GR&R in A.T.E. (Automatic Test Equipment)

G

G.Pito

Dear All,

First of all I apologize for my "stupid" questions but I'm quite a "newbie" in SPC.

This is the use case: my company produces Automated Test Equipments where the operator contribution to variance is (almost) null. Since most of the test Stations we produce are made of several (up to 40) Test bays (we call them "Sockets" like in NI TestStand terminology) in my view the reproducibility concept should be linked to Sockets instead of Operators (Appraisers).

My questions are:
1) Does it make sense to use Sockets/Test Bays instead of Appraisers for Gage R&R (Average and Ranges method) in ATE?

2) If "yes", How can I Manage the (possible) high number of Sockets? So far I've been using max 3 Appraisers, and as far as I know in MSA the maximum number of appraisers is 3, so can I use for example 10 Sockets, 10 Parts and 3 trials?

3) Often I only need to verify the correlation among test sockets, which means the same UUT tested N times on N sockets, should show minimal measurement differences among sockets. What is the name of this analysis in SPC terminology? I need to evaluate results with N Box Plots or a R Chart.

Thank you in advance!
Giovanni
 

Miner

Forum Moderator
Leader
Admin
Re: GR&R in A.T.E.

This is the use case: my company produces Automated Test Equipments where the operator contribution to variance is (almost) null. Since most of the test Stations we produce are made of several (up to 40) Test bays (we call them "Sockets" like in NI TestStand terminology) in my view the reproducibility concept should be linked to Sockets instead of Operators (Appraisers).

My questions are:
1) Does it make sense to use Sockets/Test Bays instead of Appraisers for Gage R&R (Average and Ranges method) in ATE?

Yes. There are several older threads where I respond to this question.

2) If "yes", How can I Manage the (possible) high number of Sockets? So far I've been using max 3 Appraisers, and as far as I know in MSA the maximum number of appraisers is 3, so can I use for example 10 Sockets, 10 Parts and 3 trials?

If you have access to software such as Minitab, you can specify the number of sockets/test bays. If you use Excel, I have not seen an Excel file with more than three operators. I have seen some with more than 10 parts, but not operators. You do not need to include 100% of the sockets/test bays in your analysis. The three that you do include are intended to be selected randomly as representation of all of them.

3) Often I only need to verify the correlation among test sockets, which means the same UUT tested N times on N sockets, should show minimal measurement differences among sockets. What is the name of this analysis in SPC terminology? I need to evaluate results with N Box Plots or a R Chart.

I would used an Analysis of Means (ANOM). This tests the null hypothesis that none of the sockets differ from the overall group mean.
 
G

G.Pito

Re: GR&R in A.T.E.

Thank you very much Miner,
about question 2: I've been asked to write specification for our SW Development dept. in order to develop an SPC tool to be embedded in our ATE Software Framework. I know Minitab is very powerful but we need only few features and to have them embedded in our Test Framework. I'd be "free" to ask our Sw Develop. Dept. to allow any possible number of Appraisers/Sockets but the important thing in my opinion is to understand if there is any costraint for that "parameter" or any algorythm that requires to increase/decrease others parameters when increasing the appraisers number.

For question 3:
Thanks! I'll try to investigate and understand exactly how this method works.

Thank you so much!
 

Miner

Forum Moderator
Leader
Admin
If you use ANOVA to perform the analysis instead of the Xbar/R method, there are no theoretical constraints to the number of sockets, parts, replicates used.

If your purpose is to estimate the relative contributions, the most important factor is the number of parts used rather than the number of operators or the number of replicates. See my blog, parts 2 and 3 on the linked page, for a simulation study by Minitab on this question.
 
Top Bottom