iSixSigma

DPMO in complex machining operations

Six Sigma – iSixSigma Forums Old Forums General DPMO in complex machining operations

Viewing 17 posts - 1 through 17 (of 17 total)
  • Author
    Posts
  • #67852

    The Belt
    Member

    What is your defect??  Have you identified your defect? 

    0
    #67854

    Paul Steed
    Participant

    I am setting up a measuring system for our suppliers’ quality.  In one sense, the DPMO rating is one, as their responsibility is to provide material meeting spec. but we want to be able to compare suppliers and the ones doing more complex machining complain they are not fairly treated at present.
    Thanks for your interest.
    Paul

    0
    #67857

    Cone
    Participant

    Opportunities are simple and guided by a detailed process map. The opportunities are the value added transformations of the product or service accomplished by the process. Simple or complex is the same — go to the detail level and see what must happen to change the product to what the customer wants.
     
    Tight tolerances also are not relevant. Pick your supplier based on capability, measure them on execution.

    0
    #27602

    Paul Steed
    Participant

    Does anyone have a set of guidelines or rules for the calculation of DPMO for complex machining operations?  If I ask different people to do the task their answers differ wildly on the same lump of metal.
    Do people allocate more ‘opportunities’ for tighter tolerances?  Is simply counting the number of dimensions on the drawing ‘good enough’ to get started?
    Thanks
    Paul Steed
     

    0
    #67860

    Paul Steed
    Participant

    Thanks for that, I am not sure I understaood more than one word in 10.  Let me reiterate, we are a customer of various machine shops. The work we put their way varies from relatively simple to very very complex; think aerospace.  We want to evaluate them fairly and come up with a  harmonised rating system so we can focus on the poorer ones.  A failure on any dimension on any drawing could cause a catastrophic failure.  I have read about DPMO here:
    http://www.msl.com/technical_capabilities/full_prometrics.html
    and wondered if there was anything publicly available that is the equivalent in the mechanical sector.  I have spoken to a quality guy from Motorola who said he had some, but it was not in the public domain. 
    Thanks
    Paul

    0
    #67863

    MSAFAI
    Participant

    Dear Paul,
    This is what I know about DPMO, regarding Printed Circuit Boards (PCB), suppose:
    -there are 20 soldered pads on each board
    – there can be 5 types of defects on each pad (e.g. cold solder, insufficient solder, …)
    – Suppose you have taken 10 sample PCB’s, and you have found the following:
    3 sample PCB’s show in total 5 defects:  1 defect of the kind A, 3 of the kind B, and 1 of the kind C
    DPMO=  1,000,000 * (total no. of defects) / (total no. of opportunities)    = 1,000,000 *( 5) /(10*20*5)=5000
    I  hope this helps
    MSAFAI
     

    0
    #67864

    Cone
    Participant

    The opportunity you have is to do it right, not how many ways can it go wrong. 20 connections – 20 opportunities. The other way just inflates the number making it look better than it is.

    0
    #67867

    Cone
    Participant

    Paul,
    The logic in the referenced link is exactly right. They are giving one opportunity for each thing that must be right. One opportunity for each solder joint, one opportunity for each componenet. Note that even though the author says opportunity for defect, he is actuall counting each thing which must be right (not inflating the number for each thing we can imagine will go wrong).
    For machining, the logic is the same. Look at everytime a machining process changes the product toward its final state. This is from the products point of view. For example, the product goes through a broaching operation, it is changed – one opportunity. The product is transported or stored or mounted or tested, it is not changed – no opportunity. In the case of complex tools, count opportunities as if each thing done by the complex tool was done by single simple tools.
    If you want to post a simple and a complex example to walk through, I will walk you through it.
     
