Re: Centerless Grinding (SPC)
This chart is a classic example of a gage that has inadequate resolution to see the process variation. The true process variation is probably stable with the one exception where the wheel was dressed.
If SPC is a requirement for this process, you will need a gage with better resolution.
You did not state the tolerances that the process must hold. If the process capability is high, you may not need SPC.
What Miner has posted is a thoughtful post. (my fellow cove members: when the OP has the forethought to post their data, we should look at it. It usually tells the story so we can provide helpful answers or at least set us up to ask insightful questions)
this data certainly tells a story.
let's start with the gage resolution: the variation in the data is only .0004 (that's at the 4th decimal place). Given this the control chart can only look out of control because the data is "chunky" - very few possible results given the gage resolution and the variation of the data. It behaves like ordinal data. IF the parts were to vary .004 then the resolution would be sufficient.
Getting a gage with better resolution will be expensive tho - this is a very small dimensional difference, your average caliper won't do it and a CMM will have potential fixturing issues.
This leads to my next point, spec limits. Miner is again correct to ask this question. if the spec limits are very large compared to the range of data provided we have awhole different situation. if teh OP could provide that we will have more to work with.
Next, is the concept of rational subgrouping. taking 30 random parts then subgrouping wthout regard to how the parts were taken (you shoud lalways specify and know how the parts are taken; this is
essential to proper informative usefull charts) Subgroups are typically sequential parts or values. (there are times when they are not but this is a delibrate decision in order to create a rational subgroup) . Additionally, your first set of data shodul span most of teh variation in the process BEFORE you begin calculating limits. SPC is specific sampling, not random sampling.
My recommendation would be to perfrom a gage R&R (30 pieces that span the full range of variation measured twice) Get yourself a copy of Donald Wheeler's books on measurement error adn go out to his website (SPC Press) and download some of his articles on measurement error and it's effect on SPC. My experience is that he is right on the point.
Also buy two books on SPC: Wheeler's "understanding statistical process control" and Wise's "innovative contro lcharting". then read them. These books will be invaluable to your understanding and the deployment of a successful program.