MSA (Measurement System Analysis) Study on an Automated Process

[Edward_Nell]

How would I go about conducting a MSA study on a automated process where the operator loads a part, the machine resizes, measures and determines acceptable and rejection, as a customer whats me to do this on the machine as per AIAG MSA guidline

 

[qusys]

Re: MSA study on a Automated process

How would I go about conducting a MSA study on a automated process where the operator loads a part, the machine resizes, measures and determines acceptable and rejection, as a customer whats me to do this on the machine as per AIAG MSA guidline

Hi,
I think that there are some environment where the influence of the operator is not determinant. For instance, it could be the case of the seminconductor environment.
There are some industry sector where the MSA studies are equipment predominant rather than worforce predominant, expecially where there is a strong automated processes.
My advice could be that to be helped by the customer, whether it has experience in this sense. I do not see other ways. Try to get collaboration with the customer and find an agreement .

 

[NumberCruncher]

Hi Edward

Just do a "textbook" MSA study. All it will show is that your reproducibility (between operators variability) is statistically indistinguishable from zero, and if your machine is good, it will show that your repeatability (within sample variability) is wonderful. Then you can report to your customer that your measurement system is wonderful, "and here are the numbers to prove it".

You say that the machine "...determines acceptable and rejection...". In this case it sounds like an attribute study so page 131 of the MSA manual onwards (4th edition).

NC



Apologies for momentarily hijacking this thread, but does anyone know where AIAG got their numbers from on page 191, paragraph 2?

"...the expected median is about 0.861 with a standard deviation of approximately 0.439."

Huh???

It's been a while since I last tried this. But the last time I put the numbers in the table (page 190) into a spreadsheet, I failed to get this median and standard deviation no matter how ingenious I was (although that may not be very ingenious!). Are the editorial board of AIAG secret Bayesians with the "expected median" representing a prior probability estimate (ie a random guess)?

Any input welcome.

 

[bobdoering]

Just do a "textbook" MSA study. All it will show is that your reproducibility (between operators variability) is statistically indistinguishable from zero, and if your machine is good, it will show that your repeatability (within sample variability) is wonderful. Then you can report to your customer that your measurement system is wonderful, "and here are the numbers to prove it".

You say that the machine "...determines acceptable and rejection...". In this case it sounds like an attribute study so page 131 of the MSA manual onwards (4th edition).

AIAG specifies that an automated device can use Gage R (since their is no reproducibility), but it can not be used alone. Typically they expect you also do stability. I do a two-fer and just run the speciments through the machine 3 times, and then do it again later (providing a time lapse which emulates stability over time) and then even later again for the 3rd trial. Then, plug it in the standard sheet.

Apologies for momentarily hijacking this thread, but does anyone know where AIAG got their numbers from on page 191, paragraph 2?

"...the expected median is about 0.861 with a standard deviation of approximately 0.439."

Huh???

It's been a while since I last tried this. But the last time I put the numbers in the table (page 190) into a spreadsheet, I failed to get this median and standard deviation no matter how ingenious I was (although that may not be very ingenious!). Are the editorial board of AIAG secret Bayesians with the "expected median" representing a prior probability estimate (ie a random guess)?

It's not clear (as usual) if it is the median of all of the data of the whole k sheet, but if so, you have an even number of data points. So, you take the middle two and average them to calculate the median. That's why whenever I do median charts, I always use an odd number of samples.

 

[NumberCruncher]

Hi bobdoering

Good point about the stability. Thanks

It's not clear (as usual) if it is the median of all of the data of the whole k sheet

And I thought it was only the Brits who did understatement.

NC