iSixSigma

Repeats in DOE Sample Collection

Six Sigma – iSixSigma Forums General Forums Methodology Repeats in DOE Sample Collection

Viewing 9 posts - 1 through 9 (of 9 total)
  • Author
    Posts
  • #55369

    Robert
    Participant

    Hi,
    I’m trying to understand if there is any point in taking repeat samples. I have a process and from a screening DOE I have found two significant factors, one of these is very difficult to change and the process is fully automated.
    I am trying to find out if repeating the test a few times will help in any way and if so how do I plug this data into my software (Minitab17).
    Thanks.

    0
    #199698

    Robert Butler
    Participant

    You can take repeated measures at each experimental condition, however, because they are repeated measures and not independent replicates they will exhibit far less measure-to-measure variation than would be observed between independent replicates. The end result is, if you cannot tell your program that the measures are repeated and not replicated the program will treat the sample-to-sample variation of the repeated measures as though it is the sample-to-sample of replicates. This will result in an underestimate of the residual variation and a declaration of term significance when, in fact, the term is not significant.

    A couple of thoughts: If you are trying to run confirming experiments – block the experimental runs on the difficult to change variable and include a blocking factor in your subsequent analysis. If you are using the results of your initial work to construct a smaller or different design which incorporates your two significant factors do the same thing – block on the difficult to change variable.

    0
    #199699

    Chris Seider
    Participant

    @rbutler brings up excellent points as usual.

    A few other thoughts to help…
    1. Repeat measurements across an array of a product (e.g. the web of paper on a machine, or across the width and length of a die) can give you a valuable insight as to variability induced by the factorial design DOE
    2. Yes, be sure not to confuse multiple measurements on product with multiple repeats of a DOE run–ideally randomized
    3. Lastly, in MANY processes, it takes time for a process to “line out” after an input condition has changed. Be sure to know have that process knowledge from your team members so you don’t measure prematurely.

    0
    #199727

    Alex Pamatat
    Participant

    I read your question as “I want to run replicates of the DOE cells” and not as “I want to make additional measurements on my samples”, can you clarify which is correct?

    If you’re running replicates then certainly additional process runs should increase the statistical power of the test and give you better detection of effects that are active/significant. Without seeing the results of the DOE analysis it’s hard to make the call for you whether you need to make additional DOE runs. You may just need to validate your model with confirmation runs at this point.

    Repeating measurements on the same DOE cell/sample are not replicates and should not be treated as such. For repeated measures of samples from each DOE run/cell you may consider adding additional responses to the DOE. For example, if you’re only measuring one site per sample then perhaps you need to make several measurements of each sample and use the average and std dev of these measurements as responses in the model.

    I agree with @Robert Butler regarding the blocking. You may also make replicate runs of the two significant factors by augmenting your design. If augmenting you should consider using blocking to groups the new runs in a separate block if there has been significant time or setup changes since the original experiment which could induce change in the response.

    I use JMP for DOE design and analysis so I can’t help with Minitab, sorry.

    ….I feel obligated to ask if you completed an SOV and MSA before the DOE.

    0
    #199730

    Joel Smith
    Participant

    @bobbydazzler Two things for you to utilize:

    1 – If you take repeat measurements, you need to analyse the mean of those repeats as a single value. In Minitab, you can use Stat > DOE > Factorial > Pre-Process for Analyze Variability to do this whether the repeats were entered across rows or in separate rows.

    2 – If a hard-to-change factor is the real issue you have, consider utilizing a split-plot design. This is available in the factorial menu within Minitab and is intended for situations where one or more factors are difficult or costly to change and you want to minimize the number of changes.

    I hope this helps!

    Joel

    0
    #199731

    Alex Pamatat
    Participant

    Regarding the split-plot design, @joelminitab is correct however you would’ve needed to run the experiment this way, this shouldn’t/can’t be employed during DOE analysis after the experiment is complete. Remember, design considerations must be executed when running the DOE, they can’t be used to save a poor design after the expt has been completed.

    0
    #199732

    Joel Smith
    Participant

    @notoriousapp Totally correct! From the priginal post I gathered that a screening DOE had been done and now a modelling DOE was desired, so I am hoping the data has not been collected. Of course the original screening DOE could just be augmented, in which case it is too late for split-plot.

    0
    #199733

    Robert
    Participant

    Joel, Thanks but if I build a split plot design I cannot create levels and understand linearity (correct me if I’m wrong here). I am going to create blocks as suggested by Mr Butler.
    Thanks all for your help.
    @joelatminitab

    @notoriousapp


    @cseider


    @rbutler

    0
    #199734

    Joel Smith
    Participant

    @bobbdydazzler You are creating levels and estimating the effect just as you would with your other factors – a split-plot is very similar to what @rbutler is describing, but without using blocks. The power and ability to estimate is equivalent. You’re basically automating the process and randomizing the levels of the HTC factor.

    0
Viewing 9 posts - 1 through 9 (of 9 total)

You must be logged in to reply to this topic.