iSixSigma

Process Shift

Six Sigma – iSixSigma Forums Old Forums General Process Shift

This topic contains 6 replies, has 5 voices, and was last updated by  Grant Blair 18 years, 5 months ago.

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #27143

    Cone
    Participant

    Anyone out there believe their process do not shift?

    0
    #66098

    Kim Niles
    Participant

    Dear Gary:

    I understand your point and agree that I’ve never seen a process that I’ve worked on that didn’t shift. However, I still believe it must be possible to have a process who’s standard deviation contains all of the variation that the process will ever see. Therefore, it must be possible to have a process that doesn’t shift.

    We must be able to think one up. How about the process of taking a customer survey where a standard form is used with the same types of questions to the same type of crowd….nah, that could shift. How about a very simple manufacturing process that is automated and 6+ sigma capable with data taken over several years….ok it’s possible it could shift. However, there is a probability that it wouldn’t!!

    What about that?

    KN

    0
    #66099

    Cone
    Participant

    Kim,

    There probably is one out there somewhere, but think of the difficulties of keeping a process perfectly centered. If we use SPC to “know” the process is centered, what are the rules for starting the process back up when a new setup is done, when a new batch of material is introduced, when a tool breaks and it is replaced, when your measurement instrument goes out for calibration, …..?

    Juran and Shanain suggest a precontrol method of 5 in a row in green. Proof we are exactly on target? No just proof we are in the right ballpark.

    QS9000 suggests taking a normal SPC sample and if within +/- 1 std deviation, run. If not within 1 but within 3, take a second sample and if it is within +/- 1, run. If not set up again and try again. Proof we are exactly on target? No just proof we are in the right ballpark.

    Now consider some of the other discussion on the board where people are asking about C=0 plans, wondering what LTPD means and other things of this nature. You can conclude from this that most of the folks are still trying to control the inputs from outside sources through the use of incoming inspection and don’t even know how to judge the goodness of their sampling plan! Material inputs exactly on target batch to batch? Want to buy a nice bridge?

    You see the issues.

    Gary

    0
    #66103

    Ken Myers
    Participant

    Kim,

    Using what is called the Average Run Length (ARL) for a sample one can estimate the ability to see a shift in the mean given any set of detection rules over a determined period. When conducting SPC an ARL curve for a given sample can be constructed. Dr. D. Wheeler has prepared such ARL curves for many types of variables control charts in his SPC texts, see http://www.spcpress.com for details.

    The salient point is suppose the process was, in fact, perfectly centered. Suppose no other sources of variation were present that could cause the mean to shift off center. Gary presented compelling reasons why this would be a difficult feat. However, if we supposed the process was perfectly centered how would we know? Usually, we would sample the process over a particular period of time and observe if any values were found outside the control limits. How many samples would be required to show conclusively that the process was perfectly centered? Using an ARL table one could determine how many samples per subgroup would be required to conclusively prove no shift was evident in the mean. For standard control charts the required sample size would essentially be very very large, approaching infinite.

    The bottom-line is we would never be able to conclusively prove that such a process exist…

    Regards,

    Ken

    0
    #66133

    Charlie Pfaff
    Participant

    It has been my experience that how much a process shifts depends on the process itself. A 1.5 shift is excessive. In any process there are variables that will cause an Xbar shift and other variables that will cause the s or standard deviation to increase and decrease. I am sure there are variables that will cause both and xbar shift and changes in the standard deviation. This xbar shift comes from Motorola and the electronics industry. I can understand why they may see a 1.5 shift, but I would doubt it in the automotive or aluminum industry. Shifting is not the serious problem; it is controling the standard deviation.

    0
    #66146

    Ken Myers
    Participant

    Gee Charlie I thought we covered your concerns back in thread https://www.isixsigma.com/forum/showmessage.asp?messageID=1076.

    You’re right, variation is the key to improvement. But, as you know if the process average is inconsistent then variation reduction will not buy you much…

    0
    #66241

    Grant Blair
    Participant

    Motorola was NOT concerned with how much the process could shift. The concern in their study was..how much
    could the process shift before it was DETECTED.
    Simple common sense should tell anyone that if the accepted level for attaining 3 parts per thousand defective=3 sigma and a 1 sigma process shift,
    the accepted level for attaining ~4 parts per million
    will be higher.
    There is a sound statistical and practical basis for
    4.5 sigma units and 1.5 process shift.
    All processes drift…left unchecked, they will drift into oblivion….or 12 sigma units…whichever comes first. If they did not drift, then you would not need any of this stuff, because:
    1. The mere act of measuring would confer process stability.
    2. All processes would have perpetual motion, because the law of entrophy would be repealed.

    0
Viewing 7 posts - 1 through 7 (of 7 total)

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