Spc Question - When the Cpk value is bigger than 2.0

[hobbyxin]

Dear experts:
I have a question in implementing the process spc , When the Cpk value is biger than 2.0 , did the point must in control area? I have a sample that the Cpk is 2.26, but the Xbar chart has a appearance of noncotorl , why?

[jperez25]

Hi! I take it that you calculated the Cpk and control limits after more than 25 subgroups and all of a sudden one day you took another sample, calculated Xbar and R, and the R value is out of the control limits.

If that's the case, then it simply means that the process where you took the sample from is different than the the one you always had (a.k.a. out of statistical control or special cause of variation)

You can have a very high Cpk value, but whenever a special cause of variation happens, your process will be different than the one you always had, and then the statistics will fall beyond the control limits.

cheers

[Tim Folkerts]

It is also important to remember the difference between control limits and specification limits. Cpk measures how well you meet customer requirements. If the customer has given fairly broad specs, then even a process that is out of control (with points falling outside the control limits) could give a good Cpk (points are all far from the spec limits). On the other hand, if the customer had fairly tight specs, then a process that is perfectly well in control could have a poor Cpk.

For example, suppose the specs say 100 +/- 5 and you are producing parts with X-bar-bar =100.5 and sigma = 0.5. Furthermore, the control chart shows you are in control. The CpK would be a respectable (105-100.5)/1.5 = 3. The control limits of the X-bar chart would be 100.5 +/- 1.5.

Now, suppose you get some new raw materials and you start getting X-bar-bar =99.5 and sigma = 0.5. You are still well within the spec limits of 100 +/- 5, but the control chart will quickly show points out of control.


Tim F

[hobbyxin]

First , thanks a lot for your quickly reply . I like this site.
Attached is a sample , you could check , hope this is useful for you .

[Darius]

First, your process is out of control, the behabiur of the data look like is a continuos process (there is too much autocorrelation between data) so the control limits look smaller than they must be, the points 15 tru 18 look trully out of control (taken the difference between 2 consecutive data on the chart as a pattern) .

There are 2 ways to include autocorrelation to your data, one is more time between samples (making the sample bigger or more time between the elements of the sample) and the other is with an autocorrelation index, the question may be if the data must be in 6 elements samples?, if is a continuos process the best chart can be IX-MR.

Agree with Tim about capability (specification limits) and control charts, they are 2 different things. (even not talking about one specs ) , the problem with the calculation of an index in a out of control variable is that you don't know with value can be obtained next (is a wheel of the fortune).

[jperez25]

I am a bit doubtful about your measuring system (it may be just the excel file), but it looks like the minimum unit you can read from it is just 1/8th of the specification bandwith (USL-LSL: 17-9)... that may be another thing to look into.

Javier

[Arvind]

Purpose of capability is to predict future state of the process. If for some reason, your process control chart is showing out of control condition, which indicates presence of special cause, it does not make sense to talk of capability.
Always use capability index only when process variation is under common cause.

[hobbyxin]

Originally Posted by Arvind

Purpose of capability is to predict future state of the process. If for some reason, your process control chart is showing out of control condition, which indicates presence of special cause, it does not make sense to talk of capability.
Always use capability index only when process variation is under common cause.

Originally Posted by jperez25

I am a bit doubtful about your measuring system (it may be just the excel file), but it looks like the minimum unit you can read from it is just 1/8th of the specification bandwith (USL-LSL: 17-9)... that may be another thing to look into.

Javier

May the measuring be the probable cause for this uncommon , on the other hand , the process out of control may be the another cause . We studied this further and found the GAP (key character)has a unstability variance due to the machine's shaking in process .
Thanks a lot for your directing.