Let say I perform a two level full factorial design (FFD) experiment to screen the important factors between A, B, C, D, and E (total of 32 runs) in a process. The analysis showed that A, B and C are the significant factors, so I proceed with optimization study with RSM for factor A, B and C with the same range I used in screening study. Assume i choose CCD, which leads to 20 runs according to design matrix generated in Minitab. Do I need to run all the 20 experiments, or I can just make use of the data from the screening study?
If I'm not mistaken, the corner points in both mentioned dsign matrix should be the same, and maybe I can move the data from FFD to CCD? However I haven't see such practice in research paper, so please advice if this is wrong.
Dear Luing,
If you have center points in your FFD, you can test for curvature effect. If this effect is not significant, you don't need to build a CCD.
If the center point effect is significant, you can add 10 trials to estimate the quadratic effects.
With 5 factors, if you neglect interaction effects between more than 2 factors you could have use a fractional factorial design of resolution V (with only 16 trials), and the same with center points.
Hope this will help you.
Pierre
Dear Luing,
Here is a reference that might be of some help: https://www.researchgate.net/publication/221951077_Multivariate_methods_in_tablet_formulation_suitable_for_early_drug_development_Predictive_models_from_a_screening_design_of_several_linked_responses?_iepl%5BviewId%5D=2kk8cl3lq6Ht6Gd0Zudg7bxVB0LeB6ay4i03&_iepl%5Bcontexts%5D%5B0%5D=prfhpi&_iepl%5Bdata%5D%5BstandardItemCount%5D=5&_iepl%5Bdata%5D%5BuserSelectedItemCount%5D=5&_iepl%5Bdata%5D%5BtopHighlightCount%5D=1&_iepl%5Bdata%5D%5BstandardItemIndex%5D=1&_iepl%5Bdata%5D%5BstandardItem1of5%5D=1&_iepl%5BtargetEntityId%5D=PB%3A221951077&_iepl%5BinteractionType%5D=publicationTitle
Article Multivariate methods in tablet formulation suitable for earl...
It depends on your number factors what you are using. If you are opting four 5 factors with 2 levels, there are 32 runs of experimental trials. And can convert two level factorial design to RSM.
If you can do 32 trials, I think it's better to do a central composite design based on a fractional factorial design (2^5-1 with resolution V). then you'll have 16 (factorial) + 10 (axial points) + 6 (center points) = 32 trials. The the structure of data is quite perfect (VIF are very closed to 1):
Thanks for all your contributions and time to answer my question. I'll use a CCD based on FFD design.
- Badges
- Science topic
- Similar topics
- Correction