    Gary

    0
    #67871

    denton
    Participant

    Just remember, I’m sort of a heretic–
    My opinion is that DPMO only creates the illusion of communication.  The fact that you can’t get any two judges to interpret the number of opportunities in the same way illustrates my point.  Calculating DPMO, or “sigma levels” for that matter, clouds rather than clarifies.  “Works/Doesn’t Work” has meaning.  If DPMO has real meaning, I don’t understand it.
    Most machined parts have a few really critical dimensions, and several less critical ones.  I’d go with Ppk’s on the critical ones, and forget the DPMO approach.
    However, here again, don’t get carried away with some of the popular mythology.  Contrary to popular belief, you cannot predict defects down into the hundreds of PPM using Process Capability and samples of 50 or 100.  I know it is widely taught, but the model does not support that type of conclusion.
    Consider a process that runs normally distributed, except when the moon is full and garbage truck is late.  When that happens, it spits out unusually bad parts, the net result being a normally distributed output with a little 1,000 DPPM “pip” far out on one tail.  A sample of 1 has a .999 probability of not representing this “pip”.  A sample of 100 has a .999^100 probability of not representing the defective material.  If memory serves correctly, that comes out to about a .91 chance that your sample of 100 will not even contain a representative of the defects.  So trying to distinguish between 200 and 500 DPPM with normal distributions and samples of 100 is a waste of time.  (Don Wheeler does a more elegant, and more pessimistic analysis of this.)
    Denton Bramwell
    Sr. Master Black Belt
    [email protected]
     

    0
    #67875

    MSAFAI
    Participant

    Gary,
    Thanks for the comment. The definition you are giving of  DPMO seems logical but unconventional to me. Could you tell me where this definition comes form?
    Regards
    MSAFAI

    0
    #67877

    Paul Steed
    Participant

    Denton,
    I think that is a good approach.   Unfortunately we do not have large enough batches (typically five units a week), so most conventional stats. is out, and far too many of the dimensions are really critical, which is another way of saying we don’t know enough about the mechanisms of failure. 
    I think its back to the drawing board with this one. Thanks to all who responded for your help.
    Paul Steed
     

    0
    #67880

    denton
    Participant

    For small quantities like that, the best approach that I know is to carefully identify the input variables that affect the various tolerances, and to very carefully control those.   Your FMEA, a nonstatistical tool, may turn out to be your best guide in that effort. 

    0
    #67882

    “Ken”
    Participant

    Denton,Does the Ppk metric you support provide you any predictive capability?

    0
    #67910

    denton
    Participant

    I don’t know if I’m done wrestling with this pig, but here’s my present tilt on the issue:
    Your process probably isn’t any better than the Ppk measure indicates, but it might be quite a bit worse, especially if your Ppk is quite high.  Like many statistical tools, it starts to be much less effective when quality is very good. 
    Yes, Ppk has predictive value, and can be extremely useful.  Like all statistical tools, it has its limits.  Show me a Ppk of .7 with a sample of 100, and I will quickly believe that the process needs improvement, and I’ll know whether to work on centering or variation, or both.  Show me a Ppk of 2.5 with a sample of 100, and I will accept that your process is probably quite good, but I’ll want that reconfirmed over time, and I won’t believe the resulting estimate of DPPM for quite a long time.
    The best write-up I’ve found on this is Don Wheeler’s “Normality and the Process Behavior Chart”.  It’s a real eye-opener.  He also thoroughly demonstrates that Process Behavior Charts (Control Charts) do not require the data to be even remotely close to normally distributed.  (http://www.spcpress.com)
     
    Denton Bramwell
    Sr. Master Black Belt
    [email protected]
     

    0
    #67948

    Cone
    Participant

    It comes from Motorola Automotive where it has been the definition since the 80’s

    0
    #67985

    Diorio
    Member

    We also utilize a number of machine shops for parts with various degrees of complexity. My current plan is to give the same part to four different vendors. The part will be complex enough to provide a good test for their capability. I will allow one test part and then have them make 10 parts. These parts will then be delivered (without inspection — I am not interested in their “inspected” quality). A capability study will be completed for each vendor on each critical dimension or characteristic. I am not sure if this is the ideal way to begin our vendor development, but at this time I feel that it will be helpful. Any suggestions or comments will be appreciated.

    0
    #95432

    nancy
    Participant

    How do you define and find defect/opportunities and DPMO for growth of saved files on a shared drive of a company.  How do you know with so many millions of files which is “unnecessary” and should not be saved.  ( unnecesary files saved takes up space and time and cost to back-up and maintain. Please help me with this one.
    Total number of files saved =301,138  total MG from saved files 125GB.

    0
Viewing 17 posts - 1 through 17 (of 17 total)

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