Hello bob,
I just want to clear out my doubt as I'm quite confuse with subgroup size definition in doing capability study, so, you mentioned should 1 and according to gcpa81, 50 parts and need to measure 15 significant characteristics, is that mean we take only 1 part out of 50 and measure the 15 characteristics?
If he is doing a capability study where he runs 50 parts in a row through a machine, then measures the parts, then for any statistics used to analyze to data the subgroup should be 1. If you were to chart the parts and a traditional Shewhart chart was appropriate, then you would use I-MR. Some folks like to dream up a subgroup - maybe 4 or 5 - then use X bar -R or similar calculations. but, logical subgroups and made-up-to-make-software-work subgroups are two very different things. Only one is truly useful.
Of course, as an aside, no matter what process it is, you should do both a run chart and do curve fitting. Always get the data in time order sequence. If you find multi-modal variation, you will want to see if it is a time function. Dumping data into software and chugging out a "normal" curve, Cpks and Ppks is backyard stuff. The true distribution may not even support Cpks or Ppks. Also, it is critical that the measurement system not contribute to the distribution, or it may mask the true process distribution. Poor measurement technique and gage R&R can make a lot of non-normal distributions
look normal by masking. So will overcontrol.
Interesting, huh?