Hi to all,
Could someone help me to confirm the correct/estimated sample size used to do an Attribute GR&R?, this means, how many good / bad parts are needed?.
We need to develop an Attribute GR&R for an AOI (Automatic Optical Inspection).
Hope you can help.
Thanks
A common standard for a Continuous GR&R study is to use 10 parts, measured by 3 different people, 3 times each, giving you a total of 90 results.
For Attribute data GR&R studies more data is required (Due to Attribute data having less resolution) therefore at least 20 parts should be assessed at least 3 times by each appraiser giving 180 results.
You should also ensure your selection of parts includes some borderline products to challenge the abilitiy of the measurement system.
Hope this helps
Jim G
there are a couple of approaches for sample size that work for attribute (aka categorical) data guage R&R studies. The best approach is dependent on what you are trying to accomplish with the study.
Check out this attachement which details these approaches and the appropriate statistical analyses. Categorical Data MSA
. for what it's worth - I dont' recommend 'reading' each unti three times. twice is sufficient and ti allows you to increase your sampel size appropriately without increasing the amount of work the subjects have to do. This is particularly true for categorical data as most of the appropriate statistical tests are based on two readings for each unit and tehy get very complicated very fast when you increase the number of readings - without a corresponding increase in the usefulness of the data results.
Remember that a R&R study is trying to determine the measurement error, which is expressed in terms of the standard deviation. This standard deviation is from teh repeated measurements of a single unit. the improvement in the SD estimate of a sample of 2 to a sample size of 3 is negligable. However, increasing your unit sample size from 10 to 30 has a significant improvemetn in the estimate of the measurement error by virtue of using 30 sample SDs instead of 10 sample SDs.
All the info you provided is good, but nobody answer my question, I will try to explain in a another way.
What I want to know is how many parts (good & Bad) I need in order to do an attribute GR&R?
I need to do a GR&R in the Automatic optical Inspection (AOI), this equipment is detecting the lack of electrical components, in order to do the GR&R, what is the standard quantity of good pcs & the standard quantity of bad pcs in order to do the GR&R, for example if we use 30 parts, how many parts needs to be good & how many neds to be bad parts?
Currently I'm selecting between 20 to 30 test samples that represent the full range of variation encountered in actual production runs. For maximum confidence, I'm using a 50-50 mix of good/bad parts but also a 30:70 ratio is acceptable. But I'm doing this according to my experience, but I need to know if there is something in a procedure, MSA, Customer requirements in order to ensure I'm following a standard sample.
Please advise, any comments are very welcome.