iSixSigma

1.5 Sigma Process Shifts

Six Sigma – iSixSigma Forums Old Forums General 1.5 Sigma Process Shifts

This topic contains 12 replies, has 7 voices, and was last updated by  Grant Blair 18 years, 5 months ago.

Viewing 13 posts - 1 through 13 (of 13 total)
  • Author
    Posts
  • #27088

    Joe Perito
    Participant

    The 1.5 sigma shift that a number of companies are starting to use is used incorrectly. Please be advised that this number is only useful for gereralizations about Motorola’s processes since it is their processes that were studied and found to have a 1.5 sigma drift over time. This does not mean your process drifts 1.5 sigma. Your process may drift more or less. It is an error for a six sigma black belt, or trainee, to come into your company , run a short term capability study, and tack on an additional 1.5 sigma to the process variation (calculated from the capability study)to determine if your process is capable or not. Motorola’s historical process shift has no relavance outside of their own walls without data to demonstrate the same level of variation is applicable to another company. So how do you qualify your process knowing thet it will have drifts in the means also? This is what the extra cushion in the PpK calculation is for. Customers do not accept the fact that your process just barely meets the tolerances allowed and has no room to move off target. This is why you must have an additional +/- one sigma allowance between the process limits and the tolerances in the PpK requirement of 1.67 or greater. A PpK of 1.67 is a +/- 4 sigma process. The “ongoing process capability” (CpK) is a +/- 3 sigma process capability. Therefore, what the customer is saying: plus or minus 4 standard deviations must fit between their spec limits, or in other words, you must have a PpK of 1.67 or higher. It is inappropriate to add another 1.5 sigmas to it because you have already meet the more stringent requirment which includes process drift.

    0
    #65886

    Joe Perito
    Participant

    Here’s a correction to my own posting on 1.5 sigma shifts: All of the logic is OK in the posting but as soon as I signed off I realized I was not paying enough attention to the math. For beginners in the Quality field be advised that a CpK of 1.33 is a plus or minus 4 sigma process. A process potential study that yields a PpK of 1.67 is a +/- 5 sigma process. Both of these calculations are based on a 3 sigma process yielding 99.73% good quality. The 3 is always in the denominator of the two formulae. Therefore, to get a CpK of 1.33 you divide a (+/-) 4 sigma process by 3 to get a CpK of 1.33. A PpK of 1.67 is a +/- 5 sigma process divided by 3 (5/3 = 1.67).

    0
    #65919

    Kim Niles
    Participant

    Thanks for your posting. This is a very interesting subject to me.

    I see what you are saying but take on the argument that since “Six Sigma” is more of a management system than a statistical metric, any excuse one could find to keep seemingly unrealistic management goals from being obtainable is ok by me.

    Besides that, this subject will always be controversial as it was before Six Sigma came along. Articles date back to 1975 that I know of on whether or not a shift should be applied to process metrics.

    Any other comments or information?
    KN

    0
    #66002

    Cone
    Participant

    There is no debate as to whether processes shift or not. Those who wonder about such things only need to analyze some long term data from any process.

    0
    #66042

    Kim Niles
    Participant

    Dear Gary:

    I agree that processes shift.

    I believe that the 1.5 sigma shift controversy is all over whether or not processes that are “stable” shift or not. The controversy lies in the definition of one word (stable).

    If we have a perfectly stable process that includes all the variation that the process will ever see in the standard deviation calculations then adding a 1.5 sigma shift means that we have a 4.5 sigma process at 6 sigma.

    Right?!?!??!

    KN

    0
    #66135

    Charlie Pfaff
    Participant

    I may be getting lost with six sigma, but I thought the “new” cpk is done over a short term part of the process, while a ppk is done over a long term part of the process where all the variables of the process pertake in the data. A short term is where a minimum amount of the variables effect the data. Is this correct. A Ppk of 1.67 would probably be better than a Cpk of 1.33, but according to Motorola a cpk could undergo a shift of 1.5 standard deviations. I do not agree, it depends on the process. In parts manufacturing alignment and operator influence would all be in the short term. Maintenance, possibily tool life and coolant changes would not. If tool life was in the short term, I doubt if there would be that much of a shift and maintenancer variability also usually shows up in affecting the standard deviation not the X bar shift. “Short term and Long term” need to be explained to a more specific degree. If I run my process and all the variables of the process have a chance to affect my data; if I have a Ppk (old Cpk) of 1.67 my process is in great shape.

