My challenge that brings me to the community has been attempting to statistically improve control over a wire finishing process. I've watched the pq systems two part webinar on control charts and capability and it had some insights that should be valuable. However I still have some questions due the complexity of the process.
So, I have a part measured for coating thickness over wire in four places, each measuring location will have a different mean by virtue of its physical location and orientation, the objective of course is to reduce the variation as there is just one specification.
Now I have 3 different substrate materials in 5 different gages that have a tolerance for diameter of their own that I'm measuring the coating thickness over with a caliper. In addition say I have X different coating materials and Y possible positions inside the coating equipment itself.
That's before I get into any of the other more complicated process parameters of the equipment itself and the thermodynamics involved. Let's ignore those for now, I can't address them here anyway.
Currently I use an xbar and range chart that I can filter by date, part#, substrate material, and coating material. It has four data columns, one for each measurement location on any given piece.
In the past I have sampled several parts a few times a day, the webinar points out that this is actually not as good at capturing my variation as more frequently sampling just one part. Resulting in control limits that are tighter than they should be, but the process would still be out of control regardless.
My question is if I'm going to reevaluate the design of the control chart and the way I measure and sample, should I instead chart this as individuals and moving range if the end goal is to communicate just the one spec?
For example (chart includes some special causes and experimental results over one substrate type):