Understanding Subgroup Size - Multi Cavity (Minitab)

[SLEEPYMUM]

Hi Folks,

Hope someone can help me, myself and a colleague cannot come to agreement on what the subgroup size should be when calculating Cpk in Minitab.

We have taken 10 samples containing 32 parts (32 cavities in the tool, each sample contains 1 part from each cavity).

When calculating individual cpk values from the above, I enter in the field "cavity" as the sub group size as this is the subgroup that I want to analyze.. Is this correct

When I want to look at combined Cpk should I enter in 32 as the subgroup size or 10 or cavity? - This is the one that is causing the most confusion :mg:

 

[Miner]

There are two issues here. The first is that you recognize that you are mixing 32 process streams into a single analysis. This may or may not cause a problem in your analysis. It will depend on how similar each cavity is to each other. I recommend that you analyze the variation from cavity to cavity vs. over time to see whether this could be an issue.

As to the second issue. This will depend on the first issue. A subgroup is comprised of those parts collected during a very small slice of time. Again we have several possibilities.

Scenario 1: the 10 process cycles were spaced out over time: If you choose to keep your process streams mixed, your subgroup size would be 32 (all cavities from a single cycle). If you choose to separate your process streams, your subgroup size would be 1 (a single cavity from a single cycle).

Scenario 2: the 10 process cycles were consecutive: If you choose to keep your process streams mixed, your subgroup size would be 320 (all cavities from all cycles). If you choose to separate your process streams, your subgroup size would be 10 (a single cavity from all cycles).

In no scenario would you use cavity as a subgroup. I do recommend that you statistically analyze whether you have differences between subgroups using ANOVA. If you can attach your data, we can assist in this.

 

[SLEEPYMUM]

Thanks Miner for your fast response.

Please see attached data as requested.
As part of validation we have been asked to complete Cpk analysis by both cavities and combined.

All of our data is in spec, but there is a lot of variation due to some cavities being much lower than others, this is why it is failing commbined Cpk but passing on individual cavities. Hoever depending on sample size used we are almost their criteria.
I dont want to transform data unless really necessary, but I have made them aware of the cavitiy signal seen so I dont seen the need to transform it.

 

[Miner]

Your greatest source of variation is cavity-to-cavity (see boxplot). The shot-to-shot variation was very small in comparison.

Are these dimensions fixed or closure dimensions?

If fixed, you may need to do capability by cavity then close off non-capable cavities or segregate and sort if required. If they are closure dimensions, you might be able to remove material at the parting line of the mold if this will not throw out other dimensions.

Do not transform the data. This is only an option when the data are naturally non-normal such as naturally skewed. When the cause is mixed process streams, you should never try to transform the data.

 

[Bev D]

i would also ask f this is a mod that will wear such that over time the values will naturally decrease as the tool wears...when this common conditions occurs it is standard practice - and a very good practice - to start at the spec limit from which the tool will wear away from. this makes Cpk values even more meaningless than normal...