We are monitoring component flow rates. We have determined that flow rate variation over a 30 second period would be relevant to our product and have set up to capture every 2 seconds. I have attempted to process using Xbar-S
Tried to insert as a link to picture of chart but post count prevented this
What we see is a cyclic pattern as the flow rate drifts away from its set point and in then corrected by the machines logic processor. Because this drift is gradual we end up with very low short term variation. So we have a low sbar which then plays havoc with the control limits on the xbar chart
We could attempt to decrease the threshold of drift the machine reacts to but the variation is already low enough where we wouldn't be concerned.
However we still need a way for operators to identify special cause variation.
I'm sure this can't be a unique problem so I am looking for any experience, advice, solutions.
I think I might have answered my own question here but if anyone could verify.....
I am misunderstanding the concept of rational subgrouping. These are supposed to be measurements taken under the same conditions and an I know the flow is being manipulated by the logic unit this cannot be the case.
So since I am interested in average flow over a 30 second period, I should charting that directly. e.g. getting that 30 second average straight from the machine or calculating it first, then using I-MR, thus giving me direct comparison to see if the flow is stable between 30 second periods.
I then get realistic control limits and identification of special clauses
Just a comment: SPC stands for Statistical Process Control. You are already using an logic processor, which provides automated process control. SPC would be redundant and non-value add (unless I am missing something). The automated controller would also tend to create unusual patterns that would trip some of the extended rules for SPC.
You are correct that the process is being controlled but I am wish to see if that control is stable.
This is a chemical application and it is difficult to measure and detect variation in the final product. We want to make sure our inputs are in control as much as possible and must be stable. So if we start to see worn valves/lumps in the material being pumped etc. these would have an impact on the flow rate separate to the logic controller
the second approach you describe is theoretically correct.
when your average is not controlled by the same factors that control the sequential event readings 'traditional' shewhart charts won't work directly - you need to consider a rational subgrop scheme.
I have many flow systems like the one you describe and have successfully applied SPC to many of them.
Rational subgrouping requires you to put the variation that you want to control between subgroups inside the subgroup. This requries us to understand both the physics of the specific system at hand as well as its actual variation...
could you post soem of your data in an EXCEL file?
Seems to me we could use X-individuals charts to at least identify if the controller keeps the results within a predictable band.
Is there a way to identify when the controller applies a correction and in which direction? Sometimes controllers can "hunt" and keep applying a change in one direction and then a few cycles applies the opposite change and keeps cycling back and forth where you may be better off releasing the tolerances on the controller. But certainly a time-series plot, with annotations of when the controller takes action could be very useful for verification that the controller is adding value.