 |
|
| Science Forum Index » Space - Consult Forum » Feasible design of experiment: DOE... |
|
Page 1 of 1 |
|
| Author |
Message |
| 2000... |
 Posted: Wed Sep 30, 2009 5:30 am |
|
|
|
Guest
|
Hello everybody,
I am working on “improvement of formulation and process parameters for
a process”. For a proper planning of my experiments I need a design of
experiment.
I have to show the influence of 8 factors to my process. Whereas five
of this factors contain 2 levels, and three factors contain 3 levels.
I will have time for 15 runs maximally.
Someone gave me the advice to try Plackett-Burmann, but as I have
factors with 3 levels, I am not sure if it really is the right design.
You can determine the influence of 11 factors in 12 runs. But as I
need just 8 factors, I could combine factors 6 to 11 to three factors
of 3 levels, right?
Further someone suggested D-optimal design. Apparently, 2-level, 3-
level (or higher) factors can be used directly. Also, the algorithm is
more flexible and can be used for any number of runs (i. e. 15 in my
case). Unfortunately, I couldn't find too many references. Is this
design appropriate for my problem?
Any other suggestions?
As I have no experience with DOE and this question is quite essential
in my project, I would be really grateful if someone could help me.
Thank you in advance.
Cheers, |
|
|
| Back to top |
|
|
|
| Paige Miller... |
| Posted: Wed Sep 30, 2009 6:52 am |
|
|
|
Guest
|
On Sep 30, 11:30 am, 2000 <ivonne2... at (no spam) live.com> wrote:
[quote:2b77b8d028]Hello everybody,
I am working on “improvement of formulation and process parameters for
a process”. For a proper planning of my experiments I need a design of
experiment.
I have to show the influence of 8 factors to my process. Whereas five
of this factors contain 2 levels, and three factors contain 3 levels.
I will have time for 15 runs maximally.
Someone gave me the advice to try Plackett-Burmann, but as I have
factors with 3 levels, I am not sure if it really is the right design.
You can determine the influence of 11 factors in 12 runs. But as I
need just 8 factors, I could combine factors 6 to 11 to three factors
of 3 levels, right?
Further someone suggested D-optimal design. Apparently, 2-level, 3-
level (or higher) factors can be used directly. Also, the algorithm is
more flexible and can be used for any number of runs (i. e. 15 in my
case). Unfortunately, I couldn't find too many references. Is this
design appropriate for my problem?
Any other suggestions?
As I have no experience with DOE and this question is quite essential
in my project, I would be really grateful if someone could help me.
Thank you in advance.
Cheers,
[/quote:2b77b8d028]
Either method can provide an acceptable design, or it could provide a
poor design. Some of the differences depend on the statistical needs
of your project. For example, the 12 run Plackett-Burman design
usually does not allow interactions to be estimated (and I have no
idea what it will do if you try to squeeze your 3-level factors into
this design). The D-Optimal design can allow a few interactions to be
estimated, and should be able to handle 3-level factors properly.
There are other differences as well.
That's about as much as I can say without a much more detailed
understanding of your experiment and what you hope to achieve.
--
Paige Miller
paige\dot\miller \at\ kodak\dot\com |
|
|
| Back to top |
|
|
|
|
|
All times are GMT - 5 Hours
The time now is Thu Dec 03, 2009 1:17 am
|
|