iSixSigma

From sorting to random sampling

Six Sigma – iSixSigma Forums Old Forums General From sorting to random sampling

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
    Posts
  • #48210

    Juvy
    Participant

    I am in publishing and one of the plans is to convert 100% QC inspection on the pages that we produce to random sampling (by the same group), obviously, with the goal of saving on manpower, but still protecting quality at customer end.
    I started defining the sampling plan based on customer’s AQL, but then, I am faced with the dilemma on how to get the random sampling going (and start saving on manpower and eliminating unnecessary loopings).
    Do I do a pilot run on the worst product line (by improving it first via six sigma methodology), and later proliferating the best practices to other product lines? OR can I start outright with product lines already meeting the AQL? If the latter, how many months should the quality level be stable before I replace sorting with random sampling? Note that aside from QC inspection and/or QA random sampling, production does not perform a separate sorting process, aside from the inherent checks (sort of limited in scope) embedded in each production stage.

    0
    #161711

    Subramanian P V
    Member

    Hi Juvy
    If your intention is to avoid ANY (100%) errors in pages sampling cannot help you. Let me illustrate. If you sample 25 out of 100 pages you publish and find 10 errors in 25 pages (40%), what it means is that in the rest of the 75 pages there might be 30 more errors (40%).
    Sampling helps you in telling you what is happening in the population with the help of a sample. Its like this-you take a drop of blood/tissue from a person and test whether he has cancer. If he has cancer and you treat that particular drop of blood or tissue (sample) his whole body is not going to be cured similarly if you find errors in your sample it only means there will be errors in the unsampled portion also
    You could start with random sampling, study error patterns and try to find root cause using DMAIC and then reduce the over all errors. Manpower reduction will make sense only when you are meeting customer specs first 

    0
    #161754

    Juvy
    Participant

    Hi,     I understood the analogy very well, but the scenario here is slightly different.     Goal is to reduce QA inspection from “ALL pages in the journal” to “only representative pages in the journal”. Currently, QA inspects 100% of pages, and and for any error found, journal is returned to production for corrections. One journal is equivalent to one lot. ALL lots are to be inspected. Client tolerates a certain amount of error. Acceptance rate is at a mediocre 50% probably because of 100% buy off inspection, and there already is a factor of overkill or “individual preference in style”.      Given the setup above, will it be possible for me to immediately inject random sampling (based on ANSI tables) across the board, and check which product lines would qualify? Those that will not will be improved. With this approach, immediate benefit will be gained. But then again, the question I posted earlier is, how stable should a line be in terms of achieving the AQL, before it qualifies to sampling?     Or should I choose the most problematic product line, improve it first till it is stable enough in meeting AQL, and later proliferate the improvements to other lines? This approach will take months.     Thanks in advance for the advice.

    0
    #162149

    Fontanilla
    Participant

    Juvy,
    Implementing a sampling plan on an unstable process isn’t a good idea.  You will need to achieve stability prior.  Use of control charts is highly recommended here.
    Simply changing from 100% inspection to random sampling most likely will improve your acceptance rate (currently 50% which is terrible) at the expense of sending more defects to your customer. 
    So, you have to choose.  Which is more expensive?  Rejecting 50% of your lots of production, or angering a customer by sending defects.
    Good luck!

    0
    #162251

    Ashman
    Member

    If you have enough data to support that some of your processes are already meeting your AQL then you can begin by sampling those processes.  Assuming that you have a normal distribution this can be done by using confidence intervals.  The end result would read somthing like:  The 95% confidence interval lies between the mean of the population  plus or minus a variance.  The size of the sample will affect the confidence interval.  The larger the sample size generally the higher the confidence interval.  The more in control the process (smaller variance)  the smaller the sample size is needed for the same confidence interval.

    0
    #162298

    Stan Alekman
    Member

    Juvy,
    In terms of sampling, the AQL represents what the sample plan accepts, not what it rejects. The LTPD aka LQ represents what the sample plan rejects. The protection provided by a sample plan is what it rejects.
    AQL does not apply to the batch of product that is inspected. It applies to the average batch over the long term. Passing an AQL sample does not mean that the batch from which the sample was taken meets the AQL.

    0
Viewing 6 posts - 1 through 6 (of 6 total)

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