iSixSigma

CPK of a process with X bar moving over time

Six Sigma – iSixSigma Forums Old Forums General CPK of a process with X bar moving over time

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
    Posts
  • #51236

    TingaMicas
    Member

    Hi Dear all
    I would like to learn how to calculate a Cpk of a process which mean value (X bar) is moving up over time. When the process starts running the variation is centered close to LSL, but along over time and due to a process charateristic the mean is going towards the upper limit USL. Calculating the Cpk using a normal distribution results in a low Cpk value, because the process takes almost all the values between the low and high limits. Taking a picture of the process, i.e, taking a very short sample the process reveals a normal distribution but this sample is too short to be representative of the process variance.
    Regards
     

    0
    #177285

    Pramod
    Participant

    Hi Tinga,
    As I understand, you need to see the centering rather than the variation here in this case (consider Cp instead of CpK). Check for your target mean of the process and see whether the actual mean is away from it.
    Also share how does it look like in the curve? It might look like a long tailed / skewed process variation?
    Best Regards,
    Pramod
     

    0
    #177291

    TingaMicas
    Member

    Hi Pramod
    Yes, the shape of the process is a long tail with not so much variance towards the upper limit.
    Each time the values reach close to the upper control limit, the process resets and returns to lower values close to lower control limit. Then due to a erasing process, the mean increases up and up. Its a kind of a cycle process.
    To be true, all the readings are inside the spec limits, however due to the fact that process mean is increasing over time, the histogram returns a wide bell shape curve. When calculating the Capability it results in a poor score.
    If I look on the centering I will get also a poor score on Cp due to the fact that after 20 samples I’ve at least got values from 2 to 3 process cycles. Each cycle sweeps values from bottom to top chart.
    Shoul I perform some Non distribution calculation exercise ? How to do it ?
    Looking forward to your reply. Regards
     
     
     

    0
    #177294

    Brian M
    Participant

    If your mean is trending, then the process in not in control and the Cp and Cpk will be of little value. You need to figure out why the process is trending and bring it under tighter control.My $0.02

    0
    #177311

    Pramod
    Participant

    Hi Tinga,
    I believe there is a sporadic change in the process because of which you get a spike. This is normally due to the special cause variation in the process, please check for any abnormalities occuring. If you find that, ask the subject matter expert for the solution. I would suggest to go ahead after the special cause elimination.
    Best Regards,
    Pramod
     

    0
    #177343

    tungamicas
    Member

    Dear Sirs
    I do confirm that the mean is trending. This must be considered a normal cause of variation because the process is dependent on a paint coating that gets thicker along time. That’s my problem. How to measure a process that has a characteristic like this.
    Looking forward for your replys.
    tinga
     
     

    0
    #177344

    Rhineg
    Member

    Hi Tinga,I suggest that you change your perspective from internal to external. Your customer does not care about your internal process difficulties, they only want product which is reliably within USL and LSL and they might even benefit from using less of that full range. Your low Cpk is a legitimate indication of how much of your customer’s needs are in jeopardy with this drift and reset process.You might need to look into a method to continuously keep your paint viscosity uniform, rather than batching the adjustments. {easy to say, but includes lots of engineering requirements (e.g. is there an incomplete reaction, do you have solvent loss, etc.,…).Just my 2 pence.

    0
    #177358

    will
    Participant

    Hi Tinga,
    You have a process drift driven by some assignable cause which will be a thorn in your side if it is not understood. Your long term Cpk will always be poor and possibly the short term too for most observation sets, unless you are taking sample observations at the time the mean is near the target. This assignable cause must be understood. As suggested by someone else, get a process expert involved to understand the root cause (assignable cause) and see if there is away to control it.
    Cyrus

    0
Viewing 8 posts - 1 through 8 (of 8 total)

The forum ‘General’ is closed to new topics and replies.