iSixSigma

Setting up a Shelf life DOE

Six Sigma – iSixSigma Forums Old Forums General Setting up a Shelf life DOE

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #33302

    Blackbelt CS
    Participant

    We are designing a product line of blends.  I want to set up a DOE for a range of blend composites. Example:  water, organic solvent, acid.  The proposed factors affecting shelf life were time, temperature, %acid and %water.  For a full factorial with center points, and 2 replicates it comes to 37 samples.  We have 2 acids and 2 organics we want to test so a total of 4 different products, and a total of 148 samples.  Does this seem logical, using time and temperature as factors for a DOE?

    0
    #89876

    BB
    Participant

    I’d balk at the large sample size and prefer to use a smaller sample size with a single replicate, unless there is very small variation in the shelf life variable at different factor levels. While not wanting to sound miserly in choosing a smaller sample size, I’d prefer to confound the factors I thought could be done so- this reduces my overall cost of the experiment and also lets me understand whether my factors were chosen right. All this is just off the top of my head. Would like to know what others think

    0
    #89885

    Robert Butler
    Participant

      The purpose of a DOE is minimum effort for maximum information. 148 samples to check out seven variables (time, temperature, %organic, %H2O, and %Acid with two acids and two organics) is not in keeping with the spirit or intent of design.
      It sounds like you have a mixture problem with process variables. If the mixture is H2O, Acid, Organic then you need to look into building a design that includes mixture ( percent concentrations of the three additives) and process variables (time, temperature, acid type and organic type).There are a number of packages on the market that will permit you to develop a much more economical design on these variables.  
      If you don’t have a design package then 112 experiments (simplex lattices for each of the Acid/Organic types at each of the 4 time, temperature combinations ) plus a couple of replicate points for a total of 114 would be the maximum and even this is overkill.
      For a main effects study the brute force approach would reduce to 64 plus two replicate points for a total of 66.  Of course,for either case any good computer aided design package would generate designs that are much smaller.
      You need to remember that all your replication does is give you more degrees of freedom for error estimate. An unreplicated design already has degrees of freedom for the error estimate since it is highly unlikely that every model term will be significant. Consequently, rather than replicating everything, run a couple of replicates on your centers (or on a couple of the design points if centers aren’t possible) and then do the analysis.  When you are done with your initial analysis you could go ahead and run a complete replicate and do a re-analysis but you will probably find that nothing much has changed in the way of your initial findings. (If you have run an experiment in the past which you completely replicated you could check this by re-running your analysis of that work with only the single run and a random choice of one or two points from the second replication.  You will most likely find that there is no difference between the results of the reduced and the full analysis.)
      If this is not a mixture problem then you have a situation of seven variables, two of which happen to be type variables.  A rough estimate for the upper limit of a design of this type would be 6 points for the five continuous variables run on the matrix of Acid/Organic combinations to give 24 runs. Add a couple of replicate points and you are in the region of 26 or so. Again, a good computer aided design package would give you a design with fewer points that this.
     

    0
    #89910

    Blackbelt CS
    Participant

    We have Minitab available and are using that, as you suggested we believe it is a Mixture Design issue and a realiability or shelf life issue.  We are striving to understand mixture designs and then will set up a shelf life study also.  Our blends are not equal in %, one is always <1% and the other two components can range from 100% to 0%.  If/when we figure it out with the assistance of a Master Blackbelt I will report back the strategy and DOE we decide to use.
    Thank you for your comments, they will be useful as we go forward. 

    0
    #89940

    Robert Butler
    Participant

      If you have the good fortune of having an MBB who knows how to develop a constrained mixture design which includes process variables you shouldn’t have any problem.  However, if he or she does not know how to handle this class of design, the two of you should get a copy of the Volume 5 of the ASQ Basic Reference series – How to Run Mixture Experiments for Product Quality by John Cornell and study it carefully.  The problem that you described can be handled in a single design. You do not have to set up a mixture design and then go in with a second design to study process effect such as aging.  I’m not familiar with Minitab’s capabilities with respect to building designs that include mixture and process variables.  Perhaps some other reader could give you some recommendations with regard to how well Minitab can handle such issues.

    0
    #90277

    Manee
    Participant

    This is a good example to use Taguchi’s paramter design. Water, Organic solvent and acid are used for control parameters and time, temprature are used as noise factors.  With proper design you can get results with few experiments.
    Read any book on that subject to get the insights
    Regards
    Manee

    0
    #90300

    Jonathon L. Andell
    Participant

    If your blend is nothing but water, solvent, and acid, then you should heed the advice of previous postings regarding mixture designs. However, if those are additives to some other substance, then check out the book by Schmidt and Launsby. They have a clever approach to handling the experiment.

    0
Viewing 7 posts - 1 through 7 (of 7 total)

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