How to Measure DPMO in Engineering Drawings?

Six Sigma – iSixSigma Forums Old Forums General How to Measure DPMO in Engineering Drawings?

Viewing 10 posts - 1 through 10 (of 10 total)
  • Author
  • #30928

    Tony G.

    Lets say you have a company that manufactures electronic control systems.  The engineering drawings that the factory uses to build the systems sometimes contain errors (defects) such as incorrect wire nomenclature, wrong wire guage, parts don’t fit together as specified, etc.
    We calculated that on each page of the drawings there are (on average) 179 opportunities to make an error.  Each wire has to have the corect color, guage, type, start and end points, nomenclature, etc.
    Multiplying 179 by the number of pages in each design (which somtimes number in the hundreds), and adding in the opportunities for errors on each ISO form, gave us the total opportunities for error in each design.  From this we calculated a DPMO for each design (D/O*10^6).
    Our observed DPMOs translate to process Sigma levels between 4 and 5.5.  That does not seem to pass the common sense test.  I don’t think we are actually that good.
    We may miss a few defects in our data collection (10 – 15 %)but I don’t see where this would account for such high process sigma values.
    The root of the “problem” seems to be that we have such a large number of opportuities for error in our calculations.  To make matters worse, the process owners have suggested that we should count every leter in a wire nomenclature as an opportunity for error.  This would make our process sigma levels even higher. 
    I know it seems odd to consider a high process sigma level as a problem.  However, when it seems unbelievable, we run into a credibility issue with the whole Six Sigma methodology.
    It could also be that a process sigma level of 6 is just not good enough for our process.
    Has anyone else had experience in calculating opprotunites for error in engineering drwaings?  What methods have worked for you to produce reasonable reults?
    Thanks in advance for your comments.



    I came across something similar on a project:  Looked at defect levels in datasheets used to specify items of equipment.  Definition of unit, opportunity and defect is key here.  Assuming, in your case, each drawing is a unit you have, at least, two options:
    1.  As you have done, count the number of opportunities per drawing to find an average opportunity count.  Multiply this by the number of drawings to give the total number of opportunities.  Then count the number of defects and calculate DPMO.  Tends to result in a low DPMO as there are so many opportunities.
    2.  Take the number of opportunities as the number of drawings and any drawing with at least one error as being a defect.  Will probably result in a DPMO more in line with your expectations.


    Ketan Parikh

    Hello Mr. Tony,
    In response to your query, I have following inputs,
    firstly the approach that you have mentioned is quite good and on track, it was very similar to the way we approached the DPMO calculation for engineering drawings.
    We used to have a form wherein we would write the defects and was later on calculated by the person who receives the drawing.
    Secondly as the rule says, more the opportunity of defects better the DPMO (Ofcourse that doesn’t mean one should neglect all the opportunites of defects).
    Thirdly and probably most important is that, at what stage are the defects identified and calculated. i.e Is it that defects are calculated after the drawing is passed on to the Internal supplier or the defects are calculated when the person who makes the drawing and then immediately idetifies his initial error as an defect.Since the logic is if it send to the internal supplier after number of errors identified and corrected before, than the counting of defects are already discounted.
    This are just some of the suggestion out of experience, hope the reply helps you.
    thanks and best regards 



    Your measurement is too granular; to go to the other extreme, if you have any error on a drawing and count that as a defect, your sigma level drops like a rock. You might consider categorizing the kinds of defects you might have on a drawing, so you get more realistic results.


    Charles Bruce

    I think what you are all pointing out is the overwhelming sensitivity and high degree of subjectivity of the DPMO metric. This is why I discourage the use of this metric except when there is a bona fide reason to compare one operation to another and those operations are of substantially different levels of complexity.
    I think the original writer does come to a significant conclusion. His drawing service, when measured by his definition of DPMO, must operate at an astronomically high Sigma Level. So high, in fact, that it might be non-productive to communicate.



    Your methodology is right on! At 4 sigma you are creating 6,210 defects per million opportunites. What does it cost to release a drawing ( design) get to the shop floor and find them then go back and redo the drawings?  Costly I’d imagine.
    A simpler method might be to identify grouping such as wire type, measurement errors, graphical errors etc. and any defect in that group is considered a defect. Then you can pareto the defects and find which areas are most frequently missed. Another approach might be to weight the nature of the defects and count the more critical errors much higher.
    Let us know how you resolved this issue.



    2 remarks:
    1. DPMO is an odd ratio, especially the O can be tricky. You claim that you have approx. 170 errors possible per drawing so 170 O’s and you name one error example which is a wrong start and end point That alone – I hope – you counted as two opportunities, but what if there should have been no wire and there is one drawn. That too is an error, or an wire has been drawn that should not be there. Or the wire is not connected to any point. And I bet you can keep going like this if you’d brainstorm this through this some “dummies”.
    2. You know why you can’t believe you’re so good ? Because you probable aren’t… You probably confuse defects with defective units. You do not have 5 drawings defective on a million. You probably have 5 defective drawings on 6000 (1MM/175) (which still looks pretty good to me, if you’d ask)
    To conclude: if you counted your defects correctly and your defect opportunities, then I think you are OK.



    I’ve encounted the same problem and I used PPM as my metric instead of DPMO.  DPMO gives you a better picture because it multiples the Total Opportunities for a defect making your Process Sigma look better than it may be.  PPM is more of a “defective” parts per million opportunity.  PPM should correspond directly to the yield of your process.



    In a similar project, our solution was much the same ass Ron’s. We prioritized the opportunities, and counted as those with the greatest impact (e.g., dimension error made the cut, spelling error did not). One or many errors in dimensions (for example) was counted as one defect. The end result was ~20 opportunities per drawing.
    Remember, the goal is to get better. As long as you use the same method, you can measure pre and post levels.


    Gary W

    A drawing is only one element of the ‘datapack’ that needs to be checked.  The release process should include checks for:

    Part form, fit and function
    Product conformance to specification
    Conformance to procedural rules and standards
    Change & configuration management verification and approval
    With the increasing use of CA’X’ and PDM tools within the design office and the drive towards reducing the number of drawings and reliance upon digital data throughout the value stream the type and format of the check to be performed is different and the opportunity for error greater.  The drawing is no longer the master part representation; it is merely extracted from the 3D data.
    We have taken the view that an error is registered once it affects a downstream process or function normally through one of the following:

    Data is inaccurate
    Data is incorrect
    Data is incomplete
    Excess data
    Unable to manipulate and use the data
    Data is not robust
    Data exchange is impossible
     To reduce the reliance upon human, manual checking it is more effective to use an automated data quality checker (such as I/Check) that addresses the procedural and administrative errors that occur as the design is iterated; allowing the engineers to correct the engineering form, fit and function errors.
    Checking, either a drawing or a CAD model, is the end result of the design process.  To achieve a high quality product design involves building in quality during the entire design process and not just checking the end result.

Viewing 10 posts - 1 through 10 (of 10 total)

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