1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.
Dismiss Notice
You must be a registered member in order to post messages and view/download attached files in this forum.
Click here to register.

SPC, sample size, and ANSI Z1.9

Discussion in 'SPC - Statistical Process Control' started by Dylan Brunjes, Nov 25, 2015.

Tags:
  1. Dylan Brunjes

    Dylan Brunjes New Member

    Joined:
    Aug 7, 2015
    Messages:
    3
    Likes Received:
    0
    Trophy Points:
    1
    We are trying to make a transition to SPC in our plant. Currently we perform final QC inspection on attribute characteristics based on sample sizes determined from ANSI Z1.4. As we try to implement SPC using variable data it seems that we would want to transition to Z1.9. Does anyone have any experience transitioning from Z1.4 to Z1.9? If so, how complicated is it to train QC inspectors who have been inspecting by attribute to actually making calculations for variable data?
     
    Dylan Brunjes, Nov 25, 2015
    #1
  2. James

    James Active Member

    Joined:
    Aug 31, 2015
    Messages:
    50
    Likes Received:
    13
    Trophy Points:
    7
    Location:
    Oklahoma
    Sorry to hijack a little, but would you be able to also change from a final inspection to an in-process inspection while doing this? Not sure what kind of plant you have but our quality is much better now that we do our sample at the machine as parts are being made.
     
    James, Nov 30, 2015
    #2
  3. Bev D

    Bev D Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    606
    Likes Received:
    664
    Trophy Points:
    92
    Location:
    Maine
    I'm a bit confused about what Dylan wants to do. Z1.4 and Z1.9 are for acceptance sampling - at either incoming or final inspection. SPC uses neither standard to determine the sample size and is done in-process.
    Dylan - can you clarify your question? is it related to SPC or to your acceptance sampling?
     
    Bev D, Nov 30, 2015
    #3
    Bob Doering likes this.
  4. Dylan Brunjes

    Dylan Brunjes New Member

    Joined:
    Aug 7, 2015
    Messages:
    3
    Likes Received:
    0
    Trophy Points:
    1
    Hi Bev,
    Good question.
    To clarify, we are implementing SPC, but one of the challenges we have is determining the best sample size. I thought that we could use Z1.9 to help determine sample sizes for SPC. I guess my new question would be what is the best way to determine SPC sample sizes?
    As for acceptance sampling, currently we rely on Z1.4 for acceptance sampling based on attribute data. As we transition to SPC, we are planning on collecting more variable data. It makes sense to me that we could decrease our end-item inspection sample sizes by transitioning to acceptance sampling based on Z1.9. My other questions would remain the same: Is it incredibly complicated in other people's experience to retrain QC inspectors to move from Z1.4 to Z1.9? What are some other pitfalls that other people have faced as they make this transition?

    Thank you,
    Dylan
     
    Dylan Brunjes, Nov 30, 2015
    #4
  5. Bob Doering

    Bob Doering Member

    Joined:
    Jul 30, 2015
    Messages:
    37
    Likes Received:
    27
    Trophy Points:
    17
    Beth is correct, acceptance sampling has nothing to do with SPC sampling. SPC sampling depends on the process and the nature of its variation. Some processes only need one part sample, others need more. There is no way of telling based on the information provided so far. There is no "plug and chug" methodology for sampling size in SPC - sorry.

    I am not a huge fan of acceptance sampling, either...because not to many people determine if the distribution is appropriate for that technique prior to its use. The other problem is the haystack effect. How much of the haystack do you need to sample to find evidence of a needle problem if you have 250,000 needles in there? Now, if you you have a better needle avoidance process, only have one needle in the haystack, do you REALLY think you will detect it with a SMALLER sample? Hmm.....
     
    Bob Doering, Nov 30, 2015
    #5
  6. Bev D

    Bev D Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    606
    Likes Received:
    664
    Trophy Points:
    92
    Location:
    Maine
    Dylan:

    Sample size for SPC is typically fairly low and depends on the process. For example a homogenous batch of some chemical or powder would have a sample size of 1, while a high volume object such as from injection molding may have 3-4 parts sequential per cavity that is monitored. It all depends on what makes sense for the process. I spend very little time thinking about the subgroup sample size; I just pick what makes logistical sense and isn't too large. Of far greater importance is how I subgroup and how often I will take the subgroup samples. This is based on a concept called "Rational Subgrouping". The best place to start on this topic is Donald Wheeler's "What is a Rational Subgroup?”, Quality Digest, October, 1997. In fact you might want to start reading all of Dr. Wheeler's articles at Quality Digest...

    Z1.9 isn't difficult to train in my experience it just takes a little more time than sampling by attributes. However, again that isn't the difficult or important part. Z1.9 relies on the Normailty of the distribution of the characteristic under inspection. Since many processes aren't Normal this can be Problematic. It gets worse when you are trying to detect a small defect rate as even bell-shaped symmetrical curves are horribly inaccurate beyond about 2.5 standard deviations from the mean. So this approach is fraught with over rejection. And if the distribution isn't homogenous (which you can't determine by shape) you can under reject. See Dr Wheeler's recent article on this: "The Secret Foundation of Statistical Inference" Quality Digest, December 2015.

    Bob is correct that the best approach is to prevent defects. SPC in process is great for this. So is mistake proofing and understanding your inputs effects on your outputs and controlling the inputs. However, we don't live in the ideal world and inspection is still needed and 100% inspection isn't always practical. So I have used continuous data sampling but I haven't used Z1.9 in a long time. I understand the distribution and the capability and develop a sample size that empirically prove will work most of the time. Then I plot the inspection results on a control chart to detect drifts and shifts that will alert me that my capable process may be producing defects...much better insight and protection. For processes that are producing defects I use categorical inspection. It's intended to be a stop gap until the performance can be improved. It should be hard.
     
    Bev D, Dec 1, 2015
    #6
    Bob Doering and Roberticus like this.