T
To begin with, you have a really sweet looking report, and
If its alright with you, lets try to tackle one thing at a time.
Given that you have some number of tablets in each sample and multiple samples per campaign and multiple campaigns per APR.
I can understand @gzmorgan0 's comment about a variance type report, it would be a smart thing. But how to do that in this context?
It might be a good idea to get the table with all of the individual individual tablet weights:
These Columns: Campaign, Sample, Weight. Use an X-bar chart with R charts for between and within variation (three way chart) I suspect this will fix the overly tight control limit issue that you might have originally seen with an x-bar chart that caused the selection of Levey Jennings charts. In the x-bar chart use campaign as a Phase variable.
That chart might look like this:
The second control chart might compare the average sample weight to a historical, fixed, control limit from a time when the process was stable and in control. This is not a three standard deviation control limit, but a 3 sigma limit base on the moving range, normal control chart limits.
That chart might look like this.
Now about normality and control charts.
I'm about to get "Wheeler" on you. You can definitely plot non-normal data on a control chart without normalizing it (with exceptions for very lognormal data from something like a biologic potency test.)
The distribution model matters for Ppk but not for control charts. Ppk is baed on the ratio of 3 standard deviation between the spec limit and the process mean. If your data is not normal, then using standard deviation, which assumes your data is independent and identically distributed is going to give you inaccurate Ppk estimates.
The tolerance interval is a curious criteria to apply. An interval with alpha 0.95 and proportion 0.99... is saying I'm 95% sure 1% of the tablet weights will be outside this range. You'll have to check this for yourself, and without getting into it, 3 sigma is always within that interval. I don't know the rationale behind including it in the report, I'm sure its interesting. By culling thel "outliers", the data will fit into those limits most of the time; however, If I was auditing your facility and I saw that your tablet weight consistency was based on data with the outliers removed before reporting, that would open the door for a lot of other exploration. It might be better to report all the data and means of the samples, just like in an x-bar chart?
JMP Systems Engineer, Health and Life Sciences (Pharma)