as operator error and part-to-part variation. Whether or not 10% is appropriate is dependent on the criticality of the measurements and the amount of tolerance available. If the process is already using 95% of the tolerance, 10% might be too much. On the other hand, if the process is using only 10% of the tolerance, more gage error is tolerable.

Jim,

When calculating the process variation using a given set of measuring devices, the figure arived at already includes both process variation as well as measurement variation. The acceptability or otherwise of the process is defined using this combined figure.

MSA is a separate analysis where process variation is excluded from the calculation and only measuring system variation is considered. The acceptability or otherwise of this result is an absolute decision. Not related to the process capability.

The example given by you is extreme. A 95% process variation by itself is not acceptable.

Going to other extreme, if the process variation is only 10%, that cannot make a GRR of 50% acceptable.

 

[Jim Wynne]

When calculating the process variation using a given set of measuring devices, the figure arived at already includes both process variation as well as measurement variation.

I think you might need to back up a little and reconsider what you wrote. Did you really mean to say that calculation of process variation assumes knowledge ("...already includes...") of process variation? How can you know something before you know it?

MSA is a separate analysis where process variation is excluded from the calculation and only measuring system variation is considered.

Sorry, potdar, but process variation may or may not be included in calculations, depending on what one hopes to accomplish by the analysis.

The example given by you is extreme. A 95% process variation by itself is not acceptable.

I referred to a process that is using 95% of the tolerance, which is a common occurrence, not "95% process variation."

Going to other extreme, if the process variation is only 10%, that cannot make a GRR of 50% acceptable.

My point was that measurement system evaluation should take into account the amount of the tolerance that's being used by the process. Of course, the AIAG methods leave us in a sort of Catch-22 in this regard, as we can't be sure of the process variation if we measure using systems that have not been qualified.

 

[Statistical Steven]

In most industries (except for AIAG documents) the goal of a measurement system is to be adequate for its intended use. Therefore, the tolerance determines the capability of a device. A ruler is sufficient to measure the size of a nail, but insufficient to measure the distance from Maine to California. Additionally, I concentrate with my clients on sources of variation and not total variation. Knowing the GRR% or other "total system" error does not give any information if the measurement system can be improved to be used in a specific application.

 

[Jim Wynne]

In most industries (except for AIAG documents) the goal of a measurement system is to be adequate for its intended use. Therefore, the tolerance determines the capability of a device.

I think the AIAG methods are aimed at measurement efficacy, but the AIAG processes as a whole (APQP, SPC, MSA) were not well thought out, and sometimes it's obvious that the people who wrote one manual weren't talking to the ones that wrote another.

A ruler is sufficient to measure the size of a nail, but insufficient to measure the distance from Maine to California.

It works OK for me; Los Angeles is 9" away from Bangor on my map.

 

[Statistical Steven]

Mr

I think the AIAG methods are aimed at measurement efficacy, but the AIAG processes as a whole (APQP, SPC, MSA) were not well thought out, and sometimes it's obvious that the people who wrote one manual weren't talking to the ones that wrote another.



It works OK for me; Los Angeles is 9" away from Bangor on my map.

Jim, you and I think alike....

I was hoping someone would have said you can use a ruler and a map to determine the distance. Of course the discrimination of that measurement system is questionable, but nonetheless you can use it.

 

[potdar]

I think you might need to back up a little and reconsider what you wrote. Did you really mean to say that calculation of process variation assumes knowledge ("...already includes...") of process variation? How can you know something before you know it?

Well Jim, when I start I only know that the process varies and my measuring system varies. I dont have the faintest idea about "how much".

There also is some confusion regarding the definition of "process variation". Is it "variation due to the process", or "variation in process parameter measurements obtained during analysis"?

The second case is commonly understood as "process variation" and includes variation due to all aspects involved in the process and inspection.

Sorry, potdar, but process variation may or may not be included in calculations, depending on what one hopes to accomplish by the analysis.

The measurement data includes both "variation due to process" and "variation due to MS".

So far as I understand, MSA is necessarily done by eliminating effects of process variation. The calculation methods enable that. There is no relation to "what one hopes to accomplish".

If the data is analysed without any filtration technique, the output is known as the "process variation" (typically variance, Cp,...) by the world. This may be internally analysed further by splitting into various components. Any outside party is hardly ever interested in the breakup.

I referred to a process that is using 95% of the tolerance, which is a common occurrence, not "95% process variation."

Sorry, I may have used wrong words, but 95% of tolerance being used by process variation is a common occurance. And commonly, it is considered unacceptable.

My point was that measurement system evaluation should take into account the amount of the tolerance that's being used by the process. Of course, the AIAG methods leave us in a sort of Catch-22 in this regard, as we can't be sure of the process variation if we measure using systems that have not been qualified.

Exactly what I tried to communicate.

Only, MS variation, being a part of overall process variation should remain limited to an acceptable portion of overall process variation. Thats why if process variation covers 10% of tolerance, MS variation cannot be allowed to cover 50%.