12-16-2014 01:10 PM
For the approach, normally you would determine what model you want to fit, and then choose a design that allows you to fit that model. Sometimes, you can't do all the necessary runs (3**4=81 runs in your case) in which case you would have to consider fractionating the design.
The number of replicates depends on the precision you want for your estimates, and again the number of runs you can afford.
This is all covered in great detail in many standard textbooks on design of experiments. It's far too complex and detailed to really cover in an internet forum.
Here is one such book (but there are many that cover this topic)
12-16-2014 01:25 PM
i had read some journal on it and most of the journalist are using the Full factorial experimental design,which is 2**k ~ if I wish to use that, I will have to run 81 runs right?
if I wish to fractionating it, the runs I should make is which formula ? 2***(k-1)?
12-16-2014 03:49 PM
You have three level factors, and you have four of them. That's 3**4=81 runs for a full factorial. A fractional factorial would be 3**(4–k) where k=1 or k=2.
Again, you need to know what model is being fit here, or you can't really determine what to do.
12-16-2014 09:12 PM
Response output : Actual output of a model product
Input factor : Buffer capacity
Failure time of machine
Number of operator
Number of machine
each factor are 3 levels
in this case I should consider fractional? if like that how to determine the k in fractional design?