Nested MSA question

[#1]

Hi All,

I have a question about Nested MSA. I need to understand what is the Part-to-part variation and the measurement system variation. But...

• The parts are destroyed once measure by the said machine
• There is some variation in every part and the measurement system in question is the only one that can measure it
• We know the measurement system is not stable

So, how can I calculate if there is a systematic variation (error) that the measurement system is causing? Any guidance on this would be deeply appreciated!

[#2]

We do have a Golden Sample or ideal specification for the parts in case that makes a difference...

[#3]

Here are my thoughts:

1. It will be impossible to separate measurement errors from part-to-part variation in a destructive test.

2. How can you know bullets 2 & 3?

3. The strategies to handle this situation are summarized/described in the attached document. 

[#4]

Hi Vipul,

There is a Mastering JMP talk I gave last year that demonstrates 2 strategies for destructive test MSA. As Bill stated, you cannot directly measure variation from repeat measurements as the sample is destroyed before repeat measurements can be made. The way to get around this is by using pseudo repeats.

The first method I show in the talk assumes perfect samples. Test multiple samples from a lot/batch and assume that they are equal. Do this across as many batches as the experiment design demands. This is often referred to as aliquoting.

The second method is to estimate part to part variation using a perfect (non-destructive) test. Test each sample multiple times with the non-destructive test before running destructive testing. A relationship (calibration curve) between the non-destructive and destructive tests must be established in advance.  You will need the curve to report the non-destructive repeats on the same scale as the destructive test. There are sample data tables uploaded to the Mastering community page that you can play with. 

There is a 3rd circumstance I did not cover. That is when the property being measured in a test changes over the duration of the test. In these cases, each test would generate a curve. There is an excellent Discovery talk that covers this circumstance. Look for the link in the references section on the Mastering page.

I find destructive MSAs to be a lot of fun. They definitely require a bit of ingenuity when setting up, running and interpreting the results. 

Happy testing!!