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.

CCF Design Axial Points Outliers

Discussion in 'DOE - Design of Experiments' started by talez2810, Nov 9, 2017.

  1. talez2810

    talez2810 New Member

    Joined:
    Nov 9, 2017
    Messages:
    2
    Likes Received:
    0
    Trophy Points:
    1
    Hello. I have encountred a problem during analysis of response in CCF design (4 factors).
    I have a strong confidence about validity of my response data but yet two points are indicated as outliers. These are axial points. I am wondering if it is possible that in CCF design axial points differ from factorial points in its potential to become outliers (I have noticed that each factorial point in my design has two similar points differing in only one factor and axial points do not have such similar points)? Does this two axial point outliers are showing me that I have to enhance my design and perform additional experiments to find model that would explain this outliers?

    Thanks in advance for any help!!!

    upload_2017-11-9_16-22-18.png

    upload_2017-11-9_16-23-33.png
    upload_2017-11-9_16-25-46.png
     
    talez2810, Nov 9, 2017
    #1
  2. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    577
    Likes Received:
    493
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    Residuals are the difference between your model's predicted value and the actual value. These points are not outliers in the normal sense of the term, but an indication that the model does not predict well in that portion of the design space. This is verified by the significant "Lack of Fit" term in the ANOVA table. The probable cause for this is that a quadratic polynomial model is only an approximation for a nonlinear model. This means that the quadratic curve will bend more or less sharply than the actual nonlinear curve in that region of the design space. I noticed that you used a face centered design. This means that there is more uncertainty around the coefficients for the quadratic terms. Are there restrictions that would prevent you from using a central composite circumscribed (CCC) design? This would provide better estimates of those coefficients. Also, is the error (~ 0.09) large enough to cause you problems with this model?
     
    Miner, Nov 9, 2017
    #2
    talez2810 likes this.
  3. talez2810

    talez2810 New Member

    Joined:
    Nov 9, 2017
    Messages:
    2
    Likes Received:
    0
    Trophy Points:
    1
    Miner - thanks a lot for your hints. So if I get you right, in case of models containing quadratic terms CCC design could provide better estimation of these terms? Is there any way to use my existing results and perform only fraction of new experiments, rather than performing CCC design experiments from scratch to end up with a better estimation of quadratic terms?
     
    talez2810, Nov 10, 2017
    #3
  4. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    577
    Likes Received:
    493
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    Sure. You would only need to run new axial points plus a couple center points. Add these to your existing design as a new block.

    I will caution you. While this will improve the quadratic terms, there is no guarantee that the polynomial will ever be a great fit for a nonlinear function. It all depends on what you are trying to do with your model.
     
    Miner, Nov 10, 2017
    #4
    talez2810 likes this.