iSixSigma

Process Stability SPC

Six Sigma – iSixSigma Forums Old Forums General Process Stability SPC

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
    Posts
  • #33492

    GONZALEZ
    Participant

    Hi people,
    Here are some questions regarding process stability that I would like to discuss with you. Any comment you may have is highly desirable:
    1) What would your sucessful criteria for determining if a process is stable? In my opinion, even if you have a few out of control limits events it is reasonable as long as they do not exceed the expected proportion (0.27%). A 1-Proportion test is good for this purpose.
    2) What do you think about automatic SPC limits calculation? I think it is easier for administrating the system, but it may hide process shifts over the time. However, I think both approach are valid.
    3) How many data points or time do you consider that is reasonable for re-visiting control limits? I think that 100 data points or 3 months is fine, but it is highly dependable on your process.
    Any input is welcome,

    0
    #90642

    Eileen
    Participant

    Miguel,
     
    You are making a few assumptions which may not be true.
    1. Control charts are not based on any definitive probabilitites. They are based on economics and a reasonable risk of finding a change in the process. There should be only random points within the control limits and no points outside the limits. If points are outside the limits, then it is reasonable to investigate those causes of change in the process. If you don’t want to do this, then why are you using the control chart?
    2. Control chart limits should only be recalculated when an improvement has been made. This means the process has shifted towards the target and/or the process variation has been reduced. If you recalcuate based on a time frame, you could end up with very wide limits which tell you little about the process.
    Eileen

    0
    #90687

    GONZALEZ
    Participant

    Hi Eileen,
    I think I did not explain my idea as clearly as  wish. When I said that a few data points outside the control limit are reasonable, I am talking about determining if the process is stable or not in the long term for implementing SPC charts. I agree with you that those data points outside the control limit worth to be investigated, therefore having a small proportion (around 0.27%) would be reasonable to implement control charts, otherwise an unpredictable process would cause you to over-react to try to control something that is not. However, my discussion point was if this is an appropiate method for determining if the process is stable in the long term. Any additional comment you may have is really appreciate.
    Regards.
     

    0
    #90689

    Mannu Thareja
    Participant

    Hi,
    A standard control chart usses 3 std deviations from the mean. The probabilty that the point would lie outside the control limit is only 0.27% (when the process has not changed)
    If the control limits are set to 2 std deviations from the mean, then u are increasing the Type I error or the alpha error (Rejecting the null hypothesis when it is true).
    On the other hand if u set the control limit to 4 std deviations, then u would increase the Type II error or Beta error ( Not able to reject the null hypothesis when it is false).
    Therefore the control chart should keep in mind both alpha and beta error.
    Hope this helps… or initiates the discussion :)
    regards,
    Mannu

    0
    #90691

    Savage
    Participant

    In answer to:
    #1 Others have replied to this.
     #2 “What do you think about automatic SPC limits calculation?” Not much – for the reason you stated. You are using control limits to predict the future based on the past. Once you have enough data, compute your limits and don’t keep re-computing limits just because.
    #3 How much data or time is reasonable for re-visiting control limits? More data is better, but less is often adequate. Same for time, but, as you stated, it depends on the process.

    0
    #90699

    Gabriel
    Participant

    Miguel:
    1) A process is stable when it delivers the same distribution over the time. A (real life) process is not stable. And even if it was it is not possible to prove. A process has achieved a reasonable state of control when it seldom shows an out-of-control signal and when appropiate action is taken in those few cases when it do. 0.27% of points beyond control limits is not a proof of stability if those points are attribuible to special causes. If, after an investigation, no special cause is found, given the low frequency you can assume that there was no special cause and they are just because of chance.
    About your second explanation to Eileen (“having a small proportion (around 0.27%) would be reasonable to implement control charts”), in fact when you reached that level of control, and specially if you typically find no special cause associated to them, it may be more reasonable to stop using SPC than to implement it. The main idea of SPC it to distinguish between special causes of variation and common causes of variation, so one reacts only to special causes (avoiding the “overcontrol”). The most profitable scenario for SPC is when the process is clearly not stable and all the “obvious” special causes have been eliminated. Then the SPC will help ou identify and correct 1 by one all those “not so obvious” special causes that make your process unstable, until you reach a good level of stability. Why would you wait to have a stable process to implement SPC? That just make no sense.
    What would your sucessful criteria for determining if a process is stable? In my opinion, even if you have a few out of control limits events it is reasonable as long as they do not exceed the expected proportion (0.27%). A 1-Proportion test is good for this purpose.
    2) I recomend you to not to apply automatic control limits calculation. The control limits are based on estimations of the process average and the process variation. These estimations are based on a sample, and then they are subject to sampling variation (that is, chance). Then, even if the process does not change, you wuld be setting new control limits after each new calculation. That’s to overreact, and that’s another way of “overcontrol”. It is more or less the same that to adjust the process after each part, instead of reacting only on special causes.
    A gideline to recalculating control limits is that you would do that when all the following is true:
    a) There is data-based evidence that the process behavior has changed For example, most of the ranges are below the average, too many points are too close from Xbarabr, etc. Note that tehy are out-of-control signals.
    b) The new behavior is desirable (the process has improved, or it has worsened but you don’t mind).
    c) The causes for that change of behavior is known (those are the special causes).
    d) There is a willingness to keep those special causes ongoing and to make them systematic (for example, the special cause could be that you used a special tool that is much more expensive and you are not willing to keep using, in which case you won’t change the control limits).
    3) It does not depends on how many points or how much time. If the control limits show to keep valid (because the process does not shows a change in it’s behavior) then you can keep them forever (of course that won’t speak well about your continous improverment efforts).

    0
    #91299

    Eileen
    Participant

    Miguel,
    You can’t use a probability value (.27%, for example) to define process stability. This is a very dangerous way to look at process stability.  If you have a process with many out-of-control signals, you need to work to identify those special causes of variation and worked to eliminate them as appropriate.
    Whether you are looking at the process in the short-term or long-term, it doesn’t matter. The theory is the same. You have to use SPC to identify instability and work to remove special causes of variation. So simple.
    Eileen
     

    0
    #91302

    Eileen
    Participant

    Mannu,
    Sorry, you are wrong. There are no definite probabilities associated with SPC. In addition, they are no tests of hypotheses. So, you cannot talk about a type I or alpha error nor a type II error (Beta).
    I would advise you to read Dr. Walter Shewhart’s book “Economic Control of Quality of Manufactured Product.” This will teach you the basic theory of SPC.
    Eileen
     

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

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