iSixSigma

Supplier Defects Per Million metric

Six Sigma – iSixSigma Forums Old Forums General Supplier Defects Per Million metric

Viewing 15 posts - 1 through 15 (of 15 total)
  • Author
    Posts
  • #37914

    quality_ab
    Participant

    I am trying to figure out what could be the best way to depict monthly supplier performance.

    I would appreciate any suggestions on how to measure and portray supplier quality.
    At incoming inspection lets say we receive a lot with 500 parts. We sample 10 parts.
    1)       If all 10 are good we allow all 500 parts to go to the assembly line.
    2)       If even 1 out of the 10 is bad, we reject the entire lot of 500 and consider 500 pieces as defective in our DMR (damage material report) database.
    If the 500 parts are allowed to go through to the line (based on the incoming/receiving inspection), if the assembler/operator finds 1 or 2 parts defective, these 1 or 2 pieces are added as defects to the DMR database.
     Now at the end of the month when we want to portray the month’s supplier quality, we pull up information from the DMR database which gives us the defects per million (DPM).
    Parts defective divided by the parts received and the result multiplied by a million. The problem with this metric is that there is a large variation in the data since it is a combination of 500 pieces rejected and 1 or 2 line defect pieces rejected and in some cases the DMR’d pieces are as many as 10000 and give an unnecessarily high number of DPM (defects per million).
    Thanks in advance!

    0
    #113279

    Sanjeev Gupta
    Member

    Hello
    The process we have been following for this purpose is that we calculate the PPM rejected at the end of the month. Let us say is we finf 2 pcs defective outof the 10 samples tested out of a 500 lot  we reject the entire lot but while calculating monthly PPM , we don’t consider entire 500 pcs as deefctive but 2/10 ie 20% defectives i.e. (2/10)*500 parts as defective i.e. 100 pieces as rejected and consider the same in the monthly metrics.
    Hope this helps.
     

    0
    #113280

    Sathish Chandran A.R
    Member

    hi,
    There are two angles to the current scenario.
    1) Is your current sample sizs of 10 nos for every 500 parts a true representation of your population behaviour. Pls recalcualte your sample size.

    Here you need to check if you can handle the increased sample size.
    If the variation of the parts are bad, look at educating the vendor and train them to control this. (increasing inspection will not solve the problem).
    2) Putting a vendor rating system. here you should not only measure the defect rates of the products that you are able to detect, but also the timelyness of delivery made and lead time to develop new products and vendor’s capability to make changes in products’s specification in the time frame as desired by  you.
    In this case i would suggest the following;
    1) Recalcualte your sample size.
    2) Sit with vendor and find out the possible reasons for defects and correct them.
    3) Work towards control the variation.
    then evolve a measurement system for vendor performance.
     
    regards
     
    sathish 

    0
    #113281

    Dhananjay Joshi
    Participant

    Hi,
    1.Your sampling need to have some statistical base or some standard 
       (e.g. ISO 2859). Hence I suggest follow some standard.
    2. Supplier quality is very difficult for you to measure.
    3. It is required that supplier only measures his quality level & sends you the reoprts & only sampling & histogram can be used to verify (not the quality levels) but the general trends. How to do requires lond explanation. Can be done if you need.
    I had faced similar problem & it is now succesfully managed.
     
     

    0
    #113285

    Markert
    Participant

    The problem is that you are trying to answer the wrong question.
    Inspection sampling plans are designed to give an average quality level over an infinite series of lots. They do not assess the conformance of any individual lot, nor do they assess variation from lot-to-lot.
    The variation in outcome (accept/reject) is due purely to sampling variation. You will be rejecting lots on one day that are identical to lots which you may accept the following day.
    YOU CANNOT USE SAMPLING PLANS TO DESIGN SAMPLES FOR ASSESSING THE PERFORMANCE OF A PROCESS!
    Hope this helps
    Phil

    0
    #113290

    Da Agent
    Participant

    Phil,
    Can you touch a little further on the comment you made…..”YOU CANNOT USE SAMPLING PLANS TO DESIGN SAMPLES FOR ASSESSING THE PERFORMANCE OF A PROCESS!”
    Thanks for any help in advance…