    0
    #66136

    Ken Myers
    Participant

    Charlie,

    You are correct in your thinking about Cpk and Ppk. There is no change here. Not sure what is meant by the “new” Cpk, but the your thinking appears reasonable to me. Cpk is a measure of process capability using short term variation, and Ppk is the same measure using long term variation. Certainly, a Ppk of 1.67 would be better than a Cpk of 1.33.

    Much discussion has come and gone concerning Motorola’s +/-1.5 sigma shift. We have all covered this subject extensively in the last week or two, and most are probably exhausted talking about it further. If you view the message board closer I believe you will find your thoughts have generally been conveyed.

    As Gary Cone and others have indicated if you do not have any long term data supporting the process, then the +/-1.5 sigma shift advocated by Motorola can be used as a conservative estimate. As you collect the additional data necessary to compute the long term variation a better estimate of the process shift can be made. Hopefully, I’ve done a reasonable job of conveying our collective understanding here. There is considerable controversy and emotion surrounding this topic, but I believe most would agree with my comments above.

    Good Luck,

    Ken

    0
    #66137

    Ken Myers
    Participant

    Kim,

    Yes, you are correct…

    Ken

    0
    #66142

    Cone
    Participant

    Amen. If you really have Ppk’s of 1.67 for all your critical processes, you are in great shape. If fact, if you have such a situation, I would pay my own expenses to see it. You would be better than any of the folks implementing Six Sigma

    0
    #66150

    Cone
    Participant

    Charlie,

    You are right, but I want to make a couple of points on what you wrote.

    “I may be getting lost with six sigma, but I thought the “new” cpk is done over a short term part of the process, while a ppk is done over a long term part of the process where all the variables of the process pertake in the data.”

    While this is technically correct, I would suggest that “new” Cpk is meaningless except where you have a one sided spec, off center target, or some type of process where the mean cannot be economically adjusted to center. In all other cases Cp is important, not Cpk.

    “A short term is where a minimum amount of the variables effect the data. Is this correct?”

    This is correct.

    “A Ppk of 1.67 would probably be better than a Cpk of 1.33”

    Orders of magnitude better.

    “but according to Motorola a cpk could undergo a shift of 1.5 standard deviations. I do not agree, it depends on the process.”

    Absolutely, only assume 1.5 when you do not have real data to tell you what to expect.

    “In parts manufacturing alignment and operator influence would all be in the short term.”

    This is only true where all management and operators are trained and guaranteed to be consistent prior to being left alone to do their jobs. If not, this is a huge source of long term variability.

    “Maintenance, possibily tool life and coolant changes would not. If tool life was in the short term, I doubt if there would be that much of a shift and maintenancer variability also usually shows up in affecting the standard deviation not the X bar shift. “Short term and Long term” need to be explained to a more specific degree. If I run my process and all the variables of the process have a chance to affect my data; if I have a Ppk (old Cpk) of 1.67 my process is in great shape.”

    Absolutely true. Short term is just snapshots of your process. Long term is everything. The 1.5 Motorola suggests is usable for two things. It is a rational assumption if you do not have real data, but have basic controls in place including a guarantee of management and worker training and alignment. It is a gross underestimation if the guarantee is not in place. God bless you and your organization if you can have more than a theoretical conversation about all of your process being at a Ppk of 1.67. Your organization is rare and in good company only with Toyota as far as I know.

    0
    #66159

    Grant Blair
    Participant

    This is my first try at a post on this discussion group.
    Stop thinking about how much a process MIGHT shift.
    ALL processes shift…it’s a law of nature called
    entrophy which all processes MUST follow (Otherwise,
    you could build perpetual motion machines )
    Now, given that a process WILL shift, what is the
    maximum shift which COULD occur before you detect it
    and correct the process to keep from making defectives?
    If you want to only protect PPthousand level, allow 1 sigma,but if you want to protect PPmillion, allow 1.5 sigma as protection.

    0
    #66170

    Anonymous
    Participant

    Say what?

    0
    #66212

    Grant Blair
    Participant

    If I control chart a Cpk=1.5 process with a 0 target and a sigma of 1 and the X-bar (or X)chart indicates a point beyond 3 sigma, then how much should the process be adjusted to bring it back to target?
    1.What would happen to defect levels if you ALWAYS adjusted 3 units?
    2. What would happen if you ALWAYS adjusted 1 unit?
    3.What would happen if you ALWAYS adjusted 1.5 units?
    If you also include trend/runs rules to detect special casue, then how should the process be adjusted?
    1.What would happen if you ALWAYS adjusted by the distance of the last point to the target?
    2.What would happen if you ALWAYS adjusted by 1/2 the
    distance of the last point to target?
    Would any of the earlier rules be better than the last two?

    0
Viewing 13 posts - 1 through 13 (of 13 total)

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