I am looking for advice regarding setting up an attribute GR&R study in a continuous process. The actual set up is a vision system (VS) mounted on an injection molding machine. The VS is being validated for its ability to detect a short shot. The VS does not have sufficient resolution to detect defectives by variables data, but by attribute, the VS should work. We are programming the VS with a master (golden part). Anything "smaller" than the golden part will be rejected. We can make shorts by machine adjustment, but how do we indicate the ability of the system to repeat its decision?
Here are my thoughts:
1) Establish acceptance criteria based upon alpha and beta errors. We want 100% of all non-conforming parts rejected, but we can accept 5% of the product to be classified as non-conforming when it is actually conforming.
2) Run the machine and make short shots by machine adjustment. We should keep track of how many "shorts" we have introduced by design.
3) Dimensionally inspect all product (good and bad) using variables data in the lab.
4) Evaluate % agreement between the attribute system compared to the variables system.
5) Compare findings to 100% / 95% criteria.
The variables gage system has been assessed for GR&R and is acceptable at < 20% of tolerance. Because this cannot be replicated, the sample size will be large (N >= 300).
Does anyone see any problems with this approach or perhaps suggest a better method?
Thanks in advance!
Here are my thoughts:
1) Establish acceptance criteria based upon alpha and beta errors. We want 100% of all non-conforming parts rejected, but we can accept 5% of the product to be classified as non-conforming when it is actually conforming.
2) Run the machine and make short shots by machine adjustment. We should keep track of how many "shorts" we have introduced by design.
3) Dimensionally inspect all product (good and bad) using variables data in the lab.
4) Evaluate % agreement between the attribute system compared to the variables system.
5) Compare findings to 100% / 95% criteria.
The variables gage system has been assessed for GR&R and is acceptable at < 20% of tolerance. Because this cannot be replicated, the sample size will be large (N >= 300).
Does anyone see any problems with this approach or perhaps suggest a better method?
Thanks in advance!