GR&R Destructive Data Analysis

[brauenbuehler]

Hi Again Everyone,

Another GRR question. I am writing a protocol for a GRR that requires the analysis of destructive data.

I currently have justified using 90 total parts divided equally among 3 operators to match our justification of 10 parts x 3 operators x 3 repetitions for non-destructive variable data. First off, is this justification robust?

Second, when it comes to data analysis, I've proposed the following:

Does this analysis strategy hold any ground? What would you all recommend?

Thanks for the advice,
Ben

 

[brauenbuehler]

I should mention also that I am trying to run a crossed GRR analysis on Minitab

 

[optomist1]

always helps to know the parts, device etc...pls share info, as well as industry if you can...

Cheers Optomist1

 

[rmyjq]

It is good, can refer, Measurement System Analysis (MSA), Fourth Edition, Chapter IV - Section A --- Practices for Complex or Non-Replicable Measurement Systems, Section C Variability Studies, V4: Split Specimens (General).
Then using the ANOVA method or Standard GRR techniques to do data analysis.

 

[Miner]

Hi Again Everyone,

Another GRR question. I am writing a protocol for a GRR that requires the analysis of destructive data.

I currently have justified using 90 total parts divided equally among 3 operators to match our justification of 10 parts x 3 operators x 3 repetitions for non-destructive variable data. First off, is this justification robust?

Second, when it comes to data analysis, I've proposed the following: View attachment 27950

Does this analysis strategy hold any ground? What would you all recommend?

Thanks for the advice,
Ben

I would need to know more about the process to determine whether parts may safely be assumed to be the same. Also, your wording could be interpreted to say that all 90 parts are identical, which would not be good. You want the parts within each set of 9 to be identical, but want differences between the sets of 9.

I should mention also that I am trying to run a crossed GRR analysis on Minitab

Use the nested GRR analysis for destructive tests.

 

[brauenbuehler]

always helps to know the parts, device etc...pls share info, as well as industry if you can...

Cheers Optomist1

I work in the Medical Device Industry. When I mentioned protocol, I should have included that it is a protocol template. We make a very wide variety of devices for small batch manufacturing. There isn't a specific device in mind but I'm working to generate sections of the document that can be removed if not applicable.

I would need to know more about the process to determine whether parts may safely be assumed to be the same. Also, your wording could be interpreted to say that all 90 parts are identical, which would not be good. You want the parts within each set of 9 to be identical, but want differences between the sets of 9.



Use the nested GRR analysis for destructive tests.

After reading what Miner said, I'm not sure of anything that this would truly apply to within what we do, however it's good to build in the option.

Would we still use a nested study if we make the assumption that the parts are identical or is the nested study what we use if we CANNOT assume they are identical?

 

[brauenbuehler]

It is good, can refer, Measurement System Analysis (MSA), Fourth Edition, Chapter IV - Section A --- Practices for Complex or Non-Replicable Measurement Systems, Section C Variability Studies, V4: Split Specimens (General).
Then using the ANOVA method or Standard GRR techniques to do data analysis.

This is where I was also pulling my information. I wanted to double check that my understanding was sound.

 

[Miner]

Would we still use a nested study if we make the assumption that the parts are identical or is the nested study what we use if we CANNOT assume they are identical?

If you can measure the same parts multiple times, use a crossed study. If you cannot measure a part more than one time, use a nested study.

 

[Miner]

This is where I was also pulling my information. I wanted to double check that my understanding was sound.

Section A (p. 154) lists two methods that may be used for destructive R&R studies (i.e., V3 or V4). Both methods use one of two approaches:

• Split specimens - The product is subdivided into multiple specimens, each of which are tested (e.g., cutting a continuous extrusion into small lengths; die cutting tensile specimens from a sheet of material; sampling 5cc samples from a 100L batch of chemical, etc.).
• Consecutive parts - When the process is very homogeneous, consecutive parts may be used instead (e.g., precision machining, stamping, etc.).

Both of the above methods are used to provide the multiple measurements required for an "individual' part. The samples for the next "individual" part should be taken from a different "batch" of material or after an extended length of time between..

 

[brauenbuehler]

Section A (p. 154) lists two methods that may be used for destructive R&R studies (i.e., V3 or V4). Both methods use one of two approaches:

• Split specimens - The product is subdivided into multiple specimens, each of which are tested (e.g., cutting a continuous extrusion into small lengths; die cutting tensile specimens from a sheet of material; sampling 5cc samples from a 100L batch of chemical, etc.).
• Consecutive parts - When the process is very homogeneous, consecutive parts may be used instead (e.g., precision machining, stamping, etc.).

Both of the above methods are used to provide the multiple measurements required for an "individual' part. The samples for the next "individual" part should be taken from a different "batch" of material or after an extended length of time between..

Okay then a follow up question would be: What if we can't assume our parts are identical? I'll use an example of a UV bond that we tensile test. If there is slight variation in the amount of adhesive used then the parts would not be identical. I would say that most of our testing would actually fall into this category. What would you recommend in the event that the identical part assumption can't be made?