Search the Elsmar Cove!
**Search ALL of Elsmar.com** with DuckDuckGo including content not in the forum - Search results with No ads.
  • Google has changed ad sizes for some reason. I am investigating and will get the sizes reduced to what they have been until now. I apologise for the inconvenience.

Process Control Chart - Capable, but not stable question

C

CMfgT

#1
I have a control chart for a customer attached with Data and supporting CpK.
I have a control chart that is out of control, so technically the Cpk value is not valid because you have to be stable before you can be capable. Any thoughts on this. I think it is beating a dead horse to try and be stable with a high cpk.


FYI: the process is grinding an OD on a CNC Grinder.

Thanks,
 

Attachments

R

Rob Nix

#2
Re: Capable, but not stable question

Part of the problem here is your gage resolution (or discrimination). You need to go out one more decimal place at least to get fairly valid data. But you are right, a process cannot be considered capable unless it is both in control (stable) and meets the Cpk requirement.
 

bobdoering

Stop X-bar/R Madness!!
Trusted
#3
Re: Capable, but not stable question

I have a control chart for a customer attached with Data and supporting CpK.
I have a control chart that is out of control, so technically the Cpk value is not valid because you have to be stable before you can be capable. Any thoughts on this. I think it is beating a dead horse to try and be stable with a high cpk.


FYI: the process is grinding an OD on a CNC Grinder.
First of all, your process is perfectly capable. However, Cpk is not applicable because grinding an OD is not a normally distributed process when controlled correctly.

Second, you are not out of control and it is stable. - but you need to prove it. First, this is the wrong chart for precision grinding. You should be using the X hi/lo-R Chart. Your control limits should be LCL: 2.99895 UCL: 2.99985. You have no points out of control.

The variation you see is either gage error (insufficient GR&R) or measurement error (from roundness), not process variation. You do need at least another decimal place of resolution to detect that. You have no process variation in 30 consecutive parts on a CNC grinder (unless you have the wrong wheel material, or a great big hard part in a tiny wheel). What you need is to take every 10th part from your 300 pc run and look for the positive slope from the wheel wear. If you see that, you have the classic continuous uniform distribution that you should see if your grinding process is in control. Capability for that distribution is: (USL-LSL)/(UCL-LCL) or 1.33. Don't mess with that normal distribution stuff - it has nothing to do with your process unless it is out of control!

To maintain that control, you need to follow the directions for the X hi/lo-R charting methodology.

Precision grinding is the gold standard for the continuous uniform distribution.
 
Last edited:

Bev D

Heretical Statistician
Staff member
Super Moderator
#4
additional: to assess stability you also need a lot more parts AND they should span a longer period of time. which of course begs the question: what possible value add information are you providing from a Cpk value on so few parts?


some additional clarification: yes a process should be stable for a capability study to be predictive of future performance. and an appropropraite "control chart" is ONE way of assessing this. However as Bob points out many processes do not have a homogenous process stream even when they are stable, predictiable and capable. We must use logic and our knowledge of physics as well as statistics.
 

bobdoering

Stop X-bar/R Madness!!
Trusted
#5
additional: to assess stability you also need a lot more parts AND they should span a longer period of time. which of course begs the question: what possible value add information are you providing from a Cpk value on so few parts?
An example that I use for implementing X hi/lo-R charts is what occurred the first time I convinced a grinder operator to adjust to the lower control limit, and not adjust again until it reached the upper control limit. He had been running to the mean by making adjustments every 15 minutes. By allowing the machine to do its job and not touching it, it did not need adjustment for a week! (Talk about overcontrol!) Imagine what little 30 parts in a row will tell about such a slow wheel wear process! Not much...even less for the range calculated for 5 parts in a row if one was to attempt X-bar R. That's why control limits based on the range of 5 parts in a row are so tight on an X bar R chart - they do not represent the full variation over time as they need to for a normally distributed process sample. In fact, on occasion have to extrapolate the slope determined in a shift's production to determine the frequency of adjustment for very slow wheel or tool wear. Fortunately, none of that affects the capability calculation for the continuous uniform distribution - it is a constant based on the control limits.

Of course, Bev has heard this all before...this is for any readers new to the concept!
 
Top Bottom