iSixSigma

How to create DOE?

Six Sigma – iSixSigma Forums Old Forums General How to create DOE?

  • This topic has 4 replies, 4 voices, and was last updated 20 years ago by SK.
Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #30012

    SK
    Member

    We have a concern of engine water pipe comes off.  So, we will do an experiment to see what combination of pipe and hole dimension will give the best removal force.
    There are 2 factors  1) pipe dimension  : +/-      2) hole dimension : +/-
    For the pipe, there are 4 types of material    1) 1.4 t  brazing        2) 1.4 t w/o brazing      3) 1.6 t      4) CKD
    I plan to do full factorial (2×2) with 4 blocks.   So I have to do totally 16 runs.  But the problem is the test piece of hole is costly.    In this case, how can I confound each block to minimize the number of total runs ?
    Thank you very much.

    0
    #77768

    Manee
    Participant

    Here is a DOE for you.  I hope it helps

    Expt No.
    Pipe Dimension
    Hole Dimension
    Type of Material

    1
    1)  Small
    1)  small
    1)  1.4 t Brazing

    2
    2)  Large
    2)  Large
    1)  1.4 t Brazing

    3
    1)  Small
    1)  small
    2)  1.4 t Without brazing

    4
    2)  Large
    2)  Large
    2)  1.4 t Without brazing

    5
    1)  Small
    2)  Large
    3)  1.6 t

    6
    2)  Large
    1)  small
    3)  1.6 t

    7
    1)  Small
    2)  Large
    4)  ckd

    8
    2)  Large
    1)  small
    4)  ckd
     
    Manee

    0
    #77776

    Seregni
    Participant

    You must have choose  more 2 factors and realize 2** (4-1). In this case, you can do 1/2 fraction with 8 runs.
    If you have doubts , send me your e-mail.
     

    0
    #77815

    Selden
    Member

    SK,
    Here’s a modern view of how to approach your DOE problem. You may be able to significantly reduce the expense of your experimentation by using an optimal design.
    Your problem statement has two factors with two levels each and one factor with four levels.
    1. For the factors with only two levels, one can only fit a linear model, that is, a model with a constant, plus linear terms in each factor, and possibly an interaction term that is proportional to the product of these two factors. You should decide whether an interaction term might be important. If you can rule it out, then you can save on your experimentation costs, because you have more information at the start. If you can’t rule it out, leave it in.
    2. For the qualitative factor, you should decide how changing the factor value might influence the response, perhaps based on engineering knowledge. For example, you might know, or strongly suspect, that each particular value of the qualitative factor influences the response by only a constant amount, with the amount depending upon the value of the factor. Alternatively, the value of the qualitative factor might influence the slope of the response along the direction of one or both of the first two factors. Use what information you have.
    You must think about the model function. If you do not, then you will be making assumptions about the model, perhaps unwittingly and thus dangerously.
    The simplest model for the first two factors has only three terms. The simplest model of the effects of the qualitative factor is that its value affects only the constant. For four possible values of this factor, this introduces three additional terms. So you have a total of six terms in the simplest model. In the optimal-design approach, you can perform experiments for this model with any number of trials, as long as there are at least the minimum number, viz., six.
    Thus, if your trials are very expensive, you could design an experiment with just six trials. If you do, however, you will be very suseptible to any error in the experimentation, so you would need to be very careful. You may choose to make more than the minimum number of trials. The number you choose will depend on economic considerations. In the optimal design approach you can control the number of trials. In the case at hand, this could be 6, 7, 8, 9, …, 16, 17, … , that is, any integer from 6 on up.
    I have created a free service that can be used to find optimal designs of this kind at http://www.webdoe.com. You won’t find any ads there, just information.
    Good luck and contact me via [email protected] if you have any questions.
    –Selden Crary

    0
    #77823

    SK
    Member

    Dear Manee, Alessandro and Selden,
    Thank you very much for your input.  I’m very appreciate on that. 
    Manee : 
    From your design, it seems you fold it into half fraction.   Could you let me know what is the defining relation that makes you come up with such design.  
    Also, in my understanding, there are only 2 factors which are “pipe dimension” and “hole dimension”.   Whilst, “type of material” is subset of pipe only, so firstly I regard it as the “blocks”.   Is it right ?   Or I should separate it as another factor ?
    Alessandro :
    “You must have choose  more 2 factors and realize 2** (4-1). In this case, you can do 1/2 fraction with 8 runs.”     I have no idea how to choose more 2 factors.  Could you please give me more explanation at [email protected].
    Selden :
    Thanks for recommend me a new helpful source of knowledge.  I have shortly visited Webdoe, according to your suggestion.   Optimal design is completely new to me.  I will take time to go through it later on.
    Any further comment please kindly let me know.
    Regards,
    SK
     

    0
Viewing 5 posts - 1 through 5 (of 5 total)

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