    0
    #113291

    Markert
    Participant

    Hi
    Acceptance sampling plans are intended to guarantee a specific average quality level (AQL) in the long term by using a designed sample based on a lot size. It does not guarantee that any one lot has a specific quality. The acceptance or rejection of a lot is not related to the quality of that individual lot. In fact some lots can be passed which are significantly worse than the AQL and some lots can be rejected which are considerably worse.
    The best way to estimate the quality of the process is to take samples of say 50 pieces at intervals (but more than one per lot), and plot the count of defects per sample on a control chart. Assuming the defect rate is statistically stable over time, you can estimate the average defective rate from the overall average, and estimate the range of natural variation which would be regarded as sampling variation.
    Phil
     

    0
    #113292

    brun
    Participant

    Just a couple of thoughts (off the top of my head) for you to consider.
    What are you measuring supplier performance for? Is it to drive corrective action? Rate your suppliers? Are you trying to qualify their process?
    I agree with a previous poster about incoming acceptance not being a good way to measure a supplier’s process. However, IMHO, you shouldn’t have to do this. Your supplier should be doing this and providing/retaining for you to review as needed.
    If you’re driving c/a. I would suggest your measurement weighting be based on the std cost of the material. I would rather focus my energy on the supplier who delivered 3 bad assemblies assemblies at $10,000 each, than the one who provided 1,000 bad screws.
    If your rating your suppliers, here again I would use std cost to weight the performance measurement. However, I would also assume you’ve established target levels of performance based on the type of material/parts.
    The above may not be suitable or effective for your type of business. In any case the most important things are to be consistent and disciplined, and communicate to your suppliers what your process is and what the expectations are.
    Good luck.

    0
    #113295

    Box Cox transformed
    Participant

    If this is an example of how process performance is measured in your org. the answer to your question is to hire a Black Belt.  You will probably have enough significant for at least 2 yrs.

    0
    #113309

    Dayton
    Member

    Well written, but wrong.   AQL is (to most of us) Acceptance Quality Limit – not Average Quality Level.   And, as Acceptance Quality Limit, AQL is most assuredly, upon failure to meet, a lot-by-lot rejection criterion.   Are you just trying to pick a fight?
     
    Vinny

    0
    #113311

    quality_ab
    Participant

    Thanks to everyone for sharing their thoughts and trying to help.
    We have a seperate metric that takes into account lead time, on-time delivery, line stoppages, financial risk, etc. for every vendor to come up with a rating for each vendor.
    The Supplier Defects Per Million Metric is reported monthly to reflect the pain to the business from having to reject parts. And this is not meant to be an individual supplier metric. It is reported as one number for all parts received/rejected for that month.
    The suppliers are responsible for inspection at their end, but we still have an incoming inspection based on previous performance and defects.
    The biggest problem was variation in the number based on the fact that it includes lots rejected at incoming inspection as well as unit rejects from the assembly line.
    I had given almost given up  hope on receiving a response to my question. Thanks again to everyone for responding.

    0
    #113565

    quality_ab
    Participant

    Sanjeev
    You have suggested a GREAT idea.
    The only problem is cases where the vendor / supplier sends in the wrong parts, not dimensionally wrong, but 100% of the parts are the wrong type.
    The above is the only exception I can think of where your method will still cause a noise in the metric.
    Thanks again.
    AB
     

    0
    #114342

    SixSigma_QE_Rios
    Member

    I just ran into this old topic and wanted to clarify something for other readers. 
    AQL is defined as the worst case quality level and has nothing to do with average or limits as you both stated (average quality level and acceptable quality limit). AQL stands for Acceptable Quality Levels. I think you guys are confusing this with Average Outgoing Quality Limit (AOQL).

    0
    #114359

    Dayton
    Member

    Something tells me that you have been reading ASQ’s Glossary and Tables for Statistical Quality Control, Fourth Edition.   In definition for AQL Note 4., ASQ says:
     
    “The use of the abbreviation AQL to mean acceptable quality level is no longer recommended since modern thinking is that no fraction defective is really acceptable.  Using “acceptance quality limit” rather than “acceptable quality level” indicates a technical value where acceptance occurs.”  
     
    As I explained to my friends at ASQ (some of whom still take my calls), the statement is tantamount to academically nuanced fluff.   How is it used?   What has changed?   AQL in application is as it has been for many years.   Hence my statement to the esteemed Dr. Phil, “AQL to most of us is…..”    
     Vinny

    0
    #116819

    Clowers
    Participant

    Hi,
    For supplier performance measurement, there are few ways to monitor them such as monthly review, scorecard and etc.
    And for the DPPM calculation for incoming and line fall out, I suggest we separate them and plot to different control chart monitoring.
    Thanks.

    0
Viewing 15 posts - 1 through 15 (of 15 total)

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