Fractional factorial design
Six Sigma – iSixSigma › Forums › Old Forums › General › Fractional factorial design
 This topic has 9 replies, 9 voices, and was last updated 18 years, 12 months ago by asa.

AuthorPosts

December 1, 2002 at 6:08 pm #30920
I am studying DOE in BB training and need some help. What does (kp) signify in a fractional factorial design. For example what is a 2 73 with resolution IV? Also, how many different +/ treatment combinations does this design have? Thanks for your help.
0December 1, 2002 at 11:39 pm #81157I suppose you understand that a 2^k design provides all combinations of k variables. A 2^2 design has four treatment combinations (tc’s) involving two variables. A 2^3 design has eight tc’s involving three variables.
In a 2^(kp) design, you have 2^(kp) tc’s involving k variables, instead of the usual 2^k tc’s. When p=1, we’d call this a half fraction. When p=2, this is a quarter fraction. You can think of the fraction as 1/(2*p). What you are doing is running just a fraction of the tc’s that might have been run in the 2^k design.
To create one of these fractional factorials, you actually start out with the 2^(kp) full factorial — for a 2^(52), which looks at five variables, this would be starting out with a 2^3 full factorial, with 8 runs.
This 2^3 design already has 3 variables. To run the other 2 variables you would assign each to an interaction’s terms. This gets kind of complicated, but in Excel you can create these by assigning 1’s to low levels and +1’s to high levels. The interations columns are then calculated by multiplying the respective 1’s & +1’s for the involved terms (such as AB or ABC).
But which interaction should you use? This can be tricky, since you want to pick interactions that allow you to understand interactions cleanly – without nasty confounding. This is where good statistical software or DOE design tables are very usefull.
To answer your specific question, a 2^(73) design has 2^4 tc’s, or 16.
Again, this all becomes MUCH easier with the help of good DOE software such as MINITAB, JMP, StatGraphics, Statistica, and others.
Hope that helps.0December 2, 2002 at 12:55 am #81158
Charles HParticipant@CharlesH Include @CharlesH in your post and this person will
be notified via email.The source for the following is: Understanding Industrial Designed Experiments, 4th Edition, Schmidt/Launsby, Air Academy Press
k = number of factors
q = size of fraction q = 1 (1/2 fraction) q = 2 (1/4 fraction)
Resolutions:
Res V: main effects OK (aliased with 4way interactions) 2way interactions OK (aliased with 3 way interactions)
Res IV: main effects OK (aliased with 3way interactions) some (or all) 2way interactions aliased with other 2ways
Res III: some (or all) main effects alieased with 2way interactions.
A 2^73 will have 16 runs, n reps = 30December 2, 2002 at 2:44 pm #81173If you have 7 variables and 2 levels, you can use orthogonal array L8 and make 8 tests to study all variables.
Manee0December 2, 2002 at 2:48 pm #81174Answers from others inform you what you need to know but I have a curiousity question – Did the training company / instructors internal not cover this in you training and if so – what did they give you??
0December 2, 2002 at 7:02 pm #81184JM,
Resolution IV is the length of the shortest word to the right of the defining relation.
2(73) design will have 16 runs
where7 is the nos of factors0December 4, 2002 at 5:39 am #812292^(kp)
2 = signifies two levels
k = # of factors
p = # of generators
example : 2^(41) design is a half fraction, 8 runs.
the generator is D=ABC because you aliased factor D to highest order interaction. From the generator, you can can get the aliasing pattern:
D=ABC
AD=BC
DB=AC, etc….
notice the aliasing pattern, main effect equals 3way int; 2 way equals 2 way. This is a resolution IV design.
Hope this will confuse you further :).
0December 5, 2002 at 4:51 pm #81285…and provide absolutely no ability to estimate interaction effects.
0August 24, 2003 at 8:15 am #89155
zuheyr alsalihiMember@zuheyralsalihi Include @zuheyralsalihi in your post and this person will
be notified via email.Hello
I am a newcomer to this topic. Do youknow if there is a free source code available for the Fractional Factorial design? Or an algorithm that I can program. It seems to me that a lot of data has to be input to the program.
If anybody can help please, I will greatly appreciate.
Best regards,
Zuheyr0August 24, 2003 at 3:59 pm #89166Dear Sari :
You consider only control factors and what about noise factors ?
Where we place noise factor in DOE ?
If we avoid noise factors then we will neglect products characteristics hence product will not as per customers requirement.
Cheers !!!
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.