iSixSigma

Low PPM environment

Six Sigma – iSixSigma Forums Old Forums General Low PPM environment

Viewing 16 posts - 1 through 16 (of 16 total)
  • Author
    Posts
  • #33404

    Bergeron
    Participant

    Hope someone out there can offer some advice.My process average is 300ppm. How do I control and monitor a process with such low ppm? Any control chart to use if the data is in attribute? The conventional p-chart will require huge sample size when process average is 300ppm. How do I select the right sample size in this case for process monitoring?

    0
    #90252

    Tony Burns
    Member

    Joyce,What is your process? What are you measuring?Dr Tony Burns

    0
    #90259

    Bergeron
    Participant

    I’m actually doing a visual mechanical inspection to determine how many units are rejected. For example if I inspected 500 units per shift and out of it 2 are rejected. The same procedure is repeated in the next shift.   

    0
    #90264

    Tony Burns
    Member

    Rare events may be plotted on an XmR chart using the event rate (inverse time between events). The events on the chart, should relate to a similar cause. If possible, it is always preferable to take measurements and plot the variable on a control chart, rather than counting rejects.Dr Tony Burns

    0
    #90272

    Gabriel
    Participant

    Attribute charts need defects, because you plot defects. If your defect ratye is low, you need big samples to have defects. The genral gideline for a valid p or np chart is to have, on average, 5 occurrences per sample (AIAG SPC handbook). If your defect rate is 300PPM, that’s 0.0003, you need a sample size of 5/0.0003=16667, that’s probably more than what you can handle.
    Swithching from attribute to variable, as adviced by Tony, would be a good choice, if possible. You can use, for example, a Xbar-R chart with a small samle size (for example, 5) and, as long as the average and variation ramins in control, the process is performing as it ussualy does (i.e. delivering 300PPM).
    However, you must be careful about outliers. Outliers are isolated individuals in the population that clearly do not belong to the stable process distribution. Even when outliers are the result of variation due to special causes, they are very hard to detect using SPC. And outliers are of concern specially when the defect rate is low.
    Imagine that we start with a process that has a lot of variation (both due to common and special causes) and with a high defect rate. And you decide to use SPC. The variation due to common causes can mask the special causes, so you may see the process as roughly stable, but not capable. Then you make some process improvement and achive a strong reduction of the varaition due to common causes. Now you see a great improvement in the within subgroup variation (and in the defects rate too), which leads you to update (tighten) the control limits. The full potential capability of the process with this reduced common cause variation is not achived because the process, which seemed to be stable before, is not stable now. With the new control limits you start detect all the variation due to special cuses that was masked. For example, a stop moves and you detect a shift in the mean, a tooling get loosen and you detect a shift in the ranges. In each case, you find the cause and fix it, preventing the problem to happen again. Note in these two examples, the special cause of variation is of the type “start and stay”, meaning that the process changes its behavior and keep the new one until the special cause is fixed. That ensures that, once that the special cause has occurred, you will eventually detect it in the control chart. You will note that, as the process gets more and more stable, this type of special cause will be less and less frequent. The next type of special cause to call your attention will be frequent special causes that act in short periods. With these “temporary” special causes, the process changes its behevior when the speciall cause appears and return to its original behavior when the special cause leaves. Maybe only one part was produced under the influence of that special cause. If you fail to take the sample in the exact moment when the special cause is happening, you will never detect that the special cause actually happened. However, because the same special cause repeats frequently, you will eventually detect the problem. As an example, imagine an operator that, on average, makes a specific mistake in 1 out of 50 parts. If this part is not in the sample, you will not dtect the mistake, but soon one of these parts will fall in one of the samples (if , for example, you use subgroups of 5, on average 1 out of 10 subgroup will have one of these parts). After you fix this type of special causes you will have a pretty stable process. Depending on the success of the firs effor to reduce variation due to common causes, you may be now in the 300 PPM. Further more, the Cpk may show that there should be much less deffect than that (for example Cpk=2 would be vitually defect free). In such a scenario (the process seems t be very stable and capable but the defect rate, even when low, it is much bigger than expected) you would suspect of outliers. Imagine that in 1 out of 3500 parts some special and strange combination of factors make that the part is not correctly positoned, and that lead to a part that is not only very far (many sigmas) away from the ususal process distribution but also out of tolerance. The usual process distribution can be as narrow as you want, and the Cpk will then be as high as you want, but you will still have about 300PPM (1 in 3500). If you had one of these parts in a sample, the point will surely fall beyond control limits. But if you are taking subgroups of 5 you will have one of those “lucky” subgroups every 700 subgroups (on average). Assuming a subgroup is taken every 2 hours and working 8 hours/day, that is 175 working days, or 8 months. That means that during 8 month you may have a perfecly stable chart with no single point showing an ou-of-cojntrol signal, tha calculated Cpk can be whatever you want (do you like 4?), the expected defect rate would be 0 PPM, but you will still have 300PPM without a clue about why. And the picture gets even worse if those 300PPM are not the result of a single rare special cause, but of several special causes that are different, independent, and even more rare. In that case, in 8 month you will detect not “the” special cause, but “one omong those” special causes. All this makes SPC not the right tool to deal with outliers. Other techniques are used to address outliers. Some of them are “harder” than SPC, such as automatic mistake proofing 100% checking, and some are “softer”, such as personnel awareness, commitment and attitude development.

    0
    #90278

    Reigle Stewart
    Participant

    Gabriel:I just want to commend you on such a thorough
    and outstanding response. Wow, that was a
    really good example. I would suggest you make
    it into a short paper.Again, very nice work.Reigle Stewart
    Old-Bald-Fat-Guy

    0
    #90290

    Gabriel
    Participant

    Thanks OBFG.
    By the way, is there any way I can contact you off-line?

    0
    #90295

    Mikel
    Member

    If your process is really at 300 ppm, you already have good controls.
    What is the objective you are trying to achieve?

    0
    #90321

    no name
    Participant

    As usual Stan, your rhetoric has little
    contributory value. Rates of change do make a
    difference and are quite important, especially
    from a management point of view. Learning
    curves – also called improvement curves – are
    of far more substance than just theoretical
    concern, as you unintelligibly contend. If you
    improve your baseline PPM by X%, there will be
    a reduction in the absolute PPM; of a
    proportional quantity. All Rigle is saying is that
    the rate-of-change is a better management
    focus, presumably because not everyone uses
    the same baseline metrics in an organization;
    however, the use of the same learning curve
    (rate-of-change) puts everyone on the same
    plane. Learning curves applies to all kinds of
    quality metrics (DPU, PPM, DPMO, Cycle-Time,
    and even things like budget overruns). So wake
    up and get with it buddy.

    0
    #90333

    Bergeron
    Participant

    Don’t I got to have any monitoring on the process? I cannot use a p-chart to monitor this process at all since the ppm is so low. I’m looking for a better control chart.

    0
    #90335

    Lynn Zung
    Participant

    I think Joyce’s problem  is a common problem in factory. some kinds of defect are really low ppm( less than 100 some times). but some of those are really vital defects. so we need  monitor and control. aslo for most of visual charactorestics. I dont know how to change to variable control? but those problem frequently cause customer complaint. if those visual defects have low ppm as Joyce’s situation even less and important to customer,. aslo it is much difficult to convert to variables contronl ( for example:scratches or rust on a complex shape)
    How to deal with this???
    I really like a solution based on any method . not only 300ppm is aready good(it is not good for management. why not 100,50,10 even 0) .and some theoretical method.
    Can some one come out it  and help Joyce’s and so many torturous engineer in shop floor???
    Thanks in advance
     

    0
    #90346

    Mikel
    Member

    Dear Ms. No Name,
    Again I am not your buddy.
    You are responding to the wrong post here or maybe exhausted all of your creativity on the first post.
    The question was how to control and monitor a process that already is achieving 300 PPM. You clearly don’t know what a 300 PPM process looks like, because how to control and monitor is not the issue.

    0
    #90348

    Mikel
    Member

    Control charts are the wrong place to be looking.
    Try Poka Yoke or some simple rules for operating. I know of some Sony factories that make tens of thousands of product per day that shut down on the second occurance of the same defect in the same day.
    If defects don’t occur often but are very detrimental when they do, you want everyone to know when they occur. Charts just will not do that. Discipline will.

    0
    #90350

    Anonymous
    Participant

    Please, become positive and make a solid contribution.

    0
    #90648

    marklamfu
    Participant

    The attributed chart(e.g. P-chart, np-chart,etc) do not fit the so low ppm, please use variable method to control the process, X_bar chart is OK, CPK trend is OK yet.

    0
    #90651

    vidyut bapat
    Member

    If it makes sense, start monitoring and improving process sigma.

    0
Viewing 16 posts - 1 through 16 (of 16 total)